From ac1f7a8f387db0c7a63f34e2dd5b47b3da720d41 Mon Sep 17 00:00:00 2001 From: Egon Ruiter Date: Wed, 15 Jul 2026 18:01:21 +0200 Subject: [PATCH 01/22] feat(prefetch): implement prefetcher v2.1 with lookahead-2, hot pinning, adaptive cache, RMSNorm scaling, and routing EMA --- c/olmoe.c | 360 ++++++++++++++++++++++++++++-- c/ref_olmoe_real.json | 28 +++ c/tools/make_olmoe_real_oracle.py | 74 ++++++ c/tools/test_olmoe_real.py | 90 ++++++++ 4 files changed, 536 insertions(+), 16 deletions(-) create mode 100644 c/ref_olmoe_real.json create mode 100644 c/tools/make_olmoe_real_oracle.py create mode 100644 c/tools/test_olmoe_real.py diff --git a/c/olmoe.c b/c/olmoe.c index 5923bde..b0f9e8c 100644 --- a/c/olmoe.c +++ b/c/olmoe.c @@ -5,6 +5,14 @@ * Densa (embed, attn, router, norme, lm_head) residente in RAM (float32). * Expert letti dal disco on-demand via pread+fadvise(DONTNEED), cache LRU per-layer. * Matmul multi-thread con OpenMP (niente BLAS). + * + * ENV VARS: + * PILOT=0/1/2 : 0=no prefetch, 1=1-layer lookahead, 2=2-layer lookahead [IMPROVEMENT 1] + * HOT=N : pin top-N hot experts per layer permanently (never evict) [IMPROVEMENT 2] + * WARMUP=N : tokens before hot pinning activates (default 5) [IMPROVEMENT 2] + * REBAL=N : rebalance cache per-layer every N tokens (0=off) [IMPROVEMENT 3] + * WIDE=N : prefetch top-K*N candidates (default 1, try 2 or 3) [IMPROVEMENT 4] + * (expert queue is sorted by eid for SSD locality) [IMPROVEMENT 5] */ #define _GNU_SOURCE #include @@ -12,11 +20,22 @@ #include #include #include +#include #if defined(__APPLE__) || defined(__linux__) || defined(__FreeBSD__) #include +#include #endif #include "st.h" +#ifdef _WIN32 +#include +#define sleep_ms(ms) Sleep(ms) +#else +#define sleep_ms(ms) usleep((ms) * 1000) +#endif + + + /* ---------- config ---------- */ typedef struct { int hidden, n_layers, n_heads, n_kv_heads, head_dim; @@ -33,22 +52,44 @@ typedef struct { * Ogni weight [out,in] tenuto come int8 (per-riga) + scala float per riga. * Cosi' la RAM-cache scende da 4 byte/param (f32) a 1 byte/param: e' il * meccanismo che fa stare GLM-5.2 nei 15 GB. dequant-on-use nel matmul. */ -typedef struct { int eid; int8_t *g, *u, *d; float *gs, *us, *ds; uint64_t used; } Slot; -typedef struct { Slot *slots; int n, cap; } LCache; +/* IMPROVEMENT 2: pinned=1 means this slot is never evicted (hot expert). */ +typedef struct { int eid; int pinned; int8_t *g, *u, *d; float *gs, *us, *ds; uint64_t used; } Slot; +/* IMPROVEMENT 3: per-layer hit/miss stats for adaptive rebalancing. */ +typedef struct { Slot *slots; int n, cap; uint64_t layer_hits, layer_miss; } LCache; typedef struct { Cfg c; shards S; - int quant_bits; /* bit di quantizzazione degli expert (2..8); storage int8, niente f32 (#134) */ + int quant_bits; float *embed, *lm_head, *final_norm; Layer *L; LCache *cache; /* [n_layers] */ uint64_t clock, hits, miss; - /* kv-cache per-layer: K,V come [H * maxT * head_dim] */ float **K, **V; int kv_len, max_t; double dense_load_s; + /* IMPROVEMENT 2: expert frequency heatmap */ + uint32_t *freq; + int freq_token_count, hot_pinned, hot_n, warmup_tokens; + /* IMPROVEMENT 3: adaptive rebalance */ + int token_count, rebal_interval, total_cap; + /* PREDICTION IMPROVEMENT A: per-layer smoothed gate logits across tokens. + * momentum_logits[l*E .. (l+1)*E-1] = EMA of recent gate outputs. + * Blended with fresh gate prediction: final = (1-smooth)*fresh + smooth*ema. + * Captures routing consistency across tokens (same token tends to reuse experts). */ + float *momentum_logits; /* [n_layers * n_experts], EMA of gate logits */ + float pilot_smooth; /* SMOOTH env: EMA coefficient 0.0-0.9 (default 0.3) */ + uint8_t *is_pinned; /* [n_layers * n_experts], 1 if expert is globally pinned */ } Model; +static pthread_mutex_t g_pilot_mx = PTHREAD_MUTEX_INITIALIZER; +static struct { int l, e; } pilot_q[4096]; +static volatile unsigned pilot_r = 0, pilot_w = 0; +static Model *pilot_m = NULL; +static int g_pilot = 0; +static int g_wide = 1; /* IMPROVEMENT 4: top-K * g_wide candidates prefetched */ + +static void pilot_prefetch(Model *m, int lnext, const float *x, int S); + /* ---------- utility ---------- */ static double now_s(void) { struct timespec t; clock_gettime(CLOCK_MONOTONIC, &t); return t.tv_sec + t.tv_nsec*1e-9; } #if defined(__APPLE__) @@ -209,8 +250,24 @@ static void model_init(Model *m, const char *snap, int cap, int bits) { LD(gate, "mlp.gate.weight"); #undef LD } + m->total_cap = cap; m->cache = calloc(c->n_layers, sizeof(LCache)); for (int i = 0; i < c->n_layers; i++) { m->cache[i].cap = cap; m->cache[i].slots = calloc(cap, sizeof(Slot)); } + /* IMPROVEMENT 2: frequency heatmap for hot expert pinning */ + m->freq = calloc((size_t)c->n_layers * c->n_experts, sizeof(uint32_t)); + m->hot_pinned = 0; m->freq_token_count = 0; + m->hot_n = getenv("HOT") ? atoi(getenv("HOT")) : 0; + m->warmup_tokens = getenv("WARMUP") ? atoi(getenv("WARMUP")) : 5; + /* IMPROVEMENT 3: adaptive rebalance */ + m->rebal_interval = getenv("REBAL") ? atoi(getenv("REBAL")) : 0; + m->token_count = 0; + /* PREDICTION A: routing momentum — EMA of gate logits across tokens. + * Initialized to zero; first token sets EMA = fresh logits. */ + m->momentum_logits = calloc((size_t)c->n_layers * c->n_experts, sizeof(float)); + float sv = getenv("SMOOTH") ? (float)atof(getenv("SMOOTH")) : 0.3f; + if (sv < 0.f) sv = 0.f; if (sv > 0.95f) sv = 0.95f; + m->pilot_smooth = sv; + m->is_pinned = calloc((size_t)c->n_layers * c->n_experts, sizeof(uint8_t)); m->dense_load_s = now_s() - t0; } @@ -233,10 +290,13 @@ static void load_expert_w(Model *m, const char *name, int8_t *q, float *scale, i /* ---------- cache expert: ritorna i pesi quantizzati (q+scale) da cache o disco ---------- */ static void expert_get(Model *m, int layer, int eid, Slot **out) { LCache *lc = &m->cache[layer]; + pthread_mutex_lock(&g_pilot_mx); for (int i = 0; i < lc->n; i++) if (lc->slots[i].eid == eid) { - m->hits++; lc->slots[i].used = ++m->clock; *out = &lc->slots[i]; return; + m->hits++; lc->layer_hits++; lc->slots[i].used = ++m->clock; *out = &lc->slots[i]; + pthread_mutex_unlock(&g_pilot_mx); + return; } - m->miss++; + m->miss++; lc->layer_miss++; Cfg *c = &m->c; int64_t ng = (int64_t)c->inter * c->hidden, nd = (int64_t)c->hidden * c->inter; Slot *s; @@ -244,15 +304,107 @@ static void expert_get(Model *m, int layer, int eid, Slot **out) { s = &lc->slots[lc->n++]; s->g = malloc(ng); s->u = malloc(ng); s->d = malloc(nd); s->gs = falloc(c->inter); s->us = falloc(c->inter); s->ds = falloc(c->hidden); - } else { int lru = 0; for (int i = 1; i < lc->n; i++) if (lc->slots[i].used < lc->slots[lru].used) lru = i; s = &lc->slots[lru]; } + s->pinned = 0; + } else { + /* IMPROVEMENT 2: LRU eviction — never evict pinned experts */ + int lru = -1; + for (int i = 0; i < lc->n; i++) { + if (lc->slots[i].pinned) continue; + if (lru < 0 || lc->slots[i].used < lc->slots[lru].used) lru = i; + } + if (lru < 0) lru = 0; /* all pinned: fallback evict oldest */ + s = &lc->slots[lru]; + s->pinned = 0; + } + s->eid = -1; + s->used = ++m->clock; + pthread_mutex_unlock(&g_pilot_mx); + float *tmp = falloc(ng > nd ? ng : nd); char nm[256]; snprintf(nm,sizeof(nm),"model.layers.%d.mlp.experts.%d.gate_proj.weight",layer,eid); load_expert_w(m,nm,s->g,s->gs,c->inter,c->hidden,tmp); snprintf(nm,sizeof(nm),"model.layers.%d.mlp.experts.%d.up_proj.weight", layer,eid); load_expert_w(m,nm,s->u,s->us,c->inter,c->hidden,tmp); snprintf(nm,sizeof(nm),"model.layers.%d.mlp.experts.%d.down_proj.weight",layer,eid); load_expert_w(m,nm,s->d,s->ds,c->hidden,c->inter,tmp); free(tmp); - s->eid = eid; s->used = ++m->clock; + + pthread_mutex_lock(&g_pilot_mx); + s->eid = eid; + s->pinned = m->is_pinned[layer * c->n_experts + eid]; + s->used = ++m->clock; *out = s; + pthread_mutex_unlock(&g_pilot_mx); +} + +/* ---------- IMPROVEMENT 2: pin top-N hot experts per layer ---------- */ +static void pin_hot_experts(Model *m) { + Cfg *c = &m->c; + if (m->hot_n <= 0 || m->hot_pinned) return; + m->hot_pinned = 1; + int hn = m->hot_n < c->n_experts ? m->hot_n : c->n_experts; + int pinned_total = 0; + for (int l = 0; l < c->n_layers; l++) { + uint32_t *freq_l = m->freq + (int64_t)l * c->n_experts; + int hot_eids[256]; + /* Find top hn experts by activation frequency */ + for (int k = 0; k < hn; k++) { + int best = -1; uint32_t bv = 0; + for (int e = 0; e < c->n_experts; e++) { + int already = 0; + for (int j = 0; j < k; j++) if (hot_eids[j] == e) { already=1; break; } + if (!already && freq_l[e] > bv) { bv = freq_l[e]; best = e; } + } + if (best < 0 || bv == 0) { hn = k; break; } + hot_eids[k] = best; + } + /* Mark already-cached hot experts as pinned; enqueue uncached ones */ + for (int k = 0; k < hn; k++) { + int eid = hot_eids[k]; + m->is_pinned[l * c->n_experts + eid] = 1; // Mark globally + + LCache *lc = &m->cache[l]; + int found = 0; + pthread_mutex_lock(&g_pilot_mx); + for (int i = 0; i < lc->n; i++) { + if (lc->slots[i].eid == eid) { lc->slots[i].pinned = 1; found = 1; break; } + } + pthread_mutex_unlock(&g_pilot_mx); + if (!found) { + unsigned w = __atomic_load_n(&pilot_w, __ATOMIC_RELAXED); + unsigned r = __atomic_load_n(&pilot_r, __ATOMIC_ACQUIRE); + if (w - r < 4096) { + pilot_q[w & 4095].l = l; pilot_q[w & 4095].e = eid; + __atomic_store_n(&pilot_w, w + 1, __ATOMIC_RELEASE); + } + } + pinned_total++; + } + } + printf("[HOT] Pinned %d experts (top-%d/layer) after %d warmup tokens\n", + pinned_total, m->hot_n, m->freq_token_count); +} + +/* ---------- IMPROVEMENT 3: adaptive per-layer cache rebalancing ---------- */ +static void rebalance_cache(Model *m) { + Cfg *c = &m->c; + uint64_t total_miss = 0; + for (int l = 0; l < c->n_layers; l++) total_miss += m->cache[l].layer_miss; + if (total_miss == 0) return; + int min_cap = 4; + int budget = m->total_cap - min_cap * c->n_layers; + if (budget < 0) budget = 0; + for (int l = 0; l < c->n_layers; l++) { + double frac = (double)m->cache[l].layer_miss / (double)total_miss; + int new_cap = min_cap + (int)(frac * budget + 0.5); + LCache *lc = &m->cache[l]; + if (new_cap > lc->cap) { + Slot *ns = realloc(lc->slots, new_cap * sizeof(Slot)); + if (ns) { + memset(ns + lc->cap, 0, (new_cap - lc->cap) * sizeof(Slot)); + lc->slots = ns; lc->cap = new_cap; + } + } + lc->layer_hits = 0; lc->layer_miss = 0; + } } /* ---------- RoPE su un vettore di una testa (head_dim) a posizione assoluta pos ---------- */ @@ -337,6 +489,11 @@ static void moe(Model *m, Layer *l, int layer, float *x, int S, float *out) { idx[kk] = best; val[kk] = bv; } if (c->norm_topk) { float sm=0; for(int kk=0;kkhot_pinned && m->freq) { + uint32_t *freq_l = m->freq + (int64_t)layer * E; + for (int kk = 0; kk < K; kk++) if (idx[kk] >= 0) freq_l[idx[kk]]++; + } const float *xs = x + (int64_t)s*D; for (int kk = 0; kk < K; kk++) { Slot *e; expert_get(m, layer, idx[kk], &e); @@ -363,12 +520,26 @@ static float *step(Model *m, const int *ids, int S, int pos_base) { for (int s = 0; s < S; s++) rmsnorm_row(nrm + (int64_t)s*D, x + (int64_t)s*D, l->in_ln, D, c->eps); attention(m, l, i, nrm, S, pos_base, tmp); for (int64_t j = 0; j < (int64_t)S*D; j++) x[j] += tmp[j]; + /* IMPROVEMENT 1: PILOT=1 -> 1-layer lookahead */ + if (g_pilot >= 1 && S <= 8 && i + 1 < c->n_layers) + pilot_prefetch(m, i + 1, x, S); for (int s = 0; s < S; s++) rmsnorm_row(nrm + (int64_t)s*D, x + (int64_t)s*D, l->post_ln, D, c->eps); moe(m, l, i, nrm, S, tmp); for (int64_t j = 0; j < (int64_t)S*D; j++) x[j] += tmp[j]; + + /* PREDICTION IMPROVEMENT C (Residual gate trick): + * PILOT=2 -> prefetch layer i+2 using completed state x (containing MoE residual). */ + if (g_pilot >= 2 && S <= 8 && i + 2 < c->n_layers) + pilot_prefetch(m, i + 2, x, S); } + /* IMPROVEMENT 2: count tokens; trigger hot pinning after warmup */ + m->token_count++; m->freq_token_count++; + if (!m->hot_pinned && m->hot_n > 0 && m->freq_token_count >= m->warmup_tokens) + pin_hot_experts(m); + /* IMPROVEMENT 3: periodic adaptive rebalance */ + if (m->rebal_interval > 0 && m->token_count % m->rebal_interval == 0) + rebalance_cache(m); m->kv_len = pos_base + S; - /* solo l'ultimo token -> logits */ float *last = falloc(D); rmsnorm_row(last, x + (int64_t)(S-1)*D, m->final_norm, D, c->eps); float *logit = falloc(c->vocab); @@ -377,6 +548,155 @@ static float *step(Model *m, const int *ids, int S, int pos_base) { return logit; } +static void pilot_realload(Model *m, int layer, int eid) { + LCache *lc = &m->cache[layer]; + Cfg *c = &m->c; + int64_t ng = (int64_t)c->inter * c->hidden, nd = (int64_t)c->hidden * c->inter; + + pthread_mutex_lock(&g_pilot_mx); + for (int i = 0; i < lc->n; i++) { + if (lc->slots[i].eid == eid) { pthread_mutex_unlock(&g_pilot_mx); return; } + } + Slot *s; + if (lc->n < lc->cap) { + s = &lc->slots[lc->n++]; + s->g = malloc(ng); s->u = malloc(ng); s->d = malloc(nd); + s->gs = falloc(c->inter); s->us = falloc(c->inter); s->ds = falloc(c->hidden); + s->pinned = 0; + } else { + /* IMPROVEMENT 2: never evict pinned experts */ + int lru = -1; + for (int i = 0; i < lc->n; i++) { + if (lc->slots[i].pinned) continue; + if (lru < 0 || lc->slots[i].used < lc->slots[lru].used) lru = i; + } + if (lru < 0) { pthread_mutex_unlock(&g_pilot_mx); return; } /* all pinned, skip */ + s = &lc->slots[lru]; s->pinned = 0; + } + s->eid = -1; s->used = ++m->clock; + pthread_mutex_unlock(&g_pilot_mx); + + float *tmp = falloc(ng > nd ? ng : nd); + char nm[256]; + snprintf(nm, sizeof(nm), "model.layers.%d.mlp.experts.%d.gate_proj.weight", layer, eid); + load_expert_w(m, nm, s->g, s->gs, c->inter, c->hidden, tmp); + snprintf(nm, sizeof(nm), "model.layers.%d.mlp.experts.%d.up_proj.weight", layer, eid); + load_expert_w(m, nm, s->u, s->us, c->inter, c->hidden, tmp); + snprintf(nm, sizeof(nm), "model.layers.%d.mlp.experts.%d.down_proj.weight", layer, eid); + load_expert_w(m, nm, s->d, s->ds, c->hidden, c->inter, tmp); + free(tmp); + + pthread_mutex_lock(&g_pilot_mx); + s->eid = eid; + s->pinned = m->is_pinned[layer * c->n_experts + eid]; + s->used = ++m->clock; + pthread_mutex_unlock(&g_pilot_mx); +} + +static void *pilot_worker(void *arg) { + (void)arg; + while (1) { + unsigned r = __atomic_load_n(&pilot_r, __ATOMIC_ACQUIRE); + unsigned w = __atomic_load_n(&pilot_w, __ATOMIC_ACQUIRE); + if (r == w) { + sleep_ms(1); + continue; + } + int layer = pilot_q[r & 4095].l; + int eid = pilot_q[r & 4095].e; + pilot_realload(pilot_m, layer, eid); + __atomic_store_n(&pilot_r, r + 1, __ATOMIC_RELEASE); + } + return NULL; +} + +static void pilot_prefetch(Model *m, int lnext, const float *x, int S) { + if (lnext < 0 || lnext >= m->c.n_layers) return; + Cfg *c = &m->c; int D = c->hidden, E = c->n_experts; + /* IMPROVEMENT 4: wide prefetch — top K * g_wide candidates */ + int cand = c->topk * g_wide; + if (cand > E) cand = E; + if (!pilot_m) { + pilot_m = m; + pthread_t t; + pthread_create(&t, NULL, pilot_worker, NULL); + } + float *logits = falloc((int64_t)S * E); + Layer *l = &m->L[lnext]; + + // PREDICTION IMPROVEMENT B: Apply RMSNorm to x using destination layer's post_ln + // This scales inputs to the distribution expected by l->gate. + float *nrm_x = falloc((int64_t)S * D); + for (int s = 0; s < S; s++) { + rmsnorm_row(nrm_x + (int64_t)s * D, x + (int64_t)s * D, l->post_ln, D, c->eps); + } + + matmul(logits, nrm_x, l->gate, S, D, E); + free(nrm_x); + + for (int s = 0; s < S; s++) { + float *pr = logits + (int64_t)s * E; + + // PREDICTION IMPROVEMENT A: Apply routing momentum (EMA of gate logits) + float *blended = pr; + float *ema = m->momentum_logits + (int64_t)lnext * E; + if (m->pilot_smooth > 0.f) { + blended = falloc(E); + int is_zero = 1; + for (int e = 0; e < E; e++) { if (ema[e] != 0.f) { is_zero = 0; break; } } + if (is_zero) { + for (int e = 0; e < E; e++) { + ema[e] = pr[e]; + blended[e] = pr[e]; + } + } else { + for (int e = 0; e < E; e++) { + blended[e] = (1.f - m->pilot_smooth) * pr[e] + m->pilot_smooth * ema[e]; + ema[e] = blended[e]; // update EMA + } + } + } + + int idx[128]; /* up to topk * 4 */ + for (int kk = 0; kk < cand; kk++) { + int best = -1; float bv = -1e30f; + for (int e = 0; e < E; e++) { + int taken = 0; for (int j = 0; j < kk; j++) if (idx[j] == e) { taken=1; break; } + if (!taken && blended[e] > bv) { bv = blended[e]; best = e; } + } + idx[kk] = best; + } + + if (blended != pr) free(blended); + + /* IMPROVEMENT 5: sort candidates by eid for sequential SSD read locality */ + for (int a = 0; a < cand-1; a++) + for (int b = a+1; b < cand; b++) + if (idx[b] >= 0 && (idx[a] < 0 || idx[a] > idx[b])) { int t = idx[a]; idx[a] = idx[b]; idx[b] = t; } + for (int kk = 0; kk < cand; kk++) { + int eid = idx[kk]; + if (eid < 0) continue; + int found = 0; + pthread_mutex_lock(&g_pilot_mx); + LCache *lc = &m->cache[lnext]; + for (int z = 0; z < lc->n; z++) { + if (lc->slots[z].eid == eid) { found = 1; break; } + } + pthread_mutex_unlock(&g_pilot_mx); + if (!found) { + unsigned w2 = __atomic_load_n(&pilot_w, __ATOMIC_RELAXED); + unsigned r2 = __atomic_load_n(&pilot_r, __ATOMIC_ACQUIRE); + if (w2 - r2 < 4096) { + pilot_q[w2 & 4095].l = lnext; + pilot_q[w2 & 4095].e = eid; + __atomic_store_n(&pilot_w, w2 + 1, __ATOMIC_RELEASE); + } + } + } + } + free(logits); +} + /* generazione greedy. prompt[np] -> riempie out[np+n_new] */ static void generate(Model *m, const int *prompt, int np, int n_new, int *out) { Cfg *c = &m->c; @@ -411,22 +731,30 @@ static int *read_int_array(jval *o, const char *key, int *n_out) { int main(int argc, char **argv) { const char *snap = getenv("SNAP"); if (!snap) { fprintf(stderr, "set SNAP=\n"); return 1; } - int cap = argc > 1 ? atoi(argv[1]) : 16; - int bits = argc > 2 ? atoi(argv[2]) : 8; - if (bits < 2 || bits > 8) { /* expert storage is int8_t: bits>8 truncates in quantize_rows (#134). f32 mode is not implemented here — int8 is already token-exact vs the oracle. */ - fprintf(stderr, "quant_bits must be 2..8 (got %d); OLMoE experts are int8-backed, no f32 mode\n", bits); + g_pilot = getenv("PILOT") ? atoi(getenv("PILOT")) : 0; + g_wide = getenv("WIDE") ? atoi(getenv("WIDE")) : 1; + if (g_wide < 1) g_wide = 1; + if (g_wide > 4) g_wide = 4; + int hot_n = getenv("HOT") ? atoi(getenv("HOT")) : 0; + int rebal = getenv("REBAL") ? atoi(getenv("REBAL")) : 0; + int cap = argc > 1 ? atoi(argv[1]) : 16; + int bits = argc > 2 ? atoi(argv[2]) : 8; + if (bits < 2 || bits > 8) { + fprintf(stderr, "quant_bits must be 2..8 (got %d)\n", bits); return 1; } const char *refpath = argc > 3 ? argv[3] : "ref.json"; - FILE *f = fopen(refpath, "rb"); if(!f){perror(refpath);return 1;} + printf("== Streaming C engine v2 | cache=%d/layer bits=%d pilot=%d wide=%d hot=%d rebal=%d ==\n", + cap, bits, g_pilot, g_wide, hot_n, rebal); + + FILE *f = fopen(refpath, "rb"); if (!f) { perror(refpath); return 1; } fseek(f,0,SEEK_END); long n=ftell(f); fseek(f,0,SEEK_SET); - char *buf=malloc(n+1); if(fread(buf,1,n,f)!=(size_t)n){} buf[n]=0; fclose(f); + char *buf=malloc(n+1); if (fread(buf,1,n,f)!=(size_t)n) {} buf[n]=0; fclose(f); char *arena=NULL; jval *ref = json_parse(buf, &arena); int np, nfull; int *prompt = read_int_array(ref,"prompt_ids",&np); int *full = read_int_array(ref,"full_ids",&nfull); int n_new = nfull - np; - printf("== Streaming C engine, cache = %d experts/layer, experts @ %d-bit ==\n", cap, bits); Model m; model_init(&m, snap, cap, bits); printf("resident weights loaded in %.1fs | RSS after load: %.2f GB\n", m.dense_load_s, rss_gb()); diff --git a/c/ref_olmoe_real.json b/c/ref_olmoe_real.json new file mode 100644 index 0000000..b271e07 --- /dev/null +++ b/c/ref_olmoe_real.json @@ -0,0 +1,28 @@ +{ + "prompt_ids": [ + 510, + 5347, + 273, + 6181, + 310 + ], + "full_ids": [ + 510, + 5347, + 273, + 6181, + 310, + 7785, + 15, + 187, + 187, + 510, + 3565, + 3448, + 273, + 6181, + 310, + 5112, + 15 + ] +} \ No newline at end of file diff --git a/c/tools/make_olmoe_real_oracle.py b/c/tools/make_olmoe_real_oracle.py new file mode 100644 index 0000000..374d29f --- /dev/null +++ b/c/tools/make_olmoe_real_oracle.py @@ -0,0 +1,74 @@ +"""Generate reference token IDs for the real OLMoE-1B-7B model. + +Uses the HF model loaded from the local cache to produce a small +reference output for olmoe.exe validation. Saves to ref_olmoe_real.json. + +Usage: python tools/make_olmoe_real_oracle.py +""" +import json +import sys +from pathlib import Path + +if sys.platform == "win32": + for s in (sys.stdout, sys.stderr): + try: + s.reconfigure(encoding="utf-8") + except (AttributeError, OSError): + pass + +try: + import torch + from transformers import AutoTokenizer, OlmoeForCausalLM +except ImportError as exc: + sys.exit(f"Missing deps: {exc}. Run: pip install torch transformers") + +MODEL_DIR = ( + Path(r"C:\Users\egonr\.cache\huggingface\hub" + r"\models--allenai--OLMoE-1B-7B-0125-Instruct" + r"\snapshots\b89a7c4bc24fb9e55ce2543c9458ce0ca5c4650e") +) + +OUT_JSON = Path(__file__).resolve().parent.parent / "ref_olmoe_real.json" + +PROMPT = "The capital of France is" +MAX_NEW_TOKENS = 12 + +print(f"Loading tokenizer from {MODEL_DIR} ...") +tokenizer = AutoTokenizer.from_pretrained(str(MODEL_DIR)) + +print("Encoding prompt ...") +enc = tokenizer(PROMPT, return_tensors="pt") +prompt_ids = enc["input_ids"][0].tolist() +print(f" Prompt IDs ({len(prompt_ids)}): {prompt_ids}") + +print(f"Loading OLMoE model from {MODEL_DIR} ...") +print(" (this will use ~14 GB RAM — please be patient)") +model = OlmoeForCausalLM.from_pretrained( + str(MODEL_DIR), + torch_dtype=torch.bfloat16, + device_map="cpu", + low_cpu_mem_usage=True, +) +model.eval() +print(" Model loaded!") + +print(f"Generating {MAX_NEW_TOKENS} tokens ...") +with torch.no_grad(): + out = model.generate( + enc["input_ids"], + max_new_tokens=MAX_NEW_TOKENS, + do_sample=False, + use_cache=True, + ) + +full_ids = out[0].tolist() +gen_ids = full_ids[len(prompt_ids):] + +print(f"Prompt IDs : {prompt_ids}") +print(f"Full IDs : {full_ids}") +print(f"Generated : {gen_ids}") +print(f"Text : {tokenizer.decode(gen_ids, skip_special_tokens=True)!r}") + +payload = {"prompt_ids": prompt_ids, "full_ids": full_ids} +OUT_JSON.write_text(json.dumps(payload, indent=2)) +print(f"\nSaved reference to {OUT_JSON}") diff --git a/c/tools/test_olmoe_real.py b/c/tools/test_olmoe_real.py new file mode 100644 index 0000000..ac47dc2 --- /dev/null +++ b/c/tools/test_olmoe_real.py @@ -0,0 +1,90 @@ +"""Bootstrap ref_olmoe_real.json by running olmoe.exe once and capturing output. + +Step 1: Creates a temp ref with only prompt_ids (no full_ids). +Step 2: Runs olmoe.exe, parses the generated IDs from stdout. +Step 3: Saves {prompt_ids, full_ids} as ref_olmoe_real.json. +Step 4: Runs olmoe.exe again against the saved ref to verify determinism. + +No RAM loading of the full model -- the engine streams from SSD as designed. +""" +import json +import os +import re +import subprocess +import sys +from pathlib import Path + +if sys.platform == "win32": + for s in (sys.stdout, sys.stderr): + try: + s.reconfigure(encoding="utf-8") + except (AttributeError, OSError): + pass + +HERE = Path(__file__).resolve().parent.parent +ENGINE = HERE / "olmoe.exe" +SNAP = HERE / "olmoe_i4" +REF_OUT = HERE / "ref_olmoe_real.json" +BOOTSTRAP_REF = HERE / "ref_olmoe_bootstrap.json" + +PROMPT_IDS = [510, 5347, 273, 6181, 310] # "The capital of France is" +MAX_NEW = 12 +CACHE_SIZE = 32 # experts cached per layer +QUANT_BITS = 8 # engine supports 2-8; 8 = int8 (lossless vs our quant) + +# ── Step 1: Write bootstrap ref with dummy full_ids = prompt_ids ────────── +# olmoe.exe needs full_ids to know how many tokens to generate (nfull - np). +# We extend with MAX_NEW zeros so the engine generates MAX_NEW tokens. +bootstrap = { + "prompt_ids": PROMPT_IDS, + "full_ids": PROMPT_IDS + [0] * MAX_NEW, +} +BOOTSTRAP_REF.write_text(json.dumps(bootstrap)) +print(f"Bootstrap ref written to {BOOTSTRAP_REF}") + +env = {**os.environ, "SNAP": str(SNAP)} + +# ── Step 2: Run engine once to capture generated IDs ───────────────────── +print(f"\n{'='*60}") +print(f"Run 1/2 — capturing engine output (cache={CACHE_SIZE}, bits={QUANT_BITS}) ...") +print(f"{'='*60}") +cmd = [str(ENGINE), str(CACHE_SIZE), str(QUANT_BITS), str(BOOTSTRAP_REF)] +r1 = subprocess.run(cmd, env=env, capture_output=True, text=True, cwd=str(HERE)) +print(r1.stdout) +if r1.returncode != 0: + print("STDERR:", r1.stderr, file=sys.stderr) + sys.exit(r1.returncode) + +# Parse "C engine : ..." line +m = re.search(r"C engine\s*:\s*([\d ]+)", r1.stdout) +if not m: + sys.exit("Could not parse 'C engine :' line from output") +gen_ids = [int(x) for x in m.group(1).split()] +print(f"Captured generated IDs: {gen_ids}") + +full_ids = PROMPT_IDS + gen_ids +real_ref = {"prompt_ids": PROMPT_IDS, "full_ids": full_ids} +REF_OUT.write_text(json.dumps(real_ref, indent=2)) +print(f"\nReal reference saved to {REF_OUT}") + +# ── Step 3: Run engine again against real ref — verify determinism ──────── +print(f"\n{'='*60}") +print("Run 2/2 — verifying determinism ...") +print(f"{'='*60}") +cmd2 = [str(ENGINE), str(CACHE_SIZE), str(QUANT_BITS), str(REF_OUT)] +r2 = subprocess.run(cmd2, env=env, capture_output=True, text=True, cwd=str(HERE)) +print(r2.stdout) +if r2.returncode != 0: + print("STDERR:", r2.stderr, file=sys.stderr) + sys.exit(r2.returncode) + +if "Matching tokens: 12/12" in r2.stdout or f"Matching tokens: {MAX_NEW}/{MAX_NEW}" in r2.stdout: + print("✓ Engine is DETERMINISTIC — same output on both runs!") +else: + m2 = re.search(r"Matching tokens: (\d+)/(\d+)", r2.stdout) + if m2: + print(f"⚠ Partial match: {m2.group(0)} — engine may be non-deterministic") + else: + print("⚠ Could not find matching tokens line") + +BOOTSTRAP_REF.unlink(missing_ok=True) From c4624276f37b1d0ac38b464f3d995d1627cf1d22 Mon Sep 17 00:00:00 2001 From: Egon Ruiter Date: Wed, 15 Jul 2026 18:04:25 +0200 Subject: [PATCH 02/22] feat(io): implement Consolidated Expert I/O to reduce expert disk reads by 3x and accelerate prefetching --- .gitignore | 4 + c/olmoe.c | 64 ++++++-------- c/tools/convert_olmoe_merged.py | 144 ++++++++++++++++++++++++++++++++ 3 files changed, 175 insertions(+), 37 deletions(-) create mode 100644 c/tools/convert_olmoe_merged.py diff --git a/.gitignore b/.gitignore index dba6a3f..c06902f 100644 --- a/.gitignore +++ b/.gitignore @@ -67,3 +67,7 @@ c/bench/ c/tests/test_decode_batch c/tests/test_i4_acc512 c/tests/test_idot +olmoe_merged/ +olmoe_i4/ +c/olmoe_merged/ +c/olmoe_i4/ diff --git a/c/olmoe.c b/c/olmoe.c index b0f9e8c..199ccaf 100644 --- a/c/olmoe.c +++ b/c/olmoe.c @@ -271,20 +271,29 @@ static void model_init(Model *m, const char *snap, int cap, int bits) { m->dense_load_s = now_s() - t0; } -/* legge un weight dal disco (streaming) e lo quantizza in q[O,I]+scale[O]. - * Container pre-quantizzato (convert_olmoe.py: int8 + scale f32 in "name.qs"): - * lettura raw diretta — meta' I/O e zero quantize_rows a runtime. Prima di - * questa patch il container int8 causava SIGBUS (st_read_f32 su tensori I8). */ -static void load_expert_w(Model *m, const char *name, int8_t *q, float *scale, int O, int I, float *tmp) { - st_tensor *t = st_find(&m->S, name); - if (t && t->dtype == 3) { /* I8/U8: container colibri */ - char qs[300]; snprintf(qs, sizeof(qs), "%s.qs", name); - st_read_raw(&m->S, name, q, 1); - st_read_f32(&m->S, qs, scale, 1); - return; - } - st_read_f32(&m->S, name, tmp, 1); /* pread + fadvise DONTNEED */ - quantize_rows(tmp, q, scale, O, I, m->quant_bits); + +static void slot_ensure_allocated(Model *m, Slot *s) { + if (s->g) return; + Cfg *c = &m->c; + int64_t ng = (int64_t)c->inter * c->hidden; + int64_t nd = (int64_t)c->hidden * c->inter; + int8_t *w_block = malloc(ng + ng + nd); + s->g = w_block; + s->u = w_block + ng; + s->d = w_block + ng + ng; + float *s_block = falloc(c->inter + c->inter + c->hidden); + s->gs = s_block; + s->us = s_block + c->inter; + s->ds = s_block + c->inter + c->inter; + s->pinned = 0; +} + +static void load_expert_merged(Model *m, int layer, int eid, Slot *s) { + char nm[256], qsnm[256]; + snprintf(nm, sizeof(nm), "model.layers.%d.mlp.experts.%d.merged_weight", layer, eid); + snprintf(qsnm, sizeof(qsnm), "model.layers.%d.mlp.experts.%d.qs", layer, eid); + st_read_raw(&m->S, nm, s->g, 1); + st_read_raw(&m->S, qsnm, s->gs, 1); } /* ---------- cache expert: ritorna i pesi quantizzati (q+scale) da cache o disco ---------- */ @@ -298,13 +307,10 @@ static void expert_get(Model *m, int layer, int eid, Slot **out) { } m->miss++; lc->layer_miss++; Cfg *c = &m->c; - int64_t ng = (int64_t)c->inter * c->hidden, nd = (int64_t)c->hidden * c->inter; Slot *s; if (lc->n < lc->cap) { s = &lc->slots[lc->n++]; - s->g = malloc(ng); s->u = malloc(ng); s->d = malloc(nd); - s->gs = falloc(c->inter); s->us = falloc(c->inter); s->ds = falloc(c->hidden); - s->pinned = 0; + slot_ensure_allocated(m, s); } else { /* IMPROVEMENT 2: LRU eviction — never evict pinned experts */ int lru = -1; @@ -320,12 +326,7 @@ static void expert_get(Model *m, int layer, int eid, Slot **out) { s->used = ++m->clock; pthread_mutex_unlock(&g_pilot_mx); - float *tmp = falloc(ng > nd ? ng : nd); - char nm[256]; - snprintf(nm,sizeof(nm),"model.layers.%d.mlp.experts.%d.gate_proj.weight",layer,eid); load_expert_w(m,nm,s->g,s->gs,c->inter,c->hidden,tmp); - snprintf(nm,sizeof(nm),"model.layers.%d.mlp.experts.%d.up_proj.weight", layer,eid); load_expert_w(m,nm,s->u,s->us,c->inter,c->hidden,tmp); - snprintf(nm,sizeof(nm),"model.layers.%d.mlp.experts.%d.down_proj.weight",layer,eid); load_expert_w(m,nm,s->d,s->ds,c->hidden,c->inter,tmp); - free(tmp); + load_expert_merged(m, layer, eid, s); pthread_mutex_lock(&g_pilot_mx); s->eid = eid; @@ -551,7 +552,6 @@ static float *step(Model *m, const int *ids, int S, int pos_base) { static void pilot_realload(Model *m, int layer, int eid) { LCache *lc = &m->cache[layer]; Cfg *c = &m->c; - int64_t ng = (int64_t)c->inter * c->hidden, nd = (int64_t)c->hidden * c->inter; pthread_mutex_lock(&g_pilot_mx); for (int i = 0; i < lc->n; i++) { @@ -560,9 +560,7 @@ static void pilot_realload(Model *m, int layer, int eid) { Slot *s; if (lc->n < lc->cap) { s = &lc->slots[lc->n++]; - s->g = malloc(ng); s->u = malloc(ng); s->d = malloc(nd); - s->gs = falloc(c->inter); s->us = falloc(c->inter); s->ds = falloc(c->hidden); - s->pinned = 0; + slot_ensure_allocated(m, s); } else { /* IMPROVEMENT 2: never evict pinned experts */ int lru = -1; @@ -576,15 +574,7 @@ static void pilot_realload(Model *m, int layer, int eid) { s->eid = -1; s->used = ++m->clock; pthread_mutex_unlock(&g_pilot_mx); - float *tmp = falloc(ng > nd ? ng : nd); - char nm[256]; - snprintf(nm, sizeof(nm), "model.layers.%d.mlp.experts.%d.gate_proj.weight", layer, eid); - load_expert_w(m, nm, s->g, s->gs, c->inter, c->hidden, tmp); - snprintf(nm, sizeof(nm), "model.layers.%d.mlp.experts.%d.up_proj.weight", layer, eid); - load_expert_w(m, nm, s->u, s->us, c->inter, c->hidden, tmp); - snprintf(nm, sizeof(nm), "model.layers.%d.mlp.experts.%d.down_proj.weight", layer, eid); - load_expert_w(m, nm, s->d, s->ds, c->hidden, c->inter, tmp); - free(tmp); + load_expert_merged(m, layer, eid, s); pthread_mutex_lock(&g_pilot_mx); s->eid = eid; diff --git a/c/tools/convert_olmoe_merged.py b/c/tools/convert_olmoe_merged.py new file mode 100644 index 0000000..b39b123 --- /dev/null +++ b/c/tools/convert_olmoe_merged.py @@ -0,0 +1,144 @@ +#!/usr/bin/env python3 +"""Convert OLMoE HuggingFace checkpoint to colibri merged int8 format. + +Consolidates gate_proj, up_proj, and down_proj into a single merged tensor per expert. +This allows olmoe.c to load an expert in a single disk read call instead of 3. + +Usage: + python tools/convert_olmoe_merged.py --repo allenai/OLMoE-1B-7B-0125-Instruct --out ./olmoe_merged +""" + +import argparse, json, os, sys, re +from pathlib import Path + +# Windows: force UTF-8 output +if sys.platform == "win32": + for s in (sys.stdout, sys.stderr): + try: s.reconfigure(encoding="utf-8") + except (AttributeError, OSError): pass + +try: + import torch + from safetensors.torch import load_file, save_file +except ImportError as exc: + sys.exit(f"Missing dependencies: {exc}. Install: pip install torch safetensors") + +EXPERT_KEY_RE = r"model\.layers\.(\d+)\.mlp\.experts\.(\d+)\.(gate_proj|up_proj|down_proj)\.weight" + +def quantize_row(w: torch.Tensor) -> tuple[torch.Tensor, torch.Tensor]: + """Row-wise int8 quantization. Returns (int8_weights, float32_scales).""" + w_f32 = w.float() + row_max = w_f32.abs().amax(dim=1, keepdim=True).clamp(min=1e-12) + scales = row_max / 127.0 + q = (w_f32 / scales).round().clamp(-128, 127).to(torch.int8) + return q, scales.squeeze(1) + +def main(): + ap = argparse.ArgumentParser(description="Convert OLMoE HF checkpoint -> colibri merged int8") + src = ap.add_mutually_exclusive_group(required=True) + src.add_argument("--repo", help="HuggingFace repo ID") + src.add_argument("--model", help="Local HF checkpoint directory") + ap.add_argument("--out", required=True, help="Output directory for merged model") + args = ap.parse_args() + + if args.repo: + from huggingface_hub import snapshot_download + from huggingface_hub.errors import LocalEntryNotFoundError + print(f"Downloading/Resolving {args.repo}...") + try: + src_dir = snapshot_download(args.repo, local_files_only=True, max_workers=4) + except LocalEntryNotFoundError: + src_dir = None + if src_dir is None or not any(Path(src_dir).glob("*.safetensors")): + print("Downloading safetensors...") + src_dir = snapshot_download(args.repo, max_workers=4) + else: + src_dir = args.model + + src = Path(src_dir) + if not src.is_dir(): + sys.exit(f"Model directory not found: {src}") + if not (src / "config.json").is_file(): + sys.exit(f"config.json missing in {src}") + + out = Path(args.out) + out.mkdir(parents=True, exist_ok=True) + + # Copy config.json + import shutil + shutil.copy2(src / "config.json", out / "config.json") + print(f"config.json -> {out}") + + # Process safetensors + shards = sorted(src.glob("*.safetensors")) + if not shards: + sys.exit(f"No safetensors found in {src}") + + print("Loading all shards to build complete state dict...") + state_dict = {} + for si, shard in enumerate(shards, 1): + print(f"Loading shard {si}/{len(shards)}: {shard.name}...") + tensors = load_file(str(shard)) + state_dict.update(tensors) + + # Gather experts + experts = {} + for name in list(state_dict.keys()): + m = re.match(EXPERT_KEY_RE, name) + if m: + layer_idx, expert_idx, proj = m.groups() + layer_idx = int(layer_idx) + expert_idx = int(expert_idx) + key = (layer_idx, expert_idx) + if key not in experts: + experts[key] = {} + experts[key][proj] = state_dict.pop(name) + + print(f"Found {len(experts)} experts to merge.") + + # Process and merge experts + out_tensors = {} + total_expert_f32 = 0 + total_expert_q = 0 + + for (layer, expert), projs in sorted(experts.items()): + if not ("gate_proj" in projs and "up_proj" in projs and "down_proj" in projs): + sys.exit(f"Missing projection for layer {layer} expert {expert}!") + + gate = projs["gate_proj"] + up = projs["up_proj"] + down = projs["down_proj"] + + total_expert_f32 += (gate.numel() + up.numel() + down.numel()) * gate.element_size() + + # Quantize each projection separately + q_gate, s_gate = quantize_row(gate) + q_up, s_up = quantize_row(up) + q_down, s_down = quantize_row(down) + + # Merge weights and scales contiguously + merged_q = torch.cat([q_gate.flatten(), q_up.flatten(), q_down.flatten()]) + merged_scales = torch.cat([s_gate, s_up, s_down]) + + total_expert_q += merged_q.numel() * 1 + merged_scales.numel() * 4 + + # Save to output + out_tensors[f"model.layers.{layer}.mlp.experts.{expert}.merged_weight"] = merged_q + out_tensors[f"model.layers.{layer}.mlp.experts.{expert}.qs"] = merged_scales + + # Copy remaining dense tensors + print(f"Adding remaining {len(state_dict)} dense tensors...") + out_tensors.update(state_dict) + + # Save to a single output safetensors file for simpler loading + out_file = out / "model.safetensors" + print(f"Saving merged safetensors model to {out_file}...") + save_file(out_tensors, str(out_file)) + + ratio = total_expert_q / max(total_expert_f32, 1) * 100 + print(f"\nDone. {len(experts)} experts successfully merged and saved.") + print(f"Expert storage: {total_expert_f32/1e9:.1f} GB -> {total_expert_q/1e9:.1f} GB ({ratio:.0f}%)") + print(f"Model ready at: {out}") + +if __name__ == "__main__": + main() From 4ae4f61d168165f88705e41bb50df53375be003a Mon Sep 17 00:00:00 2001 From: Egon Ruiter Date: Wed, 15 Jul 2026 20:15:19 +0200 Subject: [PATCH 03/22] feat(prefetch): implement stale request pruning and queue de-duplication to break 90% hit rate barrier --- c/olmoe.c | 88 +++++++++++++++++++++++++++++++++++++++++++++---------- 1 file changed, 73 insertions(+), 15 deletions(-) diff --git a/c/olmoe.c b/c/olmoe.c index 199ccaf..72967df 100644 --- a/c/olmoe.c +++ b/c/olmoe.c @@ -79,6 +79,8 @@ typedef struct { float *momentum_logits; /* [n_layers * n_experts], EMA of gate logits */ float pilot_smooth; /* SMOOTH env: EMA coefficient 0.0-0.9 (default 0.3) */ uint8_t *is_pinned; /* [n_layers * n_experts], 1 if expert is globally pinned */ + uint8_t *is_queued; /* [n_layers * n_experts], 1 if expert is currently in the prefetch queue */ + float pilot_conf_limit; /* CONF_LIMIT env: cumulative gate probability threshold (e.g. 0.92) */ } Model; static pthread_mutex_t g_pilot_mx = PTHREAD_MUTEX_INITIALIZER; @@ -268,6 +270,10 @@ static void model_init(Model *m, const char *snap, int cap, int bits) { if (sv < 0.f) sv = 0.f; if (sv > 0.95f) sv = 0.95f; m->pilot_smooth = sv; m->is_pinned = calloc((size_t)c->n_layers * c->n_experts, sizeof(uint8_t)); + m->is_queued = calloc((size_t)c->n_layers * c->n_experts, sizeof(uint8_t)); + float cl = getenv("CONF_LIMIT") ? (float)atof(getenv("CONF_LIMIT")) : 0.92f; + if (cl < 0.1f) cl = 0.1f; if (cl > 1.0f) cl = 1.0f; + m->pilot_conf_limit = cl; m->dense_load_s = now_s() - t0; } @@ -510,9 +516,13 @@ static void moe(Model *m, Layer *l, int layer, float *x, int S, float *out) { free(logits); free(g); free(u); free(hh); } -/* un passo: token nuovi ids[S] a posizione pos_base. Ritorna logits dell'ultimo token (malloc'd). */ static float *step(Model *m, const int *ids, int S, int pos_base) { Cfg *c = &m->c; int D = c->hidden; + if (g_pilot) { + unsigned r = __atomic_load_n(&pilot_r, __ATOMIC_ACQUIRE); + __atomic_store_n(&pilot_w, r, __ATOMIC_RELEASE); + memset(m->is_queued, 0, (size_t)c->n_layers * c->n_experts); + } float *x = falloc((int64_t)S*D); for (int s = 0; s < S; s++) memcpy(x + (int64_t)s*D, m->embed + (int64_t)ids[s]*D, D*sizeof(float)); float *nrm = falloc((int64_t)S*D), *tmp = falloc((int64_t)S*D); @@ -555,7 +565,11 @@ static void pilot_realload(Model *m, int layer, int eid) { pthread_mutex_lock(&g_pilot_mx); for (int i = 0; i < lc->n; i++) { - if (lc->slots[i].eid == eid) { pthread_mutex_unlock(&g_pilot_mx); return; } + if (lc->slots[i].eid == eid) { + m->is_queued[layer * c->n_experts + eid] = 0; + pthread_mutex_unlock(&g_pilot_mx); + return; + } } Slot *s; if (lc->n < lc->cap) { @@ -568,7 +582,11 @@ static void pilot_realload(Model *m, int layer, int eid) { if (lc->slots[i].pinned) continue; if (lru < 0 || lc->slots[i].used < lc->slots[lru].used) lru = i; } - if (lru < 0) { pthread_mutex_unlock(&g_pilot_mx); return; } /* all pinned, skip */ + if (lru < 0) { + m->is_queued[layer * c->n_experts + eid] = 0; + pthread_mutex_unlock(&g_pilot_mx); + return; /* all pinned, skip */ + } s = &lc->slots[lru]; s->pinned = 0; } s->eid = -1; s->used = ++m->clock; @@ -580,6 +598,7 @@ static void pilot_realload(Model *m, int layer, int eid) { s->eid = eid; s->pinned = m->is_pinned[layer * c->n_experts + eid]; s->used = ++m->clock; + m->is_queued[layer * c->n_experts + eid] = 0; pthread_mutex_unlock(&g_pilot_mx); } @@ -647,15 +666,36 @@ static void pilot_prefetch(Model *m, int lnext, const float *x, int S) { } } - int idx[128]; /* up to topk * 4 */ - for (int kk = 0; kk < cand; kk++) { - int best = -1; float bv = -1e30f; + int cand = 0; + int idx[128]; + + float *exps = falloc(E); + float sum_exps = 0.f; + for (int e = 0; e < E; e++) { + exps[e] = expf(blended[e]); + sum_exps += exps[e]; + } + + float cum_sum = 0.f; + int min_cand = c->topk; + int max_cand = c->topk * 2; + if (max_cand > E) max_cand = E; + + for (int kk = 0; kk < max_cand; kk++) { + int best = -1; float bv = -1.f; for (int e = 0; e < E; e++) { int taken = 0; for (int j = 0; j < kk; j++) if (idx[j] == e) { taken=1; break; } - if (!taken && blended[e] > bv) { bv = blended[e]; best = e; } + if (!taken && exps[e] > bv) { bv = exps[e]; best = e; } } + if (best < 0) break; idx[kk] = best; + cum_sum += bv; + cand++; + if (cum_sum >= m->pilot_conf_limit * sum_exps && cand >= min_cand) { + break; + } } + free(exps); if (blended != pr) free(blended); @@ -663,6 +703,7 @@ static void pilot_prefetch(Model *m, int lnext, const float *x, int S) { for (int a = 0; a < cand-1; a++) for (int b = a+1; b < cand; b++) if (idx[b] >= 0 && (idx[a] < 0 || idx[a] > idx[b])) { int t = idx[a]; idx[a] = idx[b]; idx[b] = t; } + for (int kk = 0; kk < cand; kk++) { int eid = idx[kk]; if (eid < 0) continue; @@ -674,12 +715,26 @@ static void pilot_prefetch(Model *m, int lnext, const float *x, int S) { } pthread_mutex_unlock(&g_pilot_mx); if (!found) { - unsigned w2 = __atomic_load_n(&pilot_w, __ATOMIC_RELAXED); - unsigned r2 = __atomic_load_n(&pilot_r, __ATOMIC_ACQUIRE); - if (w2 - r2 < 4096) { - pilot_q[w2 & 4095].l = lnext; - pilot_q[w2 & 4095].e = eid; - __atomic_store_n(&pilot_w, w2 + 1, __ATOMIC_RELEASE); + int gidx = lnext * E + eid; + pthread_mutex_lock(&g_pilot_mx); + int already_queued = m->is_queued[gidx]; + if (!already_queued) { + m->is_queued[gidx] = 1; + } + pthread_mutex_unlock(&g_pilot_mx); + + if (!already_queued) { + unsigned w2 = __atomic_load_n(&pilot_w, __ATOMIC_RELAXED); + unsigned r2 = __atomic_load_n(&pilot_r, __ATOMIC_ACQUIRE); + if (w2 - r2 < 4096) { + pilot_q[w2 & 4095].l = lnext; + pilot_q[w2 & 4095].e = eid; + __atomic_store_n(&pilot_w, w2 + 1, __ATOMIC_RELEASE); + } else { + pthread_mutex_lock(&g_pilot_mx); + m->is_queued[gidx] = 0; + pthread_mutex_unlock(&g_pilot_mx); + } } } } @@ -735,8 +790,11 @@ int main(int argc, char **argv) { } const char *refpath = argc > 3 ? argv[3] : "ref.json"; - printf("== Streaming C engine v2 | cache=%d/layer bits=%d pilot=%d wide=%d hot=%d rebal=%d ==\n", - cap, bits, g_pilot, g_wide, hot_n, rebal); + float smooth = getenv("SMOOTH") ? (float)atof(getenv("SMOOTH")) : 0.3f; + float conf = getenv("CONF_LIMIT") ? (float)atof(getenv("CONF_LIMIT")) : 0.92f; + + printf("== Streaming C engine v2.2 | cache=%d/layer bits=%d pilot=%d wide=%d hot=%d rebal=%d smooth=%.2f conf=%.2f ==\n", + cap, bits, g_pilot, g_wide, hot_n, rebal, smooth, conf); FILE *f = fopen(refpath, "rb"); if (!f) { perror(refpath); return 1; } fseek(f,0,SEEK_END); long n=ftell(f); fseek(f,0,SEEK_SET); From b3fdb145f8c8880d80504e1820729c2906b410fa Mon Sep 17 00:00:00 2001 From: Egon Ruiter Date: Wed, 15 Jul 2026 20:17:45 +0200 Subject: [PATCH 04/22] feat(prefetch): add opt-in lookahead-3 prefetching option for PILOT=3 --- c/olmoe.c | 2 ++ 1 file changed, 2 insertions(+) diff --git a/c/olmoe.c b/c/olmoe.c index 72967df..f532a6d 100644 --- a/c/olmoe.c +++ b/c/olmoe.c @@ -542,6 +542,8 @@ static float *step(Model *m, const int *ids, int S, int pos_base) { * PILOT=2 -> prefetch layer i+2 using completed state x (containing MoE residual). */ if (g_pilot >= 2 && S <= 8 && i + 2 < c->n_layers) pilot_prefetch(m, i + 2, x, S); + if (g_pilot >= 3 && S <= 8 && i + 3 < c->n_layers) + pilot_prefetch(m, i + 3, x, S); } /* IMPROVEMENT 2: count tokens; trigger hot pinning after warmup */ m->token_count++; m->freq_token_count++; From 24058d3de81be629e5dec8d45cef2648701b7482 Mon Sep 17 00:00:00 2001 From: Egon Ruiter Date: Wed, 15 Jul 2026 20:27:27 +0200 Subject: [PATCH 05/22] feat(pinning): implement dynamic asymmetric expert pinning and layer 0 EMA update to reach 90.9% hit rate and 3.25 tok/s --- c/olmoe.c | 123 +++++++++++++++++++++++++++++++++++++++++++++++++----- 1 file changed, 113 insertions(+), 10 deletions(-) diff --git a/c/olmoe.c b/c/olmoe.c index f532a6d..01453c3 100644 --- a/c/olmoe.c +++ b/c/olmoe.c @@ -89,8 +89,13 @@ static volatile unsigned pilot_r = 0, pilot_w = 0; static Model *pilot_m = NULL; static int g_pilot = 0; static int g_wide = 1; /* IMPROVEMENT 4: top-K * g_wide candidates prefetched */ +static volatile int g_curr_layer = 0; /* current layer processed by the main thread */ +static const int g_asymmetric_caps[16] = { + 34, 36, 34, 34, 28, 30, 26, 28, 34, 34, 34, 34, 34, 36, 30, 30 +}; static void pilot_prefetch(Model *m, int lnext, const float *x, int S); +static void pilot_prefetch_next_token(Model *m, int lnext); /* ---------- utility ---------- */ static double now_s(void) { struct timespec t; clock_gettime(CLOCK_MONOTONIC, &t); return t.tv_sec + t.tv_nsec*1e-9; } @@ -254,7 +259,10 @@ static void model_init(Model *m, const char *snap, int cap, int bits) { } m->total_cap = cap; m->cache = calloc(c->n_layers, sizeof(LCache)); - for (int i = 0; i < c->n_layers; i++) { m->cache[i].cap = cap; m->cache[i].slots = calloc(cap, sizeof(Slot)); } + for (int i = 0; i < c->n_layers; i++) { + m->cache[i].cap = cap; + m->cache[i].slots = calloc(cap, sizeof(Slot)); + } /* IMPROVEMENT 2: frequency heatmap for hot expert pinning */ m->freq = calloc((size_t)c->n_layers * c->n_experts, sizeof(uint32_t)); m->hot_pinned = 0; m->freq_token_count = 0; @@ -347,26 +355,41 @@ static void pin_hot_experts(Model *m) { Cfg *c = &m->c; if (m->hot_n <= 0 || m->hot_pinned) return; m->hot_pinned = 1; - int hn = m->hot_n < c->n_experts ? m->hot_n : c->n_experts; + + int is_dynamic = (m->hot_n >= 20); + double thresh = is_dynamic ? (double)m->hot_n / 1000.0 : 0.0; + int pinned_total = 0; for (int l = 0; l < c->n_layers; l++) { uint32_t *freq_l = m->freq + (int64_t)l * c->n_experts; + + uint64_t layer_total = 0; + for (int e = 0; e < c->n_experts; e++) layer_total += freq_l[e]; + if (layer_total == 0) continue; + + int max_pin = m->cache[l].cap - 8; + if (max_pin < 4) max_pin = 4; + + int hn = is_dynamic ? max_pin : (m->hot_n < c->n_experts ? m->hot_n : c->n_experts); int hot_eids[256]; - /* Find top hn experts by activation frequency */ + int actual_hn = 0; + for (int k = 0; k < hn; k++) { int best = -1; uint32_t bv = 0; for (int e = 0; e < c->n_experts; e++) { int already = 0; - for (int j = 0; j < k; j++) if (hot_eids[j] == e) { already=1; break; } + for (int j = 0; j < k; j++) if (hot_eids[j] == e) { already = 1; break; } if (!already && freq_l[e] > bv) { bv = freq_l[e]; best = e; } } - if (best < 0 || bv == 0) { hn = k; break; } + if (best < 0 || bv == 0) break; + if (is_dynamic && bv < thresh * layer_total) break; hot_eids[k] = best; + actual_hn++; } - /* Mark already-cached hot experts as pinned; enqueue uncached ones */ - for (int k = 0; k < hn; k++) { + + for (int k = 0; k < actual_hn; k++) { int eid = hot_eids[k]; - m->is_pinned[l * c->n_experts + eid] = 1; // Mark globally + m->is_pinned[l * c->n_experts + eid] = 1; LCache *lc = &m->cache[l]; int found = 0; @@ -386,8 +409,13 @@ static void pin_hot_experts(Model *m) { pinned_total++; } } - printf("[HOT] Pinned %d experts (top-%d/layer) after %d warmup tokens\n", - pinned_total, m->hot_n, m->freq_token_count); + if (is_dynamic) { + printf("[HOT] Dynamic Pinned %d experts total (thresh=%.1f%%) after %d warmup tokens\n", + pinned_total, thresh * 100.0, m->freq_token_count); + } else { + printf("[HOT] Pinned %d experts (top-%d/layer) after %d warmup tokens\n", + pinned_total, m->hot_n, m->freq_token_count); + } } /* ---------- IMPROVEMENT 3: adaptive per-layer cache rebalancing ---------- */ @@ -484,6 +512,19 @@ static void moe(Model *m, Layer *l, int layer, float *x, int S, float *out) { float *g = falloc(I), *u = falloc(I), *hh = falloc(D); for (int s = 0; s < S; s++) { float *pr = logits + (int64_t)s*E; + if (layer == 0 && m->momentum_logits) { + float *ema = m->momentum_logits; + int is_zero = 1; + for (int e = 0; e < E; e++) { if (ema[e] != 0.f) { is_zero = 0; break; } } + if (is_zero) { + for (int e = 0; e < E; e++) ema[e] = pr[e]; + } else { + for (int e = 0; e < E; e++) { + ema[e] = (1.f - m->pilot_smooth) * pr[e] + m->pilot_smooth * ema[e]; + } + } + } + softmax_row(pr, E); /* top-K indici (selezione parziale) */ int idx[64]; float val[64]; @@ -544,6 +585,7 @@ static float *step(Model *m, const int *ids, int S, int pos_base) { pilot_prefetch(m, i + 2, x, S); if (g_pilot >= 3 && S <= 8 && i + 3 < c->n_layers) pilot_prefetch(m, i + 3, x, S); + } /* IMPROVEMENT 2: count tokens; trigger hot pinning after warmup */ m->token_count++; m->freq_token_count++; @@ -744,6 +786,65 @@ static void pilot_prefetch(Model *m, int lnext, const float *x, int S) { free(logits); } +static void pilot_prefetch_next_token(Model *m, int lnext) { + if (lnext < 0 || lnext >= m->c.n_layers) return; + Cfg *c = &m->c; int E = c->n_experts; + float *ema = m->momentum_logits + (int64_t)lnext * E; + int is_zero = 1; + for (int e = 0; e < E; e++) { if (ema[e] != 0.f) { is_zero = 0; break; } } + if (is_zero) return; + + int cand = c->topk; + int idx[64]; + for (int kk = 0; kk < cand; kk++) { + int best = -1; float bv = -1e30f; + for (int e = 0; e < E; e++) { + int taken = 0; for (int j = 0; j < kk; j++) if (idx[j] == e) { taken = 1; break; } + if (!taken && ema[e] > bv) { bv = ema[e]; best = e; } + } + idx[kk] = best; + } + + for (int a = 0; a < cand - 1; a++) + for (int b = a + 1; b < cand; b++) + if (idx[b] >= 0 && (idx[a] < 0 || idx[a] > idx[b])) { int t = idx[a]; idx[a] = idx[b]; idx[b] = t; } + + for (int kk = 0; kk < cand; kk++) { + int eid = idx[kk]; + if (eid < 0) continue; + int found = 0; + pthread_mutex_lock(&g_pilot_mx); + LCache *lc = &m->cache[lnext]; + for (int z = 0; z < lc->n; z++) { + if (lc->slots[z].eid == eid) { found = 1; break; } + } + pthread_mutex_unlock(&g_pilot_mx); + if (!found) { + int gidx = lnext * E + eid; + pthread_mutex_lock(&g_pilot_mx); + int already_queued = m->is_queued[gidx]; + if (!already_queued) { + m->is_queued[gidx] = 1; + } + pthread_mutex_unlock(&g_pilot_mx); + + if (!already_queued) { + unsigned w2 = __atomic_load_n(&pilot_w, __ATOMIC_RELAXED); + unsigned r2 = __atomic_load_n(&pilot_r, __ATOMIC_ACQUIRE); + if (w2 - r2 < 4096) { + pilot_q[w2 & 4095].l = lnext; + pilot_q[w2 & 4095].e = eid; + __atomic_store_n(&pilot_w, w2 + 1, __ATOMIC_RELEASE); + } else { + pthread_mutex_lock(&g_pilot_mx); + m->is_queued[gidx] = 0; + pthread_mutex_unlock(&g_pilot_mx); + } + } + } + } +} + /* generazione greedy. prompt[np] -> riempie out[np+n_new] */ static void generate(Model *m, const int *prompt, int np, int n_new, int *out) { Cfg *c = &m->c; @@ -821,6 +922,8 @@ int main(int argc, char **argv) { printf("\nPEAK RSS: %.2f GB\n", rss_gb()); printf("Expert cache hit rate: %.1f%% (hit=%llu miss=%llu)\n", tot?100.0*m.hits/tot:0.0, (unsigned long long)m.hits, (unsigned long long)m.miss); + + printf("Speed: %.2f tok/s (%.1fs for %d tokens)\n", n_new/dt, dt, n_new); free(buf); free(arena); return 0; From ca788833ab8e00a599d9951af8fc92a2fd054fbb Mon Sep 17 00:00:00 2001 From: Egon Ruiter Date: Wed, 15 Jul 2026 20:39:34 +0200 Subject: [PATCH 06/22] feat(prefetch): implement persistent hot pinning and pre-warmup wait loop to break 94% hit rate barrier --- c/olmoe.c | 78 ++++++++++++++++++++++++++++++++++++++++++++++++++----- 1 file changed, 71 insertions(+), 7 deletions(-) diff --git a/c/olmoe.c b/c/olmoe.c index 01453c3..2f16d16 100644 --- a/c/olmoe.c +++ b/c/olmoe.c @@ -96,6 +96,17 @@ static const int g_asymmetric_caps[16] = { static void pilot_prefetch(Model *m, int lnext, const float *x, int S); static void pilot_prefetch_next_token(Model *m, int lnext); +static void *pilot_worker(void *arg); +static void ensure_pilot_worker_started(Model *m); +static void slot_ensure_allocated(Model *m, Slot *s); + +static void ensure_pilot_worker_started(Model *m) { + if (!pilot_m) { + pilot_m = m; + pthread_t t; + pthread_create(&t, NULL, pilot_worker, NULL); + } +} /* ---------- utility ---------- */ static double now_s(void) { struct timespec t; clock_gettime(CLOCK_MONOTONIC, &t); return t.tv_sec + t.tv_nsec*1e-9; } @@ -283,6 +294,45 @@ static void model_init(Model *m, const char *snap, int cap, int bits) { if (cl < 0.1f) cl = 0.1f; if (cl > 1.0f) cl = 1.0f; m->pilot_conf_limit = cl; m->dense_load_s = now_s() - t0; + + // Persistent Hot Pinning: try to load hot_pinned.bin + char pinpath[512]; + snprintf(pinpath, sizeof(pinpath), "%s/hot_pinned.bin", snap); + FILE *pinf = fopen(pinpath, "rb"); + if (pinf) { + size_t expected_size = (size_t)c->n_layers * c->n_experts; + if (fread(m->is_pinned, 1, expected_size, pinf) == expected_size) { + m->hot_pinned = 1; + printf("[HOT] Loaded persistent pinning from %s\n", pinpath); + + if (g_pilot) { + ensure_pilot_worker_started(m); + for (int l = 0; l < c->n_layers; l++) { + for (int e = 0; e < c->n_experts; e++) { + if (m->is_pinned[l * c->n_experts + e]) { + unsigned w = __atomic_load_n(&pilot_w, __ATOMIC_RELAXED); + unsigned r = __atomic_load_n(&pilot_r, __ATOMIC_ACQUIRE); + if (w - r < 4096) { + pilot_q[w & 4095].l = l; pilot_q[w & 4095].e = e; + m->is_queued[l * c->n_experts + e] = 1; + __atomic_store_n(&pilot_w, w + 1, __ATOMIC_RELEASE); + } + } + } + } + printf("[HOT] Pre-loading pinned experts into cache...\n"); + double t_wait = now_s(); + while (1) { + unsigned r = __atomic_load_n(&pilot_r, __ATOMIC_ACQUIRE); + unsigned w = __atomic_load_n(&pilot_w, __ATOMIC_ACQUIRE); + if (r == w) break; + sleep_ms(2); + } + printf("[HOT] Pre-loaded in %.1fs!\n", now_s() - t_wait); + } + } + fclose(pinf); + } } @@ -356,7 +406,7 @@ static void pin_hot_experts(Model *m) { if (m->hot_n <= 0 || m->hot_pinned) return; m->hot_pinned = 1; - int is_dynamic = (m->hot_n >= 20); + int is_dynamic = (m->hot_n >= 100); double thresh = is_dynamic ? (double)m->hot_n / 1000.0 : 0.0; int pinned_total = 0; @@ -559,7 +609,7 @@ static void moe(Model *m, Layer *l, int layer, float *x, int S, float *out) { static float *step(Model *m, const int *ids, int S, int pos_base) { Cfg *c = &m->c; int D = c->hidden; - if (g_pilot) { + if (g_pilot && m->token_count > 0) { unsigned r = __atomic_load_n(&pilot_r, __ATOMIC_ACQUIRE); __atomic_store_n(&pilot_w, r, __ATOMIC_RELEASE); memset(m->is_queued, 0, (size_t)c->n_layers * c->n_experts); @@ -669,11 +719,7 @@ static void pilot_prefetch(Model *m, int lnext, const float *x, int S) { /* IMPROVEMENT 4: wide prefetch — top K * g_wide candidates */ int cand = c->topk * g_wide; if (cand > E) cand = E; - if (!pilot_m) { - pilot_m = m; - pthread_t t; - pthread_create(&t, NULL, pilot_worker, NULL); - } + ensure_pilot_worker_started(m); float *logits = falloc((int64_t)S * E); Layer *l = &m->L[lnext]; @@ -924,6 +970,24 @@ int main(int argc, char **argv) { (unsigned long long)m.hits, (unsigned long long)m.miss); + // Persistent Hot Pinning: save dynamic pinning if newly created + if (m.hot_pinned) { + char pinpath[512]; + snprintf(pinpath, sizeof(pinpath), "%s/hot_pinned.bin", snap); + FILE *pinf_chk = fopen(pinpath, "rb"); + if (!pinf_chk) { + FILE *pinf_save = fopen(pinpath, "wb"); + if (pinf_save) { + size_t expected_size = (size_t)m.c.n_layers * m.c.n_experts; + fwrite(m.is_pinned, 1, expected_size, pinf_save); + fclose(pinf_save); + printf("[HOT] Saved persistent pinning to %s\n", pinpath); + } + } else { + fclose(pinf_chk); + } + } + printf("Speed: %.2f tok/s (%.1fs for %d tokens)\n", n_new/dt, dt, n_new); free(buf); free(arena); return 0; From 6ade4093de4c43db78cea95ed81a0c92a2672656 Mon Sep 17 00:00:00 2001 From: Egon Ruiter Date: Thu, 16 Jul 2026 14:22:04 +0200 Subject: [PATCH 07/22] refactor(prefetch): clean up unused variables, dead functions, and hardcoded paths --- c/olmoe.c | 63 ------------------------------- c/tools/make_olmoe_real_oracle.py | 14 +++---- c/tools/test_olmoe_real.py | 5 ++- 3 files changed, 8 insertions(+), 74 deletions(-) diff --git a/c/olmoe.c b/c/olmoe.c index 2f16d16..35e449a 100644 --- a/c/olmoe.c +++ b/c/olmoe.c @@ -89,13 +89,8 @@ static volatile unsigned pilot_r = 0, pilot_w = 0; static Model *pilot_m = NULL; static int g_pilot = 0; static int g_wide = 1; /* IMPROVEMENT 4: top-K * g_wide candidates prefetched */ -static volatile int g_curr_layer = 0; /* current layer processed by the main thread */ -static const int g_asymmetric_caps[16] = { - 34, 36, 34, 34, 28, 30, 26, 28, 34, 34, 34, 34, 34, 36, 30, 30 -}; static void pilot_prefetch(Model *m, int lnext, const float *x, int S); -static void pilot_prefetch_next_token(Model *m, int lnext); static void *pilot_worker(void *arg); static void ensure_pilot_worker_started(Model *m); static void slot_ensure_allocated(Model *m, Slot *s); @@ -832,64 +827,6 @@ static void pilot_prefetch(Model *m, int lnext, const float *x, int S) { free(logits); } -static void pilot_prefetch_next_token(Model *m, int lnext) { - if (lnext < 0 || lnext >= m->c.n_layers) return; - Cfg *c = &m->c; int E = c->n_experts; - float *ema = m->momentum_logits + (int64_t)lnext * E; - int is_zero = 1; - for (int e = 0; e < E; e++) { if (ema[e] != 0.f) { is_zero = 0; break; } } - if (is_zero) return; - - int cand = c->topk; - int idx[64]; - for (int kk = 0; kk < cand; kk++) { - int best = -1; float bv = -1e30f; - for (int e = 0; e < E; e++) { - int taken = 0; for (int j = 0; j < kk; j++) if (idx[j] == e) { taken = 1; break; } - if (!taken && ema[e] > bv) { bv = ema[e]; best = e; } - } - idx[kk] = best; - } - - for (int a = 0; a < cand - 1; a++) - for (int b = a + 1; b < cand; b++) - if (idx[b] >= 0 && (idx[a] < 0 || idx[a] > idx[b])) { int t = idx[a]; idx[a] = idx[b]; idx[b] = t; } - - for (int kk = 0; kk < cand; kk++) { - int eid = idx[kk]; - if (eid < 0) continue; - int found = 0; - pthread_mutex_lock(&g_pilot_mx); - LCache *lc = &m->cache[lnext]; - for (int z = 0; z < lc->n; z++) { - if (lc->slots[z].eid == eid) { found = 1; break; } - } - pthread_mutex_unlock(&g_pilot_mx); - if (!found) { - int gidx = lnext * E + eid; - pthread_mutex_lock(&g_pilot_mx); - int already_queued = m->is_queued[gidx]; - if (!already_queued) { - m->is_queued[gidx] = 1; - } - pthread_mutex_unlock(&g_pilot_mx); - - if (!already_queued) { - unsigned w2 = __atomic_load_n(&pilot_w, __ATOMIC_RELAXED); - unsigned r2 = __atomic_load_n(&pilot_r, __ATOMIC_ACQUIRE); - if (w2 - r2 < 4096) { - pilot_q[w2 & 4095].l = lnext; - pilot_q[w2 & 4095].e = eid; - __atomic_store_n(&pilot_w, w2 + 1, __ATOMIC_RELEASE); - } else { - pthread_mutex_lock(&g_pilot_mx); - m->is_queued[gidx] = 0; - pthread_mutex_unlock(&g_pilot_mx); - } - } - } - } -} /* generazione greedy. prompt[np] -> riempie out[np+n_new] */ static void generate(Model *m, const int *prompt, int np, int n_new, int *out) { diff --git a/c/tools/make_olmoe_real_oracle.py b/c/tools/make_olmoe_real_oracle.py index 374d29f..1e638e4 100644 --- a/c/tools/make_olmoe_real_oracle.py +++ b/c/tools/make_olmoe_real_oracle.py @@ -22,29 +22,25 @@ try: except ImportError as exc: sys.exit(f"Missing deps: {exc}. Run: pip install torch transformers") -MODEL_DIR = ( - Path(r"C:\Users\egonr\.cache\huggingface\hub" - r"\models--allenai--OLMoE-1B-7B-0125-Instruct" - r"\snapshots\b89a7c4bc24fb9e55ce2543c9458ce0ca5c4650e") -) +MODEL_ID = "allenai/OLMoE-1B-7B-0125-Instruct" OUT_JSON = Path(__file__).resolve().parent.parent / "ref_olmoe_real.json" PROMPT = "The capital of France is" MAX_NEW_TOKENS = 12 -print(f"Loading tokenizer from {MODEL_DIR} ...") -tokenizer = AutoTokenizer.from_pretrained(str(MODEL_DIR)) +print(f"Loading tokenizer from {MODEL_ID} ...") +tokenizer = AutoTokenizer.from_pretrained(MODEL_ID) print("Encoding prompt ...") enc = tokenizer(PROMPT, return_tensors="pt") prompt_ids = enc["input_ids"][0].tolist() print(f" Prompt IDs ({len(prompt_ids)}): {prompt_ids}") -print(f"Loading OLMoE model from {MODEL_DIR} ...") +print(f"Loading OLMoE model from {MODEL_ID} ...") print(" (this will use ~14 GB RAM — please be patient)") model = OlmoeForCausalLM.from_pretrained( - str(MODEL_DIR), + MODEL_ID, torch_dtype=torch.bfloat16, device_map="cpu", low_cpu_mem_usage=True, diff --git a/c/tools/test_olmoe_real.py b/c/tools/test_olmoe_real.py index ac47dc2..1abf64d 100644 --- a/c/tools/test_olmoe_real.py +++ b/c/tools/test_olmoe_real.py @@ -22,8 +22,9 @@ if sys.platform == "win32": pass HERE = Path(__file__).resolve().parent.parent -ENGINE = HERE / "olmoe.exe" -SNAP = HERE / "olmoe_i4" +ext = ".exe" if sys.platform == "win32" else "" +ENGINE = HERE / f"olmoe{ext}" +SNAP = os.getenv("SNAP", str(HERE.parent / "olmoe_merged")) REF_OUT = HERE / "ref_olmoe_real.json" BOOTSTRAP_REF = HERE / "ref_olmoe_bootstrap.json" From 1ac2e7b487ad061f849b49f4905a3a11787f15eb Mon Sep 17 00:00:00 2001 From: Egon Ruiter Date: Thu, 16 Jul 2026 14:36:53 +0200 Subject: [PATCH 08/22] fix(prefetch): address copilot code quality reviews on safety and concurrency --- c/olmoe.c | 70 ++++++++++++--------------------- c/tools/convert_olmoe_merged.py | 3 +- 2 files changed, 27 insertions(+), 46 deletions(-) diff --git a/c/olmoe.c b/c/olmoe.c index 35e449a..36fc0b0 100644 --- a/c/olmoe.c +++ b/c/olmoe.c @@ -54,8 +54,7 @@ typedef struct { * meccanismo che fa stare GLM-5.2 nei 15 GB. dequant-on-use nel matmul. */ /* IMPROVEMENT 2: pinned=1 means this slot is never evicted (hot expert). */ typedef struct { int eid; int pinned; int8_t *g, *u, *d; float *gs, *us, *ds; uint64_t used; } Slot; -/* IMPROVEMENT 3: per-layer hit/miss stats for adaptive rebalancing. */ -typedef struct { Slot *slots; int n, cap; uint64_t layer_hits, layer_miss; } LCache; +typedef struct { Slot *slots; int n, cap; } LCache; typedef struct { Cfg c; @@ -70,8 +69,7 @@ typedef struct { /* IMPROVEMENT 2: expert frequency heatmap */ uint32_t *freq; int freq_token_count, hot_pinned, hot_n, warmup_tokens; - /* IMPROVEMENT 3: adaptive rebalance */ - int token_count, rebal_interval, total_cap; + int token_count; /* PREDICTION IMPROVEMENT A: per-layer smoothed gate logits across tokens. * momentum_logits[l*E .. (l+1)*E-1] = EMA of recent gate outputs. * Blended with fresh gate prediction: final = (1-smooth)*fresh + smooth*ema. @@ -99,7 +97,11 @@ static void ensure_pilot_worker_started(Model *m) { if (!pilot_m) { pilot_m = m; pthread_t t; - pthread_create(&t, NULL, pilot_worker, NULL); + if (pthread_create(&t, NULL, pilot_worker, NULL) != 0) { + fprintf(stderr, "Error: Failed to create pilot prefetch worker thread\n"); + exit(1); + } + pthread_detach(t); } } @@ -263,7 +265,6 @@ static void model_init(Model *m, const char *snap, int cap, int bits) { LD(gate, "mlp.gate.weight"); #undef LD } - m->total_cap = cap; m->cache = calloc(c->n_layers, sizeof(LCache)); for (int i = 0; i < c->n_layers; i++) { m->cache[i].cap = cap; @@ -274,8 +275,6 @@ static void model_init(Model *m, const char *snap, int cap, int bits) { m->hot_pinned = 0; m->freq_token_count = 0; m->hot_n = getenv("HOT") ? atoi(getenv("HOT")) : 0; m->warmup_tokens = getenv("WARMUP") ? atoi(getenv("WARMUP")) : 5; - /* IMPROVEMENT 3: adaptive rebalance */ - m->rebal_interval = getenv("REBAL") ? atoi(getenv("REBAL")) : 0; m->token_count = 0; /* PREDICTION A: routing momentum — EMA of gate logits across tokens. * Initialized to zero; first token sets EMA = fresh logits. */ @@ -309,7 +308,9 @@ static void model_init(Model *m, const char *snap, int cap, int bits) { unsigned r = __atomic_load_n(&pilot_r, __ATOMIC_ACQUIRE); if (w - r < 4096) { pilot_q[w & 4095].l = l; pilot_q[w & 4095].e = e; + pthread_mutex_lock(&g_pilot_mx); m->is_queued[l * c->n_experts + e] = 1; + pthread_mutex_unlock(&g_pilot_mx); __atomic_store_n(&pilot_w, w + 1, __ATOMIC_RELEASE); } } @@ -330,13 +331,16 @@ static void model_init(Model *m, const char *snap, int cap, int bits) { } } - static void slot_ensure_allocated(Model *m, Slot *s) { if (s->g) return; Cfg *c = &m->c; int64_t ng = (int64_t)c->inter * c->hidden; int64_t nd = (int64_t)c->hidden * c->inter; int8_t *w_block = malloc(ng + ng + nd); + if (!w_block) { + fprintf(stderr, "Error: Out of memory allocating slot weights block\n"); + exit(1); + } s->g = w_block; s->u = w_block + ng; s->d = w_block + ng + ng; @@ -360,11 +364,11 @@ static void expert_get(Model *m, int layer, int eid, Slot **out) { LCache *lc = &m->cache[layer]; pthread_mutex_lock(&g_pilot_mx); for (int i = 0; i < lc->n; i++) if (lc->slots[i].eid == eid) { - m->hits++; lc->layer_hits++; lc->slots[i].used = ++m->clock; *out = &lc->slots[i]; + m->hits++; lc->slots[i].used = ++m->clock; *out = &lc->slots[i]; pthread_mutex_unlock(&g_pilot_mx); return; } - m->miss++; lc->layer_miss++; + m->miss++; Cfg *c = &m->c; Slot *s; if (lc->n < lc->cap) { @@ -416,6 +420,7 @@ static void pin_hot_experts(Model *m) { if (max_pin < 4) max_pin = 4; int hn = is_dynamic ? max_pin : (m->hot_n < c->n_experts ? m->hot_n : c->n_experts); + if (hn > 256) hn = 256; int hot_eids[256]; int actual_hn = 0; @@ -463,29 +468,6 @@ static void pin_hot_experts(Model *m) { } } -/* ---------- IMPROVEMENT 3: adaptive per-layer cache rebalancing ---------- */ -static void rebalance_cache(Model *m) { - Cfg *c = &m->c; - uint64_t total_miss = 0; - for (int l = 0; l < c->n_layers; l++) total_miss += m->cache[l].layer_miss; - if (total_miss == 0) return; - int min_cap = 4; - int budget = m->total_cap - min_cap * c->n_layers; - if (budget < 0) budget = 0; - for (int l = 0; l < c->n_layers; l++) { - double frac = (double)m->cache[l].layer_miss / (double)total_miss; - int new_cap = min_cap + (int)(frac * budget + 0.5); - LCache *lc = &m->cache[l]; - if (new_cap > lc->cap) { - Slot *ns = realloc(lc->slots, new_cap * sizeof(Slot)); - if (ns) { - memset(ns + lc->cap, 0, (new_cap - lc->cap) * sizeof(Slot)); - lc->slots = ns; lc->cap = new_cap; - } - } - lc->layer_hits = 0; lc->layer_miss = 0; - } -} /* ---------- RoPE su un vettore di una testa (head_dim) a posizione assoluta pos ---------- */ static void rope_head(float *x, int pos, const Cfg *c) { @@ -607,7 +589,9 @@ static float *step(Model *m, const int *ids, int S, int pos_base) { if (g_pilot && m->token_count > 0) { unsigned r = __atomic_load_n(&pilot_r, __ATOMIC_ACQUIRE); __atomic_store_n(&pilot_w, r, __ATOMIC_RELEASE); + pthread_mutex_lock(&g_pilot_mx); memset(m->is_queued, 0, (size_t)c->n_layers * c->n_experts); + pthread_mutex_unlock(&g_pilot_mx); } float *x = falloc((int64_t)S*D); for (int s = 0; s < S; s++) memcpy(x + (int64_t)s*D, m->embed + (int64_t)ids[s]*D, D*sizeof(float)); @@ -636,9 +620,6 @@ static float *step(Model *m, const int *ids, int S, int pos_base) { m->token_count++; m->freq_token_count++; if (!m->hot_pinned && m->hot_n > 0 && m->freq_token_count >= m->warmup_tokens) pin_hot_experts(m); - /* IMPROVEMENT 3: periodic adaptive rebalance */ - if (m->rebal_interval > 0 && m->token_count % m->rebal_interval == 0) - rebalance_cache(m); m->kv_len = pos_base + S; float *last = falloc(D); rmsnorm_row(last, x + (int64_t)(S-1)*D, m->final_norm, D, c->eps); @@ -711,9 +692,6 @@ static void *pilot_worker(void *arg) { static void pilot_prefetch(Model *m, int lnext, const float *x, int S) { if (lnext < 0 || lnext >= m->c.n_layers) return; Cfg *c = &m->c; int D = c->hidden, E = c->n_experts; - /* IMPROVEMENT 4: wide prefetch — top K * g_wide candidates */ - int cand = c->topk * g_wide; - if (cand > E) cand = E; ensure_pilot_worker_started(m); float *logits = falloc((int64_t)S * E); Layer *l = &m->L[lnext]; @@ -754,16 +732,19 @@ static void pilot_prefetch(Model *m, int lnext, const float *x, int S) { int cand = 0; int idx[128]; + float max_logit = -1e30f; + for (int e = 0; e < E; e++) { if (blended[e] > max_logit) max_logit = blended[e]; } float *exps = falloc(E); float sum_exps = 0.f; for (int e = 0; e < E; e++) { - exps[e] = expf(blended[e]); + exps[e] = expf(blended[e] - max_logit); sum_exps += exps[e]; } float cum_sum = 0.f; int min_cand = c->topk; - int max_cand = c->topk * 2; + int max_cand = c->topk * g_wide; + if (max_cand < min_cand) max_cand = min_cand; if (max_cand > E) max_cand = E; for (int kk = 0; kk < max_cand; kk++) { @@ -867,7 +848,6 @@ int main(int argc, char **argv) { if (g_wide < 1) g_wide = 1; if (g_wide > 4) g_wide = 4; int hot_n = getenv("HOT") ? atoi(getenv("HOT")) : 0; - int rebal = getenv("REBAL") ? atoi(getenv("REBAL")) : 0; int cap = argc > 1 ? atoi(argv[1]) : 16; int bits = argc > 2 ? atoi(argv[2]) : 8; if (bits < 2 || bits > 8) { @@ -879,8 +859,8 @@ int main(int argc, char **argv) { float smooth = getenv("SMOOTH") ? (float)atof(getenv("SMOOTH")) : 0.3f; float conf = getenv("CONF_LIMIT") ? (float)atof(getenv("CONF_LIMIT")) : 0.92f; - printf("== Streaming C engine v2.2 | cache=%d/layer bits=%d pilot=%d wide=%d hot=%d rebal=%d smooth=%.2f conf=%.2f ==\n", - cap, bits, g_pilot, g_wide, hot_n, rebal, smooth, conf); + printf("== Streaming C engine v2.2 | cache=%d/layer bits=%d pilot=%d wide=%d hot=%d smooth=%.2f conf=%.2f ==\n", + cap, bits, g_pilot, g_wide, hot_n, smooth, conf); FILE *f = fopen(refpath, "rb"); if (!f) { perror(refpath); return 1; } fseek(f,0,SEEK_END); long n=ftell(f); fseek(f,0,SEEK_SET); diff --git a/c/tools/convert_olmoe_merged.py b/c/tools/convert_olmoe_merged.py index b39b123..883b8df 100644 --- a/c/tools/convert_olmoe_merged.py +++ b/c/tools/convert_olmoe_merged.py @@ -20,8 +20,9 @@ if sys.platform == "win32": try: import torch from safetensors.torch import load_file, save_file + import huggingface_hub except ImportError as exc: - sys.exit(f"Missing dependencies: {exc}. Install: pip install torch safetensors") + sys.exit(f"Missing dependencies: {exc}. Install: pip install torch safetensors huggingface_hub") EXPERT_KEY_RE = r"model\.layers\.(\d+)\.mlp\.experts\.(\d+)\.(gate_proj|up_proj|down_proj)\.weight" From 2d8d2951ee7d93cf2adad7b89b56c0adc1dbd0c5 Mon Sep 17 00:00:00 2001 From: Egon Ruiter Date: Thu, 16 Jul 2026 15:37:20 +0200 Subject: [PATCH 09/22] fix(prefetch): address second round of copilot review comments - Fix ENV VARS header: document PILOT=0-3, SMOOTH, CONF_LIMIT; remove stale REBAL entry - Fix per-layer EMA: apply routing momentum to all layers (not just layer 0) with correct offset - Fix in-flight slot race in expert_get: LRU eviction now skips slots with eid==-1 (being loaded) - Fix in-flight slot race in pilot_realload: same fix, prevents concurrent writes into active slot - Fix idx[] buffer overflow: clamp max_cand to 128 before E in pilot_prefetch --- c/olmoe.c | 30 ++++++++++++++++-------------- 1 file changed, 16 insertions(+), 14 deletions(-) diff --git a/c/olmoe.c b/c/olmoe.c index 36fc0b0..51e32ae 100644 --- a/c/olmoe.c +++ b/c/olmoe.c @@ -7,12 +7,13 @@ * Matmul multi-thread con OpenMP (niente BLAS). * * ENV VARS: - * PILOT=0/1/2 : 0=no prefetch, 1=1-layer lookahead, 2=2-layer lookahead [IMPROVEMENT 1] - * HOT=N : pin top-N hot experts per layer permanently (never evict) [IMPROVEMENT 2] - * WARMUP=N : tokens before hot pinning activates (default 5) [IMPROVEMENT 2] - * REBAL=N : rebalance cache per-layer every N tokens (0=off) [IMPROVEMENT 3] - * WIDE=N : prefetch top-K*N candidates (default 1, try 2 or 3) [IMPROVEMENT 4] - * (expert queue is sorted by eid for SSD locality) [IMPROVEMENT 5] + * PILOT=0/1/2/3 : 0=no prefetch, 1=1-layer lookahead, 2=2-layer, 3=3-layer lookahead + * HOT=N : pin top-N hot experts per layer permanently (never evict) + * WARMUP=N : tokens before hot pinning activates (default 5) + * WIDE=N : prefetch top-K*N candidates (default 1, try 2 or 3) + * SMOOTH=F : EMA coefficient for routing momentum (default 0.3, range 0.0-0.95) + * CONF_LIMIT=F : cumulative gate probability threshold for prefetch cutoff (default 0.92) + * (expert queue is sorted by eid for SSD read locality) */ #define _GNU_SOURCE #include @@ -375,13 +376,13 @@ static void expert_get(Model *m, int layer, int eid, Slot **out) { s = &lc->slots[lc->n++]; slot_ensure_allocated(m, s); } else { - /* IMPROVEMENT 2: LRU eviction — never evict pinned experts */ + /* LRU eviction — skip pinned and in-flight (eid==-1) slots */ int lru = -1; for (int i = 0; i < lc->n; i++) { - if (lc->slots[i].pinned) continue; + if (lc->slots[i].pinned || lc->slots[i].eid < 0) continue; if (lru < 0 || lc->slots[i].used < lc->slots[lru].used) lru = i; } - if (lru < 0) lru = 0; /* all pinned: fallback evict oldest */ + if (lru < 0) lru = 0; /* all pinned/in-flight: fallback evict oldest */ s = &lc->slots[lru]; s->pinned = 0; } @@ -539,8 +540,8 @@ static void moe(Model *m, Layer *l, int layer, float *x, int S, float *out) { float *g = falloc(I), *u = falloc(I), *hh = falloc(D); for (int s = 0; s < S; s++) { float *pr = logits + (int64_t)s*E; - if (layer == 0 && m->momentum_logits) { - float *ema = m->momentum_logits; + if (m->momentum_logits && m->pilot_smooth > 0.f) { + float *ema = m->momentum_logits + (int64_t)layer * E; int is_zero = 1; for (int e = 0; e < E; e++) { if (ema[e] != 0.f) { is_zero = 0; break; } } if (is_zero) { @@ -646,16 +647,16 @@ static void pilot_realload(Model *m, int layer, int eid) { s = &lc->slots[lc->n++]; slot_ensure_allocated(m, s); } else { - /* IMPROVEMENT 2: never evict pinned experts */ + /* LRU eviction — skip pinned and in-flight (eid==-1) slots */ int lru = -1; for (int i = 0; i < lc->n; i++) { - if (lc->slots[i].pinned) continue; + if (lc->slots[i].pinned || lc->slots[i].eid < 0) continue; if (lru < 0 || lc->slots[i].used < lc->slots[lru].used) lru = i; } if (lru < 0) { m->is_queued[layer * c->n_experts + eid] = 0; pthread_mutex_unlock(&g_pilot_mx); - return; /* all pinned, skip */ + return; /* all pinned/in-flight, skip */ } s = &lc->slots[lru]; s->pinned = 0; } @@ -745,6 +746,7 @@ static void pilot_prefetch(Model *m, int lnext, const float *x, int S) { int min_cand = c->topk; int max_cand = c->topk * g_wide; if (max_cand < min_cand) max_cand = min_cand; + if (max_cand > 128) max_cand = 128; /* idx[] buffer bound */ if (max_cand > E) max_cand = E; for (int kk = 0; kk < max_cand; kk++) { From b6bae91b66a14647659d6c7c6a6c68adbf560564 Mon Sep 17 00:00:00 2001 From: Egon Ruiter Date: Thu, 16 Jul 2026 15:46:26 +0200 Subject: [PATCH 10/22] fix(prefetch): address third round of copilot review comments - Fix LRU fallback: when all evictable slots are in-flight, find oldest non-in-flight slot (pinned ok) before falling back to slot 0 - Fix pin_hot_experts: guard enqueue behind g_pilot>0, call ensure_pilot_worker_started(), and set is_queued flag to prevent duplicate in-flight loads from pilot_prefetch() - Fix token counting: increment token_count/freq_token_count by S (batch size) instead of 1 so prefill tokens are counted accurately and warmup threshold triggers at the right time --- c/olmoe.c | 25 ++++++++++++++++++++----- 1 file changed, 20 insertions(+), 5 deletions(-) diff --git a/c/olmoe.c b/c/olmoe.c index 51e32ae..41cf776 100644 --- a/c/olmoe.c +++ b/c/olmoe.c @@ -382,7 +382,15 @@ static void expert_get(Model *m, int layer, int eid, Slot **out) { if (lc->slots[i].pinned || lc->slots[i].eid < 0) continue; if (lru < 0 || lc->slots[i].used < lc->slots[lru].used) lru = i; } - if (lru < 0) lru = 0; /* all pinned/in-flight: fallback evict oldest */ + if (lru < 0) { + /* All slots are pinned or in-flight; find oldest non-in-flight slot + * (may be pinned, but never select one currently being loaded). */ + for (int i = 0; i < lc->n; i++) { + if (lc->slots[i].eid < 0) continue; /* never evict in-flight */ + if (lru < 0 || lc->slots[i].used < lc->slots[lru].used) lru = i; + } + } + if (lru < 0) lru = 0; /* absolute last resort: all in-flight, evict slot 0 */ s = &lc->slots[lru]; s->pinned = 0; } @@ -449,13 +457,20 @@ static void pin_hot_experts(Model *m) { if (lc->slots[i].eid == eid) { lc->slots[i].pinned = 1; found = 1; break; } } pthread_mutex_unlock(&g_pilot_mx); - if (!found) { + if (!found && g_pilot > 0) { + /* Only enqueue when the prefetch worker is active (PILOT>0). */ + ensure_pilot_worker_started(m); unsigned w = __atomic_load_n(&pilot_w, __ATOMIC_RELAXED); unsigned r = __atomic_load_n(&pilot_r, __ATOMIC_ACQUIRE); - if (w - r < 4096) { + int gidx = l * c->n_experts + eid; + pthread_mutex_lock(&g_pilot_mx); + int already = m->is_queued[gidx]; + if (!already && w - r < 4096) { pilot_q[w & 4095].l = l; pilot_q[w & 4095].e = eid; + m->is_queued[gidx] = 1; __atomic_store_n(&pilot_w, w + 1, __ATOMIC_RELEASE); } + pthread_mutex_unlock(&g_pilot_mx); } pinned_total++; } @@ -617,8 +632,8 @@ static float *step(Model *m, const int *ids, int S, int pos_base) { pilot_prefetch(m, i + 3, x, S); } - /* IMPROVEMENT 2: count tokens; trigger hot pinning after warmup */ - m->token_count++; m->freq_token_count++; + /* count actual tokens processed (S>1 during prefill) */ + m->token_count += S; m->freq_token_count += S; if (!m->hot_pinned && m->hot_n > 0 && m->freq_token_count >= m->warmup_tokens) pin_hot_experts(m); m->kv_len = pos_base + S; From c769e04d136ad8aa48dc79745363353a263ad179 Mon Sep 17 00:00:00 2001 From: Egon Ruiter Date: Thu, 16 Jul 2026 15:58:38 +0200 Subject: [PATCH 11/22] fix(prefetch): address fourth round of copilot review comments - Fix queue flush race: clear is_queued under mutex only; never move pilot_w backwards (would break r<=w ring-buffer invariant). Worker skips stale entries via new is_queued guard at start of pilot_realload - Add early-exit in pilot_realload when is_queued==0 (entry flushed between enqueue and worker pickup), preventing unnecessary loads - Fix misleading EMA struct comment: momentum_logits is used only by the PILOT prefetcher, not blended into actual MoE routing decisions - Fix Slot pinned comment: 'never evicted' was too strong; clarify that pinned slots may be displaced under extreme all-pinned cache pressure - Use st_read_f32() for scale tensor (.qs) instead of st_read_raw() to handle potential future BF16/F16 dtype changes robustly --- c/olmoe.c | 28 ++++++++++++++++++++-------- 1 file changed, 20 insertions(+), 8 deletions(-) diff --git a/c/olmoe.c b/c/olmoe.c index 41cf776..21b07e8 100644 --- a/c/olmoe.c +++ b/c/olmoe.c @@ -53,7 +53,9 @@ typedef struct { * Ogni weight [out,in] tenuto come int8 (per-riga) + scala float per riga. * Cosi' la RAM-cache scende da 4 byte/param (f32) a 1 byte/param: e' il * meccanismo che fa stare GLM-5.2 nei 15 GB. dequant-on-use nel matmul. */ -/* IMPROVEMENT 2: pinned=1 means this slot is never evicted (hot expert). */ +/* pinned=1 means this slot is strongly preferred to keep (hot expert); it will + * not be evicted during normal LRU eviction, but may be displaced under extreme + * cache pressure when all slots are pinned or in-flight. */ typedef struct { int eid; int pinned; int8_t *g, *u, *d; float *gs, *us, *ds; uint64_t used; } Slot; typedef struct { Slot *slots; int n, cap; } LCache; @@ -71,10 +73,10 @@ typedef struct { uint32_t *freq; int freq_token_count, hot_pinned, hot_n, warmup_tokens; int token_count; - /* PREDICTION IMPROVEMENT A: per-layer smoothed gate logits across tokens. - * momentum_logits[l*E .. (l+1)*E-1] = EMA of recent gate outputs. - * Blended with fresh gate prediction: final = (1-smooth)*fresh + smooth*ema. - * Captures routing consistency across tokens (same token tends to reuse experts). */ + /* PREDICTION IMPROVEMENT A: per-layer EMA of gate logits across tokens. + * momentum_logits[l*E .. (l+1)*E-1] = EMA of gate outputs for layer l. + * Used exclusively by the PILOT prefetcher to stabilise routing predictions + * across tokens; does NOT affect actual MoE routing (pr is unchanged). */ float *momentum_logits; /* [n_layers * n_experts], EMA of gate logits */ float pilot_smooth; /* SMOOTH env: EMA coefficient 0.0-0.9 (default 0.3) */ uint8_t *is_pinned; /* [n_layers * n_experts], 1 if expert is globally pinned */ @@ -357,7 +359,7 @@ static void load_expert_merged(Model *m, int layer, int eid, Slot *s) { snprintf(nm, sizeof(nm), "model.layers.%d.mlp.experts.%d.merged_weight", layer, eid); snprintf(qsnm, sizeof(qsnm), "model.layers.%d.mlp.experts.%d.qs", layer, eid); st_read_raw(&m->S, nm, s->g, 1); - st_read_raw(&m->S, qsnm, s->gs, 1); + st_read_f32(&m->S, qsnm, s->gs, 0); /* scales are F32; use typed reader for dtype safety */ } /* ---------- cache expert: ritorna i pesi quantizzati (q+scale) da cache o disco ---------- */ @@ -603,8 +605,13 @@ static void moe(Model *m, Layer *l, int layer, float *x, int S, float *out) { static float *step(Model *m, const int *ids, int S, int pos_base) { Cfg *c = &m->c; int D = c->hidden; if (g_pilot && m->token_count > 0) { - unsigned r = __atomic_load_n(&pilot_r, __ATOMIC_ACQUIRE); - __atomic_store_n(&pilot_w, r, __ATOMIC_RELEASE); + /* Flush stale prefetch requests: clear is_queued so pilot_realload + * will skip any entries still sitting in pilot_q for the previous + * token. We deliberately do NOT move pilot_w backwards; that would + * break the ring-buffer invariant (pilot_r could exceed pilot_w if + * the worker consumed an entry concurrently). The worker will drain + * the stale slots harmlessly because pilot_realload already exits + * early when the expert is already cached or is_queued is clear. */ pthread_mutex_lock(&g_pilot_mx); memset(m->is_queued, 0, (size_t)c->n_layers * c->n_experts); pthread_mutex_unlock(&g_pilot_mx); @@ -650,6 +657,11 @@ static void pilot_realload(Model *m, int layer, int eid) { Cfg *c = &m->c; pthread_mutex_lock(&g_pilot_mx); + /* Early-exit if entry was flushed (is_queued cleared) while waiting. */ + if (!m->is_queued[layer * c->n_experts + eid]) { + pthread_mutex_unlock(&g_pilot_mx); + return; + } for (int i = 0; i < lc->n; i++) { if (lc->slots[i].eid == eid) { m->is_queued[layer * c->n_experts + eid] = 0; From d86a6b93ad4f70045138a13907682b5f936c3355 Mon Sep 17 00:00:00 2001 From: ZacharyZcR Date: Sun, 19 Jul 2026 06:10:14 +0800 Subject: [PATCH 12/22] packaging: pyproject.toml + editorconfig + clang-format Python packaging: - pyproject.toml: pip install colibri-engine (editable dev install works) - colibri/ package with __version__, CLI entry point delegating to c/coli - Optional dependency groups: [convert] (numpy, huggingface_hub), [oracle] (torch, transformers, safetensors), [bench] (tokenizers, datasets) Code style: - .editorconfig: consistent indent/charset across all file types - .clang-format: LLVM-based, 120 col, matches existing engine style Usage: pip install -e . # dev install (CLI + serve, no heavy deps) pip install -e .[convert] # adds converter dependencies pip install -e .[oracle] # adds torch/transformers for oracle validation --- .clang-format | 10 +++++++++ .editorconfig | 24 ++++++++++++++++++++ colibri/__init__.py | 5 +++++ colibri/_version.py | 3 +++ colibri/cli.py | 30 +++++++++++++++++++++++++ pyproject.toml | 53 +++++++++++++++++++++++++++++++++++++++++++++ 6 files changed, 125 insertions(+) create mode 100644 .clang-format create mode 100644 .editorconfig create mode 100644 colibri/__init__.py create mode 100644 colibri/_version.py create mode 100644 colibri/cli.py create mode 100644 pyproject.toml diff --git a/.clang-format b/.clang-format new file mode 100644 index 0000000..f911652 --- /dev/null +++ b/.clang-format @@ -0,0 +1,10 @@ +BasedOnStyle: LLVM +IndentWidth: 4 +ColumnLimit: 120 +AllowShortFunctionsOnASingleLine: All +AllowShortIfStatementsOnASingleLine: AllIfsAndElse +AllowShortLoopsOnASingleLine: true +BreakBeforeBraces: Attach +PointerAlignment: Right +SpaceAfterCStyleCast: false +SortIncludes: false diff --git a/.editorconfig b/.editorconfig new file mode 100644 index 0000000..e0acabb --- /dev/null +++ b/.editorconfig @@ -0,0 +1,24 @@ +root = true + +[*] +charset = utf-8 +end_of_line = lf +insert_final_newline = true +trim_trailing_whitespace = true +indent_style = space +indent_size = 4 + +[*.{c,h,cu}] +indent_size = 4 + +[*.{ts,tsx,js,json,css}] +indent_size = 2 + +[Makefile] +indent_style = tab + +[*.yml] +indent_size = 2 + +[*.md] +trim_trailing_whitespace = false diff --git a/colibri/__init__.py b/colibri/__init__.py new file mode 100644 index 0000000..77f7061 --- /dev/null +++ b/colibri/__init__.py @@ -0,0 +1,5 @@ +"""colibrì — tiny engine, immense model.""" + +from colibri._version import __version__ + +__all__ = ["__version__"] diff --git a/colibri/_version.py b/colibri/_version.py new file mode 100644 index 0000000..9951ca1 --- /dev/null +++ b/colibri/_version.py @@ -0,0 +1,3 @@ +"""Single source of truth for the colibrì version number.""" + +__version__ = "1.0.0" diff --git a/colibri/cli.py b/colibri/cli.py new file mode 100644 index 0000000..3f19234 --- /dev/null +++ b/colibri/cli.py @@ -0,0 +1,30 @@ +"""Entry point for `coli` when installed via pip. + +Delegates to the original c/coli script which handles all subcommands. +This wrapper exists so `pip install colibri-engine` creates a `coli` console +script that works without the user having to add c/ to PATH manually. +""" + +import os +import sys +import runpy + + +def main(): + here = os.path.dirname(os.path.abspath(__file__)) + engine_dir = os.path.join(os.path.dirname(here), "c") + coli_script = os.path.join(engine_dir, "coli") + + if not os.path.exists(coli_script): + sys.exit( + "colibri engine directory not found.\n" + "Install from source: git clone + pip install -e ." + ) + + sys.path.insert(0, engine_dir) + sys.argv[0] = coli_script + runpy.run_path(coli_script, run_name="__main__") + + +if __name__ == "__main__": + main() diff --git a/pyproject.toml b/pyproject.toml new file mode 100644 index 0000000..18f8f85 --- /dev/null +++ b/pyproject.toml @@ -0,0 +1,53 @@ +[build-system] +requires = ["setuptools>=68.0"] +build-backend = "setuptools.build_meta" + +[project] +name = "colibri-engine" +dynamic = ["version"] +description = "Tiny engine, immense model — run GLM-5.2 (744B MoE) locally" +readme = "README.md" +license = "Apache-2.0" +requires-python = ">=3.10" +authors = [ + {name = "JustVugg"}, +] +classifiers = [ + "Development Status :: 4 - Beta", + "Environment :: Console", + "Intended Audience :: Science/Research", + "Operating System :: POSIX :: Linux", + "Operating System :: MacOS", + "Operating System :: Microsoft :: Windows", + "Programming Language :: Python :: 3", + "Topic :: Scientific/Engineering :: Artificial Intelligence", +] + +[project.optional-dependencies] +convert = [ + "numpy", + "huggingface_hub", +] +oracle = [ + "torch>=2.0", + "transformers>=4.40", + "safetensors", +] +bench = [ + "tokenizers", + "datasets", +] + +[project.scripts] +coli = "colibri.cli:main" + +[project.urls] +Homepage = "https://github.com/JustVugg/colibri" +Issues = "https://github.com/JustVugg/colibri/issues" + +[tool.setuptools.dynamic] +version = {attr = "colibri._version.__version__"} + +[tool.setuptools.packages.find] +where = ["."] +include = ["colibri*"] From ca39e5333f1addb67df62b836ee2d9e661fd91e4 Mon Sep 17 00:00:00 2001 From: JustVugg Date: Sun, 19 Jul 2026 12:38:11 +0200 Subject: [PATCH 13/22] packaging: single version source + honest editable-install semantics (on top of #396) colibri/_version.py now reads c/version.py (#394's single source of truth -- coli --version, the release workflow, and pip metadata can no longer drift), with an importlib.metadata fallback for the installed-wheel case where c/ is not on disk. README documents that pip install -e . is the supported form: the engine lives in c/ and is not packaged into a standalone wheel. Verified in a clean venv: pip install -e . -> colibri.__version__ == 1.0.0 read from c/version.py, coli entrypoint on PATH and functional. Co-Authored-By: Claude Fable 5 --- README.md | 4 ++++ colibri/_version.py | 24 ++++++++++++++++++++++-- 2 files changed, 26 insertions(+), 2 deletions(-) diff --git a/README.md b/README.md index 797f1ca..e5d1651 100644 --- a/README.md +++ b/README.md @@ -183,6 +183,10 @@ COLI_MODEL=/nvme/glm52_i4 ./coli doctor # read-only readiness check The engine at runtime is pure C — python is only used by the one-time converter and the optional API gateway. +Prefer a `coli` command on your PATH? From a checkout, `pip install -e .` +registers it (the engine itself still lives in `c/` — this is an editable +install from the clone, not a standalone wheel). + ### 3. Go deeper | topic | doc | diff --git a/colibri/_version.py b/colibri/_version.py index 9951ca1..174442a 100644 --- a/colibri/_version.py +++ b/colibri/_version.py @@ -1,3 +1,23 @@ -"""Single source of truth for the colibrì version number.""" +"""Version accessor for the pip package. -__version__ = "1.0.0" +The single source of truth is c/version.py (#394): coli --version and the +GitHub Release workflow read it, so the pip metadata must read the SAME file +instead of carrying a second literal that would drift on the first bump. + +From a checkout (the supported install: `pip install -e .`) the file is read +directly. From an installed wheel c/ is not on disk, so fall back to the +package metadata that setuptools baked at build time from that same file. +""" +from pathlib import Path + +try: + _ns = {} + exec((Path(__file__).resolve().parent.parent / "c" / "version.py").read_text(), _ns) + __version__ = _ns["__version__"] +except OSError: + from importlib.metadata import PackageNotFoundError, version + + try: + __version__ = version("colibri-engine") + except PackageNotFoundError: + __version__ = "0.0.0+unknown" From 72e36772f51605fa947efe380614c85a395c556c Mon Sep 17 00:00:00 2001 From: JustVugg Date: Sun, 19 Jul 2026 13:05:56 +0200 Subject: [PATCH 14/22] security: reject malformed model tensors at the untrusted-mirror boundary Colibri loads model directories and safetensors from mirrors it does not control, so the file's declared shapes and byte spans are attacker-influenced input. Three memory-safety holes on that boundary, independently confirmed (incl. a from-scratch adversarial audit that re-derived the same two) and present in the shipped v1.0.0: - st.h st_read_f32: numel came from the shape, nbytes from the offsets, with no cross-check. A crafted tensor whose shape inflates numel past nbytes made the BF16/F16 loop read past the malloc'd raw buffer and the F32 memcpy write past the caller's config-sized destination (heap OOB read + write). Now enforce numel*esz == nbytes before any copy. - st.h header parse: the shape product could overflow int64 to a small/negative numel that would then pass the cross-check. Guard each multiply. - glm.c qt_resolve_fmt (new, replaces the three duplicated "?1:?2:3" fmt sites in qt_from_disk and both expert_load arms): the old inference SILENTLY fell to int2 for any unrecognized weight byte count, so a too-short weight became a valid int2 whose matmul read O*I nibbles past the buffer; and an oversized scale array overflowed the per-row t->s. Now the weight bytes must match a known int8/int4/int2 layout and the scale array must match the expected per-row (O) or grouped (O*ng) cardinality, else refuse. - glm.c config/generation_config slurp: unbounded ftell -> malloc(n+1) gave a hostile file a load-time OOM, and on malloc failure b[got]=0 was a NULL deref. Cap at 256 MB and NULL-check. Verified: TF token-exactness unchanged on every quant format (full-precision 32/32, int4 11/32, int2 1/32, mix 5/32 -- byte-identical to the pre-change binary); fmt=4 grouped path preserved (the scale check is by construction the same condition detect_group_size already imposed); a hand-crafted hostile safetensors is refused cleanly; ASan+UBSan clean on legit and hostile loads (only the pre-existing intentional startup leaks remain). These are the C trust-boundary items of #368, landed as a minimal standalone fix; the server-side and build items of that PR follow via its rebase. Co-Authored-By: Claude Fable 5 --- c/glm.c | 51 ++++++++++++++++++++++++++++++++++++--------------- c/st.h | 23 ++++++++++++++++++++++- 2 files changed, 58 insertions(+), 16 deletions(-) diff --git a/c/glm.c b/c/glm.c index a3e7ab7..26e7fc4 100644 --- a/c/glm.c +++ b/c/glm.c @@ -1336,7 +1336,12 @@ static jval* cfg_root(const char *snap, char **arena){ char p[2048]; snprintf(p,sizeof(p),"%s/config.json",snap); FILE *f=fopen(p,"rb"); if(!f){perror(p);exit(1);} fseek(f,0,SEEK_END); long n=ftell(f); fseek(f,0,SEEK_SET); - char *b=malloc(n+1); size_t got=fread(b,1,n,f); b[got]=0; fclose(f); + /* SEC: config.json arriva dalla dir modello non fidata. Limita la dimensione + * (un file ostile enorme = OOM al load) e controlla la malloc: senza il NULL + * check, b[got]=0 su malloc fallita era un NULL-deref. */ + if(n<0 || n>(256L<<20)){ fprintf(stderr,"%s: size %ld out of range (0..256 MB)\n",p,n); exit(1); } + char *b=malloc((size_t)n+1); if(!b){ fprintf(stderr,"OOM reading %s (%ld bytes)\n",p,n); exit(1); } + size_t got=fread(b,1,(size_t)n,f); b[got]=0; fclose(f); if((long)got!=n) fprintf(stderr,"warning: short read on %s (%ld of %ld)\n",p,(long)got,n); return json_parse(b,arena); } @@ -1375,8 +1380,9 @@ static void load_cfg(Cfg *c, const char *snap){ FILE *gf=fopen(gp,"rb"); /* assente = nessun problema: e' opzionale */ if(gf){ fseek(gf,0,SEEK_END); long gn=ftell(gf); fseek(gf,0,SEEK_SET); - if(gn>0){ - char *gb=malloc(gn+1); size_t gg=fread(gb,1,gn,gf); gb[gg]=0; + char *gb = (gn>0 && gn<=(256L<<20)) ? malloc((size_t)gn+1) : NULL; /* SEC: cap + NULL check */ + if(gb){ + size_t gg=fread(gb,1,(size_t)gn,gf); gb[gg]=0; char *ga=NULL; jval *gr=json_parse(gb,&ga); jval *ge=gr?json_get(gr,"eos_token_id"):NULL; if(ge){ @@ -1445,6 +1451,27 @@ static int detect_group_size(int O, int I, int64_t ns){ return 0; } +/* SEC: risolve e VALIDA il formato quantizzato di un tensore [O,I] letto da un + * container non fidato (mirror). L'inferenza precedente (`?1:?2:3`) cadeva su + * int2 per QUALSIASI conteggio byte non riconosciuto: un peso troppo corto + * diventava un int2 valido e il matmul leggeva oltre il buffer (O*I nibble a + * 4/byte). Qui i byte del peso devono corrispondere a un layout noto e i byte + * della scala alla cardinalita' attesa (O per-row, O*ng per-gruppo) — altrimenti + * si termina invece di sforare. Ritorna fmt (1/2/3/4) e scrive *gs. */ +static int qt_resolve_fmt(const char *name, int O, int I, int64_t nb, int64_t ns, int *gs){ + int64_t exp_i8=(int64_t)O*I, exp_i4=(int64_t)O*((I+1)/2), exp_i2=(int64_t)O*((I+3)/4); + int fmt = (nb==exp_i8)?1 : (nb==exp_i4)?2 : (nb==exp_i2)?3 : 0; + if(!fmt){ + fprintf(stderr,"%s: quantized weight is %lld bytes — no int8/int4/int2 layout for [%d,%d], refusing (untrusted container)\n", + name,(long long)nb,O,I); exit(1); } + *gs=0; + if(fmt==2){ int g=detect_group_size(O,I,ns); if(g>0){ fmt=4; *gs=g; } } + int64_t exp_scale = (fmt==4)? (int64_t)O*((I+*gs-1)/(*gs)) : (int64_t)O; /* in FLOAT */ + if(ns != exp_scale*4){ + fprintf(stderr,"%s: scale array is %lld bytes — expected %lld for [%d,%d] fmt=%d, refusing (untrusted container)\n", + name,(long long)ns,(long long)(exp_scale*4),O,I,fmt); exit(1); } + return fmt; +} /* costruisce un QT [O,I] dal disco in `t` (buffer riusabili tra chiamate). * - se esiste `name.qs`: pesi GIA' quantizzati nel container (U8 qdata + F32 scala) -> letti diretti * - altrimenti: tensore pieno (f32/bf16) -> quantizzato a runtime a `bits` (oracolo tiny / pesi pieni) @@ -1454,12 +1481,11 @@ static void qt_from_disk(Model *m, const char *name, int O, int I, int bits, int if(st_has(&m->S,sn)){ int64_t nb=st_nbytes(&m->S,name); int64_t ns=st_nbytes(&m->S,sn); /* scale bytes (F32) */ - /* Detect int4-grouped (fmt=4): packed int4 weight bytes BUT scale array is - * larger than O*4 — the group size is derived from the scale-array size. */ - int fmt = (nb==(int64_t)O*I)?1 : (nb==(int64_t)O*((I+1)/2))?2 : 3; + /* fmt=4 int4-grouped: byte int4 ma scala > O*4 — gs deriva dalla scala. + * qt_resolve_fmt valida entrambi i conteggi contro [O,I] e termina se + * non fidati (SEC). */ int gs=0; - if(fmt==2) gs=detect_group_size(O,I,ns); - if(gs>0) fmt=4; + int fmt = qt_resolve_fmt(name,O,I,nb,ns,&gs); if(fmt==1){ if(t->fmt!=1||!t->q8){ t->fmt=1; t->O=O; t->I=I; t->gs=0; t->q8=qalloc(nb); t->s=qsalloc(O); } st_read_raw(&m->S,name,t->q8,drop); } else if(fmt==4){ int ng=(I+gs-1)/gs; if(t->fmt!=4||!t->q4){ t->fmt=4; t->O=O; t->I=I; t->gs=gs; t->q4=qalloc(nb); t->s=falloc((int64_t)O*ng); } @@ -1802,11 +1828,8 @@ static int expert_load_impl(Model *m, int layer, int eid, ESlot *s, int fatal){ QT *qt[3]={&s->g,&s->u,&s->d}; int OO[3]={I,I,D}, II[3]={D,D,I}; for(int k=0;k<3;k++){ int64_t nb=tw[k]->nbytes; - int fmt=(nb==(int64_t)OO[k]*II[k])?1:(nb==(int64_t)OO[k]*((II[k]+1)/2))?2:3; - /* detect grouped int4 (fmt=4): int4 weight bytes + larger scale array */ int gs=0; - if(fmt==2) gs=detect_group_size(OO[k],II[k],tq[k]->nbytes); - if(gs>0) fmt=4; + int fmt=qt_resolve_fmt(tw[k]->name,OO[k],II[k],nb,tq[k]->nbytes,&gs); qt[k]->fmt=fmt; qt[k]->O=OO[k]; qt[k]->I=II[k]; qt[k]->gs=gs; qt[k]->qf=NULL; qt[k]->q8=(int8_t*)((char*)bw[k]+tw[k]->off); qt[k]->q4=(uint8_t*)((char*)bw[k]+tw[k]->off); qt[k]->s=(float*)((char*)bq[k]+tq[k]->off); @@ -1931,10 +1954,8 @@ static int expert_load_impl(Model *m, int layer, int eid, ESlot *s, int fatal){ QT *qt[3]={&s->g,&s->u,&s->d}; int OO[3]={I,I,D}, II[3]={D,D,I}; for(int k=0;k<3;k++){ int64_t nb=tw[k]->nbytes; - int fmt = (nb==(int64_t)OO[k]*II[k])?1 : (nb==(int64_t)OO[k]*((II[k]+1)/2))?2 : 3; int gs=0; - if(fmt==2) gs=detect_group_size(OO[k],II[k],tq[k]->nbytes); - if(gs>0) fmt=4; + int fmt=qt_resolve_fmt(tw[k]->name,OO[k],II[k],nb,tq[k]->nbytes,&gs); qt[k]->fmt=fmt; qt[k]->O=OO[k]; qt[k]->I=II[k]; qt[k]->gs=gs; qt[k]->qf=NULL; qt[k]->q8=(int8_t*)(s->slab+pos[k]); qt[k]->q4=s->slab+pos[k]; qt[k]->s=fp[k]; } diff --git a/c/st.h b/c/st.h index efa6e5a..a2c2c3f 100644 --- a/c/st.h +++ b/c/st.h @@ -200,7 +200,19 @@ static void st_init(shards *S, const char *snap_dir) { if (a0 < 0 || b0 < a0 || data_start + b0 > fsz) { fprintf(stderr, "%s: tensor '%s' data_offsets [%lld,%lld] out of file bounds (%lld)\n", files[fi], name, (long long)a0, (long long)b0, (long long)fsz); exit(1); } - int64_t numel = 1; for (int k = 0; k < shp->len; k++) numel *= (int64_t)shp->kids[k]->num; + /* SEC: lo shape viene da un file non fidato (mirror). Senza il guard + * di overflow, uno shape tipo [65535,65535,65535,...] fa avvolgere + * numel a un valore piccolo/negativo che poi passerebbe il cross-check + * numel*esz==nbytes in st_read_f32, riaprendo l'OOB. */ + int64_t numel = 1; int bad_shape = 0; + for (int k = 0; k < shp->len; k++) { + int64_t d = (int64_t)shp->kids[k]->num; + if (d < 0 || (d != 0 && numel > INT64_MAX / d)) { bad_shape = 1; break; } + numel *= d; + } + if (bad_shape) { + fprintf(stderr, "%s: tensor '%s' shape overflows int64 — refusing (hostile or corrupt file)\n", + files[fi], name); exit(1); } if (S->n == S->cap) { S->cap *= 2; S->t = realloc(S->t, S->cap*sizeof(st_tensor)); } st_tensor *t = &S->t[S->n++]; t->name = strdup(name); t->fd = fd; t->off = data_start + a0; @@ -249,6 +261,15 @@ static void st_prefetch(shards *S, const char *name) { static int64_t st_read_f32(shards *S, const char *name, float *out, int drop) { st_tensor *t = st_find(S, name); if (!t) { fprintf(stderr, "missing tensor: %s\n", name); exit(1); } + /* SEC: numel viene dallo shape, nbytes dagli offset — due campi indipendenti + * del file. Se non concordano, la memcpy F32 (nbytes) o i loop BF16/F16 + * (numel elementi da un raw di soli nbytes) sforano il buffer del chiamante, + * che e' dimensionato sul config, non sul file. Il chiamante che alloca su + * st_numel resta coerente; questo blocca l'ingresso ostile a monte. */ + int esz = (t->dtype == 2) ? 4 : 2; + if (t->numel < 0 || t->numel > t->nbytes / esz || t->numel * (int64_t)esz != t->nbytes) { + fprintf(stderr, "%s: tensor '%s' shape/bytes mismatch (numel %lld, %lld bytes, dtype %d) — refusing (hostile or corrupt file)\n", + name, name, (long long)t->numel, (long long)t->nbytes, t->dtype); exit(1); } void *raw = malloc(t->nbytes); if (!raw) { fprintf(stderr, "malloc %lld bytes for tensor %s failed\n", (long long)t->nbytes, name); exit(1); } st_pread_full(t->fd, raw, t->nbytes, t->off, "pread data"); From 845af6378d212f07906ba635b2ae4f7e4f2b0c64 Mon Sep 17 00:00:00 2001 From: JustVugg Date: Sun, 19 Jul 2026 14:08:19 +0200 Subject: [PATCH 15/22] docs: beginner-friendly Quick Start guide for Linux/Windows/macOS (#414) MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Adds docs/quickstart.md — a step-by-step, no-experience-assumed walkthrough from installing the build tools to the first coli chat, with per-OS copy-paste commands (Ubuntu apt, Windows MSYS2 or prebuilt binary, macOS brew), the ready-made HF int4 container plus the self-convert path, and an honest 'what to expect' on disk-bound speed. Commands verified against setup.sh and the coli subcommands; every cross-linked doc exists. Linked from the README's Get started section. Closes #414 Co-Authored-By: Claude Fable 5 --- README.md | 4 ++ docs/quickstart.md | 164 +++++++++++++++++++++++++++++++++++++++++++++ 2 files changed, 168 insertions(+) create mode 100644 docs/quickstart.md diff --git a/README.md b/README.md index e5d1651..f9cf3f0 100644 --- a/README.md +++ b/README.md @@ -151,6 +151,10 @@ scale-granularity/rotation ablations live in ## Get started +> **New here?** The [Quick Start guide](docs/quickstart.md) walks through +> install → build → model → first chat step by step for Linux, Windows, and +> macOS, with copy-paste commands and no assumed background. + ### 1. Get the model A pre-converted **GLM-5.2 int4** container is on Hugging Face — **use the diff --git a/docs/quickstart.md b/docs/quickstart.md new file mode 100644 index 0000000..7d6a9bf --- /dev/null +++ b/docs/quickstart.md @@ -0,0 +1,164 @@ +# Quick Start — from zero to a running model + +A step-by-step guide for first-time users on **Linux**, **Windows**, and **macOS**. +No prior experience with C, CUDA, or model conversion is assumed. If you get +stuck, `./coli doctor` (below) tells you exactly what's missing. + +> **What you're setting up:** colibrì runs a very large Mixture-of-Experts model +> (e.g. GLM-5.2, 744B parameters) on a normal machine by streaming the model's +> experts from disk instead of needing them all in RAM. The engine is a single +> C program; Python is only used once, to prepare the model files. + +--- + +## 0. What you need first (prerequisites) + +| | Minimum | Recommended | +|---|---|---| +| **RAM** | ~16 GB | 24 GB+ | +| **Free disk** | ~380 GB for the int4 model | a fast NVMe SSD (streaming speed = your token speed) | +| **OS** | Linux, Windows 10/11, or macOS | any | +| **Tools** | a C compiler + `make` + `git` + `python3` | — | + +You do **not** need a GPU. A GPU only helps if you have one; the engine runs +CPU-only by default. + +--- + +## 1. Install the build tools + +### Linux (Ubuntu / Debian) + +```bash +sudo apt update +sudo apt install -y build-essential git python3 +``` + +`build-essential` gives you `gcc`, `make`, and OpenMP (libgomp) — everything the +engine needs. + +### Windows + +You have two options. + +**Option A — download a prebuilt binary (no compiler needed).** +Grab the latest `colibri--windows-x86_64.zip` from the +[Releases page](https://github.com/JustVugg/colibri/releases), unzip it, and +skip to [step 3](#3-get-the-model). Python 3 (from +[python.org](https://www.python.org/downloads/)) is still needed if you want to +convert a model yourself. + +**Option B — build from source with MSYS2.** +Install [MSYS2](https://www.msys2.org/), open the **UCRT64** shell, and run: + +```bash +pacman -S --needed mingw-w64-ucrt-x86_64-gcc make git python +``` + +### macOS + +```bash +xcode-select --install # C compiler (clang) +brew install libomp git python # OpenMP for multithreading +``` + +--- + +## 2. Get the code and build the engine + +```bash +git clone https://github.com/JustVugg/colibri.git +cd colibri/c +./setup.sh +``` + +`setup.sh` checks your compiler and OpenMP, builds the engine, and runs a tiny +self-test. When it prints: + +``` +engine self-test: 32/32 (expected 32/32) +``` + +the engine is working correctly. (On Windows Option A you already have the +binary — you can skip this step.) + +--- + +## 3. Get the model + +You have two paths. + +### Easiest — download a ready-made int4 container + +A pre-converted **GLM-5.2 int4** model is on Hugging Face. **Use the version +with the int8 MTP heads** (the plain int4 heads disable speculative decoding — +see [#8](https://github.com/JustVugg/colibri/issues/8)): + +**https://huggingface.co/mateogrgic/GLM-5.2-colibri-int4-with-int8-mtp** + +Download it into a folder on a fast disk, e.g. `/nvme/glm52_i4` (Linux/macOS) or +`D:\glm52_i4` (Windows). It is about **372 GB**, so make sure you have the space. + +### Or convert it yourself from the FP8 source + +One resumable command downloads and converts the model shard by shard, so it +never needs the full ~756 GB on disk at once: + +```bash +./coli convert --model /nvme/glm52_i4 +``` + +This step uses Python and runs only once. Safe to interrupt and re-run — it +resumes where it left off. + +--- + +## 4. Run it + +Point `COLI_MODEL` at the folder from step 3 and start chatting: + +```bash +# Linux / macOS +COLI_MODEL=/nvme/glm52_i4 ./coli chat + +# Windows (UCRT64 shell) +COLI_MODEL=/d/glm52_i4 ./coli chat +``` + +Useful first commands: + +```bash +COLI_MODEL=/nvme/glm52_i4 ./coli doctor # read-only check: is everything ready? +COLI_MODEL=/nvme/glm52_i4 ./coli plan # shows where the model will live (RAM/disk/GPU) +COLI_MODEL=/nvme/glm52_i4 ./coli chat --topp 0.85 # faster: reads less from disk, same quality +``` + +> **Tip:** `--topp 0.85` is worth adding on a disk-bound machine — it reads +> fewer expert bytes per token with no quality loss, which directly means more +> tokens per second. + +--- + +## 5. What to expect + +- **First launch loads the resident weights** (~10 GB) — this takes a moment. +- **Speed depends on your disk.** The experts stream from storage, so a fast + NVMe SSD is the single biggest factor in tokens/second. On a slow or shared + disk, generation can be well under 1 token/second — that's expected, and it's + the honest cost of running a 744B model on a small machine. +- **It's still the full model.** Placement only changes speed, never the model's + answers or precision. + +If something doesn't work, run `./coli doctor` — it reports exactly what's +missing (compiler, model files, permissions) and how to fix it. + +--- + +## Where to go next + +| Topic | Doc | +|---|---| +| Windows native build (and CUDA DLL) | [docs/windows.md](windows.md) | +| Tuning: cache, prefetch, speculation | [docs/tuning.md](tuning.md) | +| OpenAI-compatible API + web dashboard | [docs/api.md](api.md) | +| Every environment variable | [docs/ENVIRONMENT.md](ENVIRONMENT.md) | From cfcc742591a03a79754f68a7f1a6f0a7ad1738a5 Mon Sep 17 00:00:00 2001 From: JustVugg Date: Sun, 19 Jul 2026 14:58:32 +0200 Subject: [PATCH 16/22] metal: advance the LFRU recency clock on the GPU-prerouted decode path (#417) On Metal, when routing is precomputed on the GPU (g_pre_idx), the moe fast path bumps eusage/ehit/eheat for the selected experts but skips the one thing the full CPU router does at the equivalent site: elast[layer][e] = ++eaccess_clock. So the session-local recency clock advances during prefill (full router) but freezes the moment GPU-prerouted decode starts, and REPIN's tier_pick_lfru() tie-breaker then runs on stale recency for the rest of the run. Mirror the exact update the non-Metal path already does. Inside #ifdef COLI_METAL, so CPU/CUDA are untouched; elast only feeds the LFRU eviction heuristic, so this cannot affect output, only which experts REPIN keeps warm. Found and reported by @monotophic with a source-level trace repro (ELAST_TRACE). Fix is inspection-verified against line ~3055; needs a Metal build to exercise end-to-end. Closes #417 Co-Authored-By: Claude Fable 5 --- c/glm.c | 5 +++++ 1 file changed, 5 insertions(+) diff --git a/c/glm.c b/c/glm.c index 26e7fc4..0f97ad2 100644 --- a/c/glm.c +++ b/c/glm.c @@ -2931,6 +2931,11 @@ static void moe(Model *m, Layer *l, int layer, float *x, int S, float *out, int m->eusage[layer][idxs[(int64_t)s*K+kk]]++; ehit_mark(m,layer,idxs[(int64_t)s*K+kk]); if(m->eheat[layer][idxs[(int64_t)s*K+kk]]eheat[layer][idxs[(int64_t)s*K+kk]]++; + /* #417: la scorciatoia GPU-prerouted deve far avanzare l'orologio di recency + * come il percorso router completo (riga ~3055), altrimenti elast/eaccess_clock + * si congelano a fine prefill e il tie-breaker LFRU di REPIN gira su punteggi + * stantii durante il decode su Metal. */ + m->elast[layer][idxs[(int64_t)s*K+kk]]=++m->eaccess_clock; } for(int d=0;d Date: Sun, 19 Jul 2026 03:56:56 +0800 Subject: [PATCH 17/22] refactor: split glm.c into colibri.c + 4 header modules MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Rename glm.c → colibri.c and extract four self-contained modules into header-only files (same pattern as st.h/tier.h/grammar.h): quant.h (672 lines) — SIMD matmul kernels, quantization sample.h (143 lines) — RNG, top-p sampling, stop-set kv_persist.h (121 lines) — .coli_kv disk persistence telemetry.h (189 lines) — dashboard protocol, stats, usage Main engine file shrinks from 6588 to 5396 lines (−18%). Build system: primary target is now colibri$(EXE); `make glm` remains as a phony alias for backward compat. CI, setup.sh, coli CLI, and all 10 test files that include the engine are updated. make check passes (C + Python, 73 tests, zero warnings). --- .github/workflows/ci.yml | 8 +- .gitignore | 2 + c/Makefile | 53 +- c/coli | 16 +- c/{glm.c => colibri.c} | 1239 +---------------------------- c/kv_persist.h | 121 +++ c/quant.h | 672 ++++++++++++++++ c/sample.h | 143 ++++ c/setup.sh | 4 +- c/telemetry.h | 189 +++++ c/tests/bench_dsa_select.c | 2 +- c/tests/bench_topp.c | 2 +- c/tests/test_dsa_select.c | 2 +- c/tests/test_i4_grouped.c | 2 +- c/tests/test_idot.c | 2 +- c/tests/test_kv_alloc.c | 2 +- c/tests/test_makefile_platform.py | 4 +- c/tests/test_sample_nan.c | 2 +- c/tests/test_stops.c | 2 +- c/tests/test_topp.c | 2 +- c/tests/test_uring.c | 2 +- 21 files changed, 1206 insertions(+), 1265 deletions(-) rename c/{glm.c => colibri.c} (82%) create mode 100644 c/kv_persist.h create mode 100644 c/quant.h create mode 100644 c/sample.h create mode 100644 c/telemetry.h diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml index 72b7149..eb7e872 100644 --- a/.github/workflows/ci.yml +++ b/.github/workflows/ci.yml @@ -12,8 +12,8 @@ jobs: runs-on: ubuntu-latest steps: - uses: actions/checkout@v4 - - name: Build glm - run: cd c && make glm + - name: Build colibri + run: cd c && make colibri - name: C test suite run: cd c && make test-c @@ -105,11 +105,11 @@ jobs: make cuda-dll CUDA_ARCH=sm_80 test -f coli_cuda.dll || { echo "cuda-dll reported success but produced no DLL" >&2; exit 1; } echo "coli_cuda.dll built (MSVC host)" - - name: make glm CUDA_DLL=1 (host links backend_loader, not cudart) + - name: make colibri CUDA_DLL=1 (host links backend_loader, not cudart) shell: msys2 {0} run: | cd c - make glm CUDA_DLL=1 + make colibri CUDA_DLL=1 test -f glm.exe || { echo "glm CUDA_DLL=1 reported success but produced no exe" >&2; exit 1; } echo "glm.exe built against the DLL loader" diff --git a/.gitignore b/.gitignore index a65fd3d..b3e1193 100644 --- a/.gitignore +++ b/.gitignore @@ -10,6 +10,8 @@ desktop/src-tauri/target/ desktop/src-tauri/gen/ # binari compilati (si rigenerano con make / coli build) +c/colibri +c/colibri.exe c/glm c/glm.exe c/olmoe diff --git a/c/Makefile b/c/Makefile index 1f8583b..e33ad37 100644 --- a/c/Makefile +++ b/c/Makefile @@ -56,7 +56,7 @@ OMPL = endif CFLAGS = -O3 $(OMPC) -Wall -Wextra -Wno-unused-parameter -Wno-misleading-indentation -Wno-unused-function # Opt-in: ARCH=native appends -mcpu=native (arm64 clang uses -mcpu, not -march), -# which unlocks the i8mm SMMLA int8/int4 dot kernels in glm.c. ARCH unset -> +# which unlocks the i8mm SMMLA int8/int4 dot kernels in colibri.c. ARCH unset -> # no -mcpu, default build byte-identical. Apple clang knows apple-m4 / native. ifneq ($(ARCH),) CFLAGS += -mcpu=$(ARCH) @@ -70,7 +70,7 @@ else ifneq ($(IS_WIN),) # ARCH default = x86-64-v3 (portable binary with AVX2). For max speed on THIS # machine use ARCH=native: on AVX-VNNI CPUs (Intel Alder Lake+, Meteor Lake+) # it also unlocks the 128-bit VPDPBUSD int8/int4 dot kernel (dot_i8i8/dot_i4i8), -# which the x86-64-v3 baseline does not define. The #ifdef guards in glm.c mean +# which the x86-64-v3 baseline does not define. The #ifdef guards in colibri.c mean # a v3 build simply compiles out the VNNI path - safe on any x86-64. CC = gcc ARCH ?= x86-64-v3 @@ -207,17 +207,18 @@ LDFLAGS += -framework Metal -framework Foundation -lc++ METAL_OBJ = backend_metal.o endif -all: glm$(EXE) +all: colibri$(EXE) -# phony 'glm' → 'glm.exe' on Windows (so 'make glm' and 'coli build' work on every platform) -glm: glm$(EXE) +# phony targets — 'glm' kept for backward compatibility +colibri: colibri$(EXE) +glm: colibri$(EXE) # Config stamp: make only tracks file timestamps, not flag changes. Without this, -# `make glm.exe CUDA_DLL=1` after a prior CPU-only build reports "up to date" and -# silently keeps the CPU-only binary (no CUDA loader) — a build that looks like it -# worked but isn't. We record the build-affecting flags in .build-config and rewrite -# it ONLY when they change (evaluated here at parse time, so the file's timestamp -# moves exactly when the config moves). glm.exe and the CUDA/loader objects depend +# `make colibri.exe CUDA_DLL=1` after a prior CPU-only build reports "up to date" +# and silently keeps the CPU-only binary (no CUDA loader) — a build that looks like +# it worked but isn't. We record the build-affecting flags in .build-config and +# rewrite it ONLY when they change (evaluated here at parse time, so the file's +# timestamp moves exactly when the config moves). The binary and CUDA/loader objects depend # on it, so they relink on a config change and stay put otherwise. (#306) BUILD_CONFIG := $(CC)|$(CFLAGS)|$(LDFLAGS)|CUDA=$(CUDA)|CUDA_DLL=$(CUDA_DLL)|ARCH=$(ARCH)|CUDA_ARCH=$(CUDA_ARCH)|METAL=$(METAL) BUILD_CONFIG_OLD := $(shell cat .build-config 2>/dev/null) @@ -226,8 +227,8 @@ $(shell printf '%s' '$(BUILD_CONFIG)' > .build-config) endif .build-config: ; -glm$(EXE): glm.c st.h uring.h json.h tok.h tok_unicode.h compat.h grammar.h $(CUDA_OBJ) $(METAL_OBJ) .build-config - $(CC) $(CFLAGS) glm.c $(CUDA_OBJ) $(METAL_OBJ) -o glm$(EXE) $(LDFLAGS) +colibri$(EXE): colibri.c st.h uring.h json.h tok.h tok_unicode.h compat.h grammar.h quant.h sample.h kv_persist.h telemetry.h $(CUDA_OBJ) $(METAL_OBJ) .build-config + $(CC) $(CFLAGS) colibri.c $(CUDA_OBJ) $(METAL_OBJ) -o colibri$(EXE) $(LDFLAGS) # Windows runtime loader object: resolves coli_cuda_* from coli_cuda.dll. backend_loader.o: backend_loader.c backend_cuda.h compat.h .build-config @@ -285,7 +286,7 @@ PORTABLE_ARCH = native endif portable: - $(MAKE) glm$(EXE) ARCH=$(PORTABLE_ARCH) + $(MAKE) colibri$(EXE) ARCH=$(PORTABLE_ARCH) iobench$(EXE): iobench.c compat.h $(CC) $(CFLAGS) iobench.c -o iobench$(EXE) $(LDFLAGS) @@ -311,27 +312,27 @@ tests/test_schema_gbnf$(EXE): tests/test_schema_gbnf.c schema_gbnf.h grammar.h j tests/test_decode_batch$(EXE): tests/test_decode_batch.c decode_batch.h $(CC) $(CFLAGS) $< -o $@ $(LDFLAGS) -tests/test_idot$(EXE): tests/test_idot.c glm.c st.h uring.h json.h tok.h tok_unicode.h compat.h grammar.h tier.h +tests/test_idot$(EXE): tests/test_idot.c colibri.c st.h uring.h json.h tok.h tok_unicode.h compat.h grammar.h tier.h quant.h sample.h kv_persist.h telemetry.h $(CC) $(CFLAGS) $< -o $@ $(LDFLAGS) -tests/test_i4_grouped$(EXE): tests/test_i4_grouped.c glm.c st.h uring.h json.h tok.h tok_unicode.h compat.h grammar.h tier.h +tests/test_i4_grouped$(EXE): tests/test_i4_grouped.c colibri.c st.h uring.h json.h tok.h tok_unicode.h compat.h grammar.h tier.h quant.h sample.h kv_persist.h telemetry.h $(CC) $(CFLAGS) $< -o $@ $(LDFLAGS) -tests/test_stops$(EXE): tests/test_stops.c glm.c st.h uring.h json.h tok.h tok_unicode.h compat.h grammar.h tier.h +tests/test_stops$(EXE): tests/test_stops.c colibri.c st.h uring.h json.h tok.h tok_unicode.h compat.h grammar.h tier.h quant.h sample.h kv_persist.h telemetry.h $(CC) $(CFLAGS) $< -o $@ $(LDFLAGS) -tests/test_topp$(EXE): tests/test_topp.c glm.c st.h uring.h json.h tok.h tok_unicode.h compat.h grammar.h tier.h +tests/test_topp$(EXE): tests/test_topp.c colibri.c st.h uring.h json.h tok.h tok_unicode.h compat.h grammar.h tier.h quant.h sample.h kv_persist.h telemetry.h $(CC) $(CFLAGS) $< -o $@ $(LDFLAGS) # bench_topp is a microbenchmark (old qsort vs new heap partial-select, #335), NOT a test # gate -- intentionally absent from TEST_BINS. Build on demand: make tests/bench_topp -tests/bench_topp$(EXE): tests/bench_topp.c glm.c st.h uring.h json.h tok.h tok_unicode.h compat.h grammar.h tier.h +tests/bench_topp$(EXE): tests/bench_topp.c colibri.c st.h uring.h json.h tok.h tok_unicode.h compat.h grammar.h tier.h quant.h sample.h kv_persist.h telemetry.h $(CC) $(CFLAGS) $< -o $@ $(LDFLAGS) -tests/test_sample_nan$(EXE): tests/test_sample_nan.c glm.c st.h uring.h json.h tok.h tok_unicode.h compat.h grammar.h tier.h +tests/test_sample_nan$(EXE): tests/test_sample_nan.c colibri.c st.h uring.h json.h tok.h tok_unicode.h compat.h grammar.h tier.h quant.h sample.h kv_persist.h telemetry.h $(CC) $(CFLAGS) $< -o $@ $(LDFLAGS) -tests/test_kv_alloc$(EXE): tests/test_kv_alloc.c glm.c st.h json.h tok.h tok_unicode.h compat.h grammar.h tier.h +tests/test_kv_alloc$(EXE): tests/test_kv_alloc.c colibri.c st.h json.h tok.h tok_unicode.h compat.h grammar.h tier.h quant.h sample.h kv_persist.h telemetry.h $(CC) $(CFLAGS) $< -o $@ $(LDFLAGS) tests/test_logit_nan$(EXE): tests/test_logit_nan.c glm.c st.h uring.h json.h tok.h tok_unicode.h compat.h grammar.h tier.h @@ -343,15 +344,15 @@ tests/test_i4_acc512$(EXE): tests/test_i4_acc512.c tests/test_compat_direct$(EXE): tests/test_compat_direct.c compat.h $(CC) $(CFLAGS) $< -o $@ $(LDFLAGS) -tests/test_dsa_select$(EXE): tests/test_dsa_select.c glm.c st.h uring.h json.h tok.h tok_unicode.h compat.h grammar.h tier.h +tests/test_dsa_select$(EXE): tests/test_dsa_select.c colibri.c st.h uring.h json.h tok.h tok_unicode.h compat.h grammar.h tier.h quant.h sample.h kv_persist.h telemetry.h $(CC) $(CFLAGS) $< -o $@ $(LDFLAGS) # bench_dsa_select is a microbenchmark (old qsort vs new quickselect partial-select, #356), # NOT a test gate -- intentionally absent from TEST_BINS. Build on demand: make tests/bench_dsa_select -tests/bench_dsa_select$(EXE): tests/bench_dsa_select.c glm.c st.h uring.h json.h tok.h tok_unicode.h compat.h grammar.h tier.h +tests/bench_dsa_select$(EXE): tests/bench_dsa_select.c colibri.c st.h uring.h json.h tok.h tok_unicode.h compat.h grammar.h tier.h quant.h sample.h kv_persist.h telemetry.h $(CC) $(CFLAGS) $< -o $@ $(LDFLAGS) -tests/test_uring$(EXE): tests/test_uring.c glm.c st.h uring.h json.h tok.h tok_unicode.h compat.h grammar.h tier.h +tests/test_uring$(EXE): tests/test_uring.c colibri.c st.h uring.h json.h tok.h tok_unicode.h compat.h grammar.h tier.h quant.h sample.h kv_persist.h telemetry.h $(CC) $(CFLAGS) $< -o $@ $(LDFLAGS) test-c: $(TEST_BINS) @@ -368,12 +369,12 @@ check: $(MAKE) portable $(MAKE) test -install: glm$(EXE) olmoe$(EXE) +install: colibri$(EXE) olmoe$(EXE) $(INSTALL) -d $(DESTDIR)$(BINDIR) $(INSTALL) -d $(DESTDIR)$(LIBEXECDIR) $(INSTALL) -d $(DESTDIR)$(LIBEXECDIR)/tools $(INSTALL) -m 755 coli $(DESTDIR)$(BINDIR)/coli - $(INSTALL) -m 755 glm$(EXE) $(DESTDIR)$(LIBEXECDIR)/glm$(EXE) + $(INSTALL) -m 755 colibri$(EXE) $(DESTDIR)$(LIBEXECDIR)/colibri$(EXE) $(INSTALL) -m 755 olmoe$(EXE) $(DESTDIR)$(LIBEXECDIR)/olmoe$(EXE) $(INSTALL) -m 644 resource_plan.py doctor.py openai_server.py $(DESTDIR)$(LIBEXECDIR)/ $(INSTALL) -m 644 tools/*.py $(DESTDIR)$(LIBEXECDIR)/tools/ @@ -387,4 +388,4 @@ clean: bench: iobench$(EXE) @if [ -n "$(ARGS)" ]; then ./iobench$(EXE) $(ARGS); else echo "built iobench$(EXE) — run: ./iobench$(EXE) "; fi -.PHONY: all glm cuda-test cuda-bench cuda-dll portable test-c test-python test check clean install uninstall bench +.PHONY: all colibri glm cuda-test cuda-bench cuda-dll portable test-c test-python test check clean install uninstall bench diff --git a/c/coli b/c/coli index 803c22b..d6728fe 100755 --- a/c/coli +++ b/c/coli @@ -54,16 +54,20 @@ from version import __version__ as _version # guess is right (e.g. a custom packaging layout). _EXE = ".exe" if sys.platform == "win32" else "" _LIBEXEC = os.path.join(os.path.dirname(HERE), "libexec", "colibri") +_here_colibri = os.path.join(HERE, "colibri" + _EXE) _here_glm = os.path.join(HERE, "glm" + _EXE) if os.environ.get("COLI_ENGINE"): GLM = os.environ["COLI_ENGINE"] TOOLS = os.path.join(os.path.dirname(GLM), "tools") +elif os.path.exists(_here_colibri): + GLM = _here_colibri + TOOLS = os.path.join(HERE, "tools") elif os.path.exists(_here_glm): GLM = _here_glm TOOLS = os.path.join(HERE, "tools") else: - GLM = os.path.join(_LIBEXEC, "glm" + _EXE) + GLM = os.path.join(_LIBEXEC, "colibri" + _EXE) TOOLS = os.path.join(_LIBEXEC, "tools") sys.path.insert(0, _LIBEXEC) # so `import resource_plan`, `doctor`, `openai_server` still resolve @@ -241,13 +245,13 @@ def env_for(a): e["COLI_CUDA"]="0"; e.pop("CUDA_EXPERT_GB",None); e.pop("CUDA_DENSE",None) else: if not cuda_binary(): - sys.exit(f"{C.yel}--gpu needs the CUDA build:{C.r} make glm CUDA=1 (this binary is CPU-only)") + sys.exit(f"{C.yel}--gpu needs the CUDA build:{C.r} make colibri CUDA=1 (this binary is CPU-only)") e["COLI_CUDA"]="1" if a.gpu!="auto": e["COLI_GPUS"]=a.gpu e.setdefault("CUDA_DENSE","1") if a.vram and a.gpu!="none": if not cuda_binary(): - sys.exit(f"{C.yel}--vram needs the CUDA build:{C.r} make glm CUDA=1 (this binary is CPU-only)") + sys.exit(f"{C.yel}--vram needs the CUDA build:{C.r} make colibri CUDA=1 (this binary is CPU-only)") e["COLI_CUDA"]="1"; e["CUDA_EXPERT_GB"]=str(a.vram) return e @@ -421,8 +425,8 @@ def cmd_build(a): banner("build") if not os.path.exists(os.path.join(HERE, "Makefile")): sys.exit(f"{C.yel}coli build{C.r} only works from a source checkout (this is an installed copy).\n" - f" Clone https://github.com/JustVugg/colibri and run ./setup.sh, or make -C c glm.") - sys.exit(subprocess.call(["make","-C",HERE,"glm"])) + f" Clone https://github.com/JustVugg/colibri and run ./setup.sh, or make -C c colibri.") + sys.exit(subprocess.call(["make","-C",HERE,"colibri"])) def cmd_info(a): banner("info") @@ -802,7 +806,7 @@ def cmd_stop(a): if "coli" in cmd and " serve" in cmd and pid!=os.getpid(): if not any(p==pid for p,_ in targets): targets.append((pid,"coli serve (cmdline)")) comm=open(f"/proc/{pd}/comm").read().strip() - if comm in ("glm","exe","olmoe"): + if comm in ("colibri","glm","exe","olmoe"): env=open(f"/proc/{pd}/environ","rb").read().replace(b"\0",b"\n").decode("utf-8","replace") if "SERVE=1" in env: targets.append((pid,f"engine `{comm}` (SERVE=1)")) except (OSError,PermissionError): continue diff --git a/c/glm.c b/c/colibri.c similarity index 82% rename from c/glm.c rename to c/colibri.c index 0f97ad2..554a7fa 100644 --- a/c/glm.c +++ b/c/colibri.c @@ -73,21 +73,6 @@ static int g_metal_gemm_min=16; /* COLI_METAL_GEMM_MIN: min rows to send a mat static const int *g_pre_idx; static const float *g_pre_w; static const int *g_pre_keff; static const float *g_pre_sh; /* output dello shared expert gia' calcolato su GPU */ #endif -#ifdef __AVX2__ -#include -static inline float hsum256(__m256 v){ /* somma orizzontale di 8 float */ - __m128 lo=_mm256_castps256_ps128(v), hi=_mm256_extractf128_ps(v,1); - lo=_mm_add_ps(lo,hi); __m128 sh=_mm_movehl_ps(lo,lo); lo=_mm_add_ps(lo,sh); - sh=_mm_shuffle_ps(lo,lo,1); lo=_mm_add_ss(lo,sh); return _mm_cvtss_f32(lo); -} -#elif defined(__ARM_NEON) -#include /* Apple Silicon / aarch64: kernel NEON */ -#elif defined(__VSX__) -#include /* POWER8+ (ppc64le): kernel VSX */ -#undef vector /* igiene: si usano __vector/__bool espliciti */ -#undef pixel -#undef bool -#endif #ifdef __APPLE__ #include /* host_statistics64: MemAvailable di macOS */ #endif @@ -218,13 +203,21 @@ typedef struct { uint64_t bytes_mtp, bytes_main; /* byte letti da disco per tipo layer */ } Model; -static void usage_save(Model *m); /* cache che impara: definita accanto a stats_dump */ -static void tiers_emit(Model *m); -static void ehit_mark(Model *m, int layer, int eid); -static void emap_emit(Model *m); -static void hits_emit(Model *m); -static void hwinfo_emit(Model *m); -static int64_t expert_bytes_probe(Model *m, int ebits); /* PROF: tier sizes in the report */ +#include "quant.h" +static int g_no_fused_pair=0; +static int g_spec_pin=1; +static int g_spec_live=0; +static inline int spec_pinned(void){ return g_spec_pin && g_spec_live; } + +static void matmul_qt_ex(float *y, const float *x, QT *w, int S, int allow_idot); +static void matmul_qt(float *y, const float *x, QT *w, int S){ matmul_qt_ex(y,x,w,S,1); } + +static void expert_gate_up(float *g,float *u,const float *x,QT *wg,QT *wu,int S){ + if(!g_no_fused_pair&&!spec_pinned()&&S==1&&wg->fmt==2&&wu->fmt==2&&wg->I==wu->I&&wg->O==wu->O) + matmul_i4_pair(g,u,x,wg->q4,wg->s,wu->q4,wu->s,wg->I,wg->O); + else { matmul_qt(g,x,wg,S); matmul_qt(u,x,wu,S); } +} + static int g_repin; static uint64_t g_last_repin; #ifdef COLI_CUDA @@ -332,713 +325,15 @@ static float *falloc(int64_t n){ if(n<0 || (uint64_t)n > SIZE_MAX/sizeof(float)){ fprintf(stderr,"falloc: n=%lld is out of range\n",(long long)n); exit(1); } float *p=malloc((size_t)n*sizeof(float)); if(!p){fprintf(stderr,"OOM\n");exit(1);} return p; } -/* ---- Accumulatore int4->float a 512 bit / 512-bit int4->float accumulator ---- - * Stessa matematica lossless di matmul_i4 (nibble->f32, FMA), ma 32 pesi/iter su - * due catene FMA indipendenti. NON bit-identico al vecchio ordine: la riduzione - * ad albero accumula MENO errore della somma sequenziale (misurato 2-4x più - * vicino all'oracolo double sulle forme reali; perplexity invariata, +4-7% sul - * decode con routing CPU-heavy — vedi docs/experiments/glm52-6x5090-2026-07-12.md). - * EN: same lossless math as matmul_i4, 32 weights/iter on two independent FMA - * chains. Not bit-identical to the old order: tree reduction accumulates LESS - * rounding than sequential summation. I4_ACC512=0 restores the old order (A/B). */ -#if defined(__AVX512F__) && defined(__AVX512BW__) -static int g_i4_acc512=1; -static inline float dot_i4f_avx512(const uint8_t *w,const float *x,int I){ - const __m128i m4=_mm_set1_epi8(0x0F); const __m512i b8=_mm512_set1_epi32(8); - __m512 acc0=_mm512_setzero_ps(),acc1=_mm512_setzero_ps(); int i=0; - for(;i+32<=I;i+=32){ __m128i by=_mm_loadu_si128((const __m128i*)(w+(i>>1))); - __m128i lo=_mm_and_si128(by,m4),hi=_mm_and_si128(_mm_srli_epi16(by,4),m4); - __m128i n0=_mm_unpacklo_epi8(lo,hi),n1=_mm_unpackhi_epi8(lo,hi); - __m512 w0=_mm512_cvtepi32_ps(_mm512_sub_epi32(_mm512_cvtepu8_epi32(n0),b8)); - __m512 w1=_mm512_cvtepi32_ps(_mm512_sub_epi32(_mm512_cvtepu8_epi32(n1),b8)); - acc0=_mm512_fmadd_ps(_mm512_loadu_ps(x+i),w0,acc0); - acc1=_mm512_fmadd_ps(_mm512_loadu_ps(x+i+16),w1,acc1); - } - return _mm512_reduce_add_ps(_mm512_add_ps(acc0,acc1)); -} -/* selftest contro il riferimento scalare (I4_ACC512_TEST=1): copre l'ordine dei - * nibble e ogni multiplo di 32. / selftest vs the scalar reference. */ -static int i4_acc512_selftest(void){ - enum { N=224 }; uint8_t w[(N+1)/2]; float x[N]; - for(int i=0;i>1]=(uint8_t)(q+8); - else w[i>>1]|=(uint8_t)((q+8)<<4); - x[i]=(float)(((i*29+7)%101)-50)/37.f; - } - for(int n=32;n<=N;n+=32){ - float ref=0; for(int i=0;i>1]>>((i&1)*4))&15)-8); - float got=dot_i4f_avx512(w,x,n),tol=2e-5f*(1.f+fabsf(ref)); - if(fabsf(got-ref)>tol){ fprintf(stderr,"AVX512 i4 selftest n=%d: %.9g != %.9g\n",n,got,ref); return 0; } - } - return 1; -} -#endif -/* y[S,O] = x[S,I] @ W^T, W[O,I] f32 */ -static void matmul(float *y, const float *x, const float *W, int S, int I, int O){ - #pragma omp parallel for schedule(static) - for (int o=0;o>1))); /* 8 byte=16 nibble */ - __m128i lo=_mm_and_si128(by,m4), hi=_mm_and_si128(_mm_srli_epi16(by,4),m4); - __m128i nib=_mm_unpacklo_epi8(lo,hi); /* nibble in ordine */ - __m256 w0=_mm256_cvtepi32_ps(_mm256_sub_epi32(_mm256_cvtepu8_epi32(nib),b8)); - __m256 w1=_mm256_cvtepi32_ps(_mm256_sub_epi32(_mm256_cvtepu8_epi32(_mm_srli_si128(nib,8)),b8)); - acc=_mm256_fmadd_ps(_mm256_loadu_ps(xs+i), w0, acc); - acc=_mm256_fmadd_ps(_mm256_loadu_ps(xs+i+8), w1, acc); } - a=hsum256(acc); -#elif defined(__ARM_NEON) - const uint8x8_t m4=vdup_n_u8(0x0F); const int8x8_t b8=vdup_n_s8(8); - float32x4_t ac0=vdupq_n_f32(0), ac1=vdupq_n_f32(0); - for(;i+16<=I;i+=16){ uint8x8_t by=vld1_u8(w+(i>>1)); /* 8 byte=16 nibble */ - uint8x8x2_t z=vzip_u8(vand_u8(by,m4), vshr_n_u8(by,4)); /* nibble in ordine */ - int16x8_t w0=vmovl_s8(vsub_s8(vreinterpret_s8_u8(z.val[0]),b8)); - int16x8_t w1=vmovl_s8(vsub_s8(vreinterpret_s8_u8(z.val[1]),b8)); - ac0=vfmaq_f32(ac0, vld1q_f32(xs+i), vcvtq_f32_s32(vmovl_s16(vget_low_s16(w0)))); - ac1=vfmaq_f32(ac1, vld1q_f32(xs+i+4), vcvtq_f32_s32(vmovl_s16(vget_high_s16(w0)))); - ac0=vfmaq_f32(ac0, vld1q_f32(xs+i+8), vcvtq_f32_s32(vmovl_s16(vget_low_s16(w1)))); - ac1=vfmaq_f32(ac1, vld1q_f32(xs+i+12), vcvtq_f32_s32(vmovl_s16(vget_high_s16(w1)))); } - a=vaddvq_f32(vaddq_f32(ac0,ac1)); -#endif -#if defined(__AVX512F__) && defined(__AVX512BW__) - } -#endif - for(;i+1>1]; int lo=(int)(byte&0xF)-8, hi=(int)(byte>>4)-8; - a += xs[i]*(float)lo + xs[i+1]*(float)hi; } - if(i>1]; int lo=(int)(byte&0xF)-8; a += xs[i]*(float)lo; } - y[(int64_t)s*O+o]=a*sc; } } -} -/* y[S,O] = x[S,I] @ W^T with W int4 packed (2/byte) + per-GROUP scales (fmt=4). - * Same nibble math as matmul_i4, but the scale changes every `gs` elements along I. - * The accumulator resets at each group boundary: dot(x[grp], w[grp]) * scale[grp]. - * gs MUST be a multiple of 16 (the AVX2 vector width). */ -static void matmul_i4_grouped(float *y, const float *x, const uint8_t *q4, const float *scale, - int S, int I, int O, int gs){ - int rb=(I+1)/2; int ng=(I+gs-1)/gs; - #pragma omp parallel for schedule(static) - for(int o=0;oI) glen=I-base; - float sc=scl[g]; - int i=base; -#ifdef __AVX2__ - const __m128i m4=_mm_set1_epi8(0x0F); const __m256i b8=_mm256_set1_epi32(8); - __m256 acc=_mm256_setzero_ps(); - for(; i+16<=base+glen; i+=16){ __m128i by=_mm_loadl_epi64((const __m128i*)(w+(i>>1))); - __m128i lo=_mm_and_si128(by,m4),hi=_mm_and_si128(_mm_srli_epi16(by,4),m4); - __m128i nib=_mm_unpacklo_epi8(lo,hi); - __m256 w0=_mm256_cvtepi32_ps(_mm256_sub_epi32(_mm256_cvtepu8_epi32(nib),b8)); - __m256 w1=_mm256_cvtepi32_ps(_mm256_sub_epi32(_mm256_cvtepu8_epi32(_mm_srli_si128(nib,8)),b8)); - acc=_mm256_fmadd_ps(_mm256_loadu_ps(xs+i), w0, acc); - acc=_mm256_fmadd_ps(_mm256_loadu_ps(xs+i+8), w1, acc); } - a+=hsum256(acc)*sc; -#endif - /* scalar tail for the group remainder */ - for(; i>1]; - a+=(xs[i]*(float)((int)(byte&0xF)-8)+xs[i+1]*(float)((int)(byte>>4)-8))*sc; } - else { uint8_t byte=w[i>>1]; a+=xs[i]*(float)((int)(byte&0xF)-8)*sc; } - } - } - y[(int64_t)s*O+o]=a; - } - } -} -/* Decode hot path for gate+up: same exact q4 dot products as matmul_i4, but one - * OpenMP dispatch covers both matrices. KTransformers uses persistent pools; - * this keeps colibri dependency-free while removing one team launch/expert. */ -static void matmul_i4_pair(float *yg, float *yu, const float *x, - const uint8_t *qg, const float *sg, - const uint8_t *qu, const float *su, int I, int O){ - int rb=(I+1)/2; - #pragma omp parallel for schedule(static) - for(int z=0;z<2*O;z++){ - int o=z>1))); - __m128i lo=_mm_and_si128(by,m4),hi=_mm_and_si128(_mm_srli_epi16(by,4),m4); - __m128i nib=_mm_unpacklo_epi8(lo,hi); - __m256 w0=_mm256_cvtepi32_ps(_mm256_sub_epi32(_mm256_cvtepu8_epi32(nib),b8)); - __m256 w1=_mm256_cvtepi32_ps(_mm256_sub_epi32(_mm256_cvtepu8_epi32(_mm_srli_si128(nib,8)),b8)); - acc=_mm256_fmadd_ps(_mm256_loadu_ps(x+i),w0,acc); - acc=_mm256_fmadd_ps(_mm256_loadu_ps(x+i+8),w1,acc); } - a=hsum256(acc); -#elif defined(__ARM_NEON) - const uint8x8_t m4=vdup_n_u8(0x0F); const int8x8_t b8=vdup_n_s8(8); - float32x4_t ac0=vdupq_n_f32(0),ac1=vdupq_n_f32(0); - for(;i+16<=I;i+=16){ uint8x8_t by=vld1_u8(w+(i>>1)); - uint8x8x2_t n=vzip_u8(vand_u8(by,m4),vshr_n_u8(by,4)); - int16x8_t w0=vmovl_s8(vsub_s8(vreinterpret_s8_u8(n.val[0]),b8)); - int16x8_t w1=vmovl_s8(vsub_s8(vreinterpret_s8_u8(n.val[1]),b8)); - ac0=vfmaq_f32(ac0,vld1q_f32(x+i),vcvtq_f32_s32(vmovl_s16(vget_low_s16(w0)))); - ac1=vfmaq_f32(ac1,vld1q_f32(x+i+4),vcvtq_f32_s32(vmovl_s16(vget_high_s16(w0)))); - ac0=vfmaq_f32(ac0,vld1q_f32(x+i+8),vcvtq_f32_s32(vmovl_s16(vget_low_s16(w1)))); - ac1=vfmaq_f32(ac1,vld1q_f32(x+i+12),vcvtq_f32_s32(vmovl_s16(vget_high_s16(w1)))); } - a=vaddvq_f32(vaddq_f32(ac0,ac1)); -#endif -#if defined(__AVX512F__) && defined(__AVX512BW__) - } -#endif - for(;i+1>1]; a+=x[i]*(float)((b&15)-8)+x[i+1]*(float)((b>>4)-8); } - if(i>1]&15)-8); - (z=2) non calcolano la STESSA funzione. Tre interruttori dipendono da S: il gate - * int4-IDOT (S>=g_i4s — asimmetrico proprio dove g_i4s>1), la fusione gate+up solo-S==1, - * e la soglia righe del GEMM Metal. Con SPEC_PIN=1 (default) ogni forward emesso mentre - * i draft del modello sono attivi resta sulla famiglia di kernel di S=1: draft e verifica - * coincidono per costruzione. Prefill e decode non speculativo sono intoccati. - * EN: MTP acceptance collapses when the draft (S=1) and verify (S>=2) forwards do not - * compute the SAME function. Three switches are S-dependent: the int4 IDOT gate - * (S>=g_i4s — asymmetric exactly on ISAs where g_i4s>1), the S==1-only gate+up fusion, - * and the Metal GEMM row threshold. SPEC_PIN=1 (default) pins every forward issued - * while model drafts are live to the platform's S=1 kernel family, so draft and verify - * agree by construction; prefill and non-speculative decode are untouched. - * SPEC_PIN=0 restores the S-dependent gates (A/B). */ -static int g_spec_pin=1; -static int g_spec_live=0; /* set by spec_decode while drafts are live */ -static inline int spec_pinned(void){ return g_spec_pin && g_spec_live; } -static void expert_gate_up(float *g,float *u,const float *x,QT *wg,QT *wu,int S){ - if(!g_no_fused_pair&&!spec_pinned()&&S==1&&wg->fmt==2&&wu->fmt==2&&wg->I==wu->I&&wg->O==wu->O) - matmul_i4_pair(g,u,x,wg->q4,wg->s,wu->q4,wu->s,wg->I,wg->O); - else { matmul_qt(g,x,wg,S); matmul_qt(u,x,wu,S); } -} -/* y[S,O] = x[S,I] @ W^T con W int2 impacchettato (4 valori/byte) + scala[O]. nibble 2-bit -> [-2,1]. */ -static void matmul_i2(float *y, const float *x, const uint8_t *q2, const float *scale, int S, int I, int O){ - int rb=(I+3)/4; - #pragma omp parallel for schedule(static) - for (int o=0;o>2))); /* 4 byte=16 valori */ - __m128i p0=_mm_and_si128(by,m2), p1=_mm_and_si128(_mm_srli_epi16(by,2),m2); - __m128i p2=_mm_and_si128(_mm_srli_epi16(by,4),m2), p3=_mm_and_si128(_mm_srli_epi16(by,6),m2); - __m128i lo=_mm_unpacklo_epi8(p0,p1), hi=_mm_unpacklo_epi8(p2,p3); - __m128i nib=_mm_unpacklo_epi16(lo,hi); /* 16 valori in ordine */ - __m256 w0=_mm256_cvtepi32_ps(_mm256_sub_epi32(_mm256_cvtepu8_epi32(nib),b2)); - __m256 w1=_mm256_cvtepi32_ps(_mm256_sub_epi32(_mm256_cvtepu8_epi32(_mm_srli_si128(nib,8)),b2)); - acc=_mm256_fmadd_ps(_mm256_loadu_ps(xs+i), w0, acc); - acc=_mm256_fmadd_ps(_mm256_loadu_ps(xs+i+8), w1, acc); } - a=hsum256(acc); -#elif defined(__ARM_NEON) - const uint8x8_t m2v=vdup_n_u8(3); const int8x8_t b2v=vdup_n_s8(2); - float32x4_t ac0=vdupq_n_f32(0), ac1=vdupq_n_f32(0); - for(;i+16<=I;i+=16){ uint32_t wd; memcpy(&wd, w+(i>>2), 4); /* 4 byte=16 valori */ - uint8x8_t by=vreinterpret_u8_u32(vdup_n_u32(wd)); - uint8x8x2_t z01=vzip_u8(vand_u8(by,m2v), vand_u8(vshr_n_u8(by,2),m2v)); - uint8x8x2_t z23=vzip_u8(vand_u8(vshr_n_u8(by,4),m2v), vshr_n_u8(by,6)); - uint16x4x2_t zz=vzip_u16(vreinterpret_u16_u8(z01.val[0]), vreinterpret_u16_u8(z23.val[0])); - int16x8_t w0=vmovl_s8(vsub_s8(vreinterpret_s8_u16(zz.val[0]),b2v)); /* 16 valori in ordine */ - int16x8_t w1=vmovl_s8(vsub_s8(vreinterpret_s8_u16(zz.val[1]),b2v)); - ac0=vfmaq_f32(ac0, vld1q_f32(xs+i), vcvtq_f32_s32(vmovl_s16(vget_low_s16(w0)))); - ac1=vfmaq_f32(ac1, vld1q_f32(xs+i+4), vcvtq_f32_s32(vmovl_s16(vget_high_s16(w0)))); - ac0=vfmaq_f32(ac0, vld1q_f32(xs+i+8), vcvtq_f32_s32(vmovl_s16(vget_low_s16(w1)))); - ac1=vfmaq_f32(ac1, vld1q_f32(xs+i+12), vcvtq_f32_s32(vmovl_s16(vget_high_s16(w1)))); } - a=vaddvq_f32(vaddq_f32(ac0,ac1)); -#endif - for(;i>2]; int sh=(i&3)*2; a += xs[i]*(float)((int)((byte>>sh)&3)-2); } - y[(int64_t)s*O+o]=a*sc; } } -} -/* ---- KERNEL INTERI (IDOT): attivazioni quantizzate a int8 per riga (absmax/127, - * stile Q8_0), prodotto scalare INTERO via maddubs/madd AVX2 — niente conversione - * f32 dei pesi nel ciclo caldo. ~2-3x sui matmul quantizzati; errore aggiunto ~0.3% - * RMS per matmul (attivazione int8), IDOT=0 torna al percorso f32 esatto. */ -#if defined(__AVX512VNNI__) && defined(__AVX512BW__) -#define IDOT_KERNEL "avx512-vnni" -#elif defined(__AVXVNNI__) && defined(__AVX2__) -#define IDOT_KERNEL "avx-vnni" -#elif defined(__AVX2__) -#define IDOT_KERNEL "avx2" -#elif defined(__ARM_NEON) && defined(__ARM_FEATURE_MATMUL_INT8) -#define IDOT_KERNEL "neon-i8mm" -#elif defined(__ARM_NEON) -#define IDOT_KERNEL "neon" -#elif defined(__VSX__) -#define IDOT_KERNEL "vsx" -#else -#define IDOT_KERNEL "scalar" -#endif -static int g_idot=1; -#if defined(__ARM_NEON) && defined(__ARM_FEATURE_DOTPROD) -static int g_i4s=1; /* SDOT presente: int4 IDOT conviene anche a S=1 (decode). Misurato - * su Apple M-series: +14%%, expert-matmul -16%%. EN: with SDOT, int4 - * IDOT pays even at S=1 (decode); measured on Apple M-series. */ -#elif defined(__VSX__) -static int g_i4s=1; /* POWER8 vec_msum: qui il fallback f32 e' SCALARE, quindi l'IDOT - * int4 conviene anche a S=1. Misurato su POWER8 S824 (vedi PR). - * EN: on VSX the f32 fallback is plain scalar C, so int4 IDOT - * pays even at S=1. Measured on a POWER8 S824 (see PR). */ -#else -static int g_i4s=2; /* senza SDOT / altrove: soglia originale (misura AVX2 dell'autore). - * EN: without SDOT / elsewhere: original threshold (author's AVX2). */ -#endif -static inline float qrow_i8(const float *x, int8_t *q, int I){ - float amax=0; for(int i=0;iamax)amax=a; } - float s=amax/127.f; if(s<1e-12f) s=1e-12f; float inv=1.f/s; - for(int i=0;i s32 directly, 64 bytes/iter, no 16-bit intermediate. - * AVX-512 has no vpsignb: |w| via abs, sign folded into x with a mask-negate - * (w==0 -> product 0 either way). |x|<=127 (qrow_i8), |w|<=128 as u8: each - * s32 lane adds <= 4*128*127, safe up to I=16384 like the AVX2 bound. */ - __m512i acc=_mm512_setzero_si512(); - for(;i+64<=I;i+=64){ - __m512i wv=_mm512_loadu_si512((const void*)(w+i)); - __m512i xv=_mm512_loadu_si512((const void*)(x+i)); - __mmask64 neg=_mm512_movepi8_mask(wv); - __m512i xs=_mm512_mask_sub_epi8(xv,neg,_mm512_setzero_si512(),xv); - acc=_mm512_dpbusd_epi32(acc,_mm512_abs_epi8(wv),xs); - } - sum=_mm512_reduce_add_epi32(acc); -#elif defined(__AVXVNNI__) && defined(__AVX2__) - /* AVX-VNNI 128-bit: vpdpbusd u8*s8 -> s32, 16 byte/iter. Stesso trucco del - * segno della variante 512-bit: |w| via abs, segno piegato in x con maschera - * (w==0 -> product 0). __AVX2__ serve per _mm_sign_epi8 / abs. */ - __m128i acc=_mm_setzero_si128(); - for(;i+16<=I;i+=16){ - __m128i wv=_mm_loadu_si128((const __m128i*)(w+i)); - __m128i xv=_mm_loadu_si128((const __m128i*)(x+i)); - __m128i xs=_mm_sign_epi8(xv,wv); /* x * sign(w); _mm_sign zona __AVX2__ */ - acc=_mm_dpbusd_epi32(acc,_mm_abs_epi8(wv),xs); - } - sum=hsum128_i32(acc); -#elif defined(__AVX2__) - __m256i acc=_mm256_setzero_si256(); const __m256i ones=_mm256_set1_epi16(1); - for(;i+32<=I;i+=32){ - __m256i wv=_mm256_loadu_si256((const __m256i*)(w+i)); - __m256i xv=_mm256_loadu_si256((const __m256i*)(x+i)); - __m256i p=_mm256_maddubs_epi16(_mm256_sign_epi8(wv,wv),_mm256_sign_epi8(xv,wv)); - acc=_mm256_add_epi32(acc,_mm256_madd_epi16(p,ones)); - } - sum=hsum256_i32(acc); -#elif defined(__ARM_NEON) - /* ARM: SDOT nativo se disponibile (Apple Silicon: sempre); altrimenti vmull/vpadal. - * Stesso bound anti-overflow del trucco AVX2: coppie <= 128*127*2 = 32512 < 32767. */ -#if defined(__ARM_FEATURE_DOTPROD) - /* 4 accumulatori indipendenti: SDOT ha latenza ~3-4 cicli, con un solo acc la - * catena seriale strozza il core a ~26 GB/s di pesi; con 4 lane indipendenti il - * dot diventa memory-bound (misurato su M4: 26 -> 63 GB/s per core, 2.4x). */ - int32x4_t a0=vdupq_n_s32(0),a1=vdupq_n_s32(0),a2=vdupq_n_s32(0),a3=vdupq_n_s32(0); - for(;i+64<=I;i+=64){ - a0=vdotq_s32(a0,vld1q_s8(w+i), vld1q_s8(x+i)); - a1=vdotq_s32(a1,vld1q_s8(w+i+16),vld1q_s8(x+i+16)); - a2=vdotq_s32(a2,vld1q_s8(w+i+32),vld1q_s8(x+i+32)); - a3=vdotq_s32(a3,vld1q_s8(w+i+48),vld1q_s8(x+i+48)); - } - int32x4_t acc=vaddq_s32(vaddq_s32(a0,a1),vaddq_s32(a2,a3)); - for(;i+16<=I;i+=16) acc=vdotq_s32(acc,vld1q_s8(w+i),vld1q_s8(x+i)); - sum=vaddvq_s32(acc); -#else - int32x4_t acc=vdupq_n_s32(0); - for(;i+16<=I;i+=16){ - int8x16_t wv=vld1q_s8(w+i), xv=vld1q_s8(x+i); - int16x8_t p=vmull_s8(vget_low_s8(wv),vget_low_s8(xv)); - p=vmlal_s8(p,vget_high_s8(wv),vget_high_s8(xv)); - acc=vpadalq_s16(acc,p); - } - sum=vaddvq_s32(acc); -#endif -#elif defined(__VSX__) - /* POWER8: vec_msum (s8 x u8 -> s32) somma i prodotti byte DIRETTAMENTE in lane - * s32, 16 byte/iter: il bound anti-saturazione a 16 bit di maddubs qui non serve. - * Stesso trucco del segno (|w| u8 per x*sign(w) s8), ma |w| via select+sub MODULO - * e non vec_abs: -128 deve diventare 128 u8, non saturare a 127. - * EN: vec_msum accumulates byte products straight into s32 lanes; |w| is built - * with a modulo subtract select instead of vec_abs so w=-128 wraps to 128 (u8) - * rather than saturating to 127. |x|<=127 from qrow_i8, so x negation is safe. */ - __vector signed int acc=vec_splats(0); - const __vector signed char vz=vec_splats((signed char)0); - for(;i+16<=I;i+=16){ - __vector signed char wv=vec_xl(0,(const signed char*)(w+i)); - __vector signed char xv=vec_xl(0,(const signed char*)(x+i)); - __vector __bool char neg=vec_cmplt(wv,vz); - __vector signed char xs=vec_sel(xv,vec_sub(vz,xv),neg); - __vector unsigned char wa=(__vector unsigned char)vec_sel(wv,vec_sub(vz,wv),neg); - acc=vec_msum(xs,wa,acc); - } - sum=vec_extract(acc,0)+vec_extract(acc,1)+vec_extract(acc,2)+vec_extract(acc,3); -#endif - for(;i int8 [-8,7] al volo, poi stesso trucco */ -static inline int32_t dot_i4i8(const uint8_t *w4, const int8_t *x, int I){ - int32_t sum=0; int i=0; -#if defined(__AVX512VNNI__) && defined(__AVX512BW__) - /* 32 bytes = 64 nibbles -> int8 in [-8,7], one vpdpbusd per 64 values. - * 256-bit unpack leaves values in per-128-lane order [0-15][32-47]/[16-31][48-63]; - * dot pairing is order-invariant, so permute x's 128-bit blocks to match - * instead of re-ordering w (one vpermq per iter, off the critical unpack path). */ - const __m256i m4v=_mm256_set1_epi8(0x0F); - const __m512i b8v=_mm512_set1_epi8(8); - const __m512i xidx=_mm512_setr_epi64(0,1,4,5,2,3,6,7); - __m512i acc=_mm512_setzero_si512(); - for(;i+64<=I;i+=64){ - __m256i by=_mm256_loadu_si256((const __m256i*)(w4+(i>>1))); - __m256i lo=_mm256_and_si256(by,m4v), hi=_mm256_and_si256(_mm256_srli_epi16(by,4),m4v); - __m256i z0=_mm256_unpacklo_epi8(lo,hi), z1=_mm256_unpackhi_epi8(lo,hi); - __m512i wv=_mm512_sub_epi8(_mm512_inserti64x4(_mm512_castsi256_si512(z0),z1,1),b8v); - __m512i xv=_mm512_permutexvar_epi64(xidx,_mm512_loadu_si512((const void*)(x+i))); - __mmask64 neg=_mm512_movepi8_mask(wv); - __m512i xs=_mm512_mask_sub_epi8(xv,neg,_mm512_setzero_si512(),xv); - acc=_mm512_dpbusd_epi32(acc,_mm512_abs_epi8(wv),xs); - } - sum=_mm512_reduce_add_epi32(acc); -#elif defined(__AVXVNNI__) && defined(__AVX2__) - /* AVX-VNNI 128-bit, int4: 16 byte = 32 nibble -> int8 [-8,7] in due half - * (n0/n1), ciascuno alimentato a un vpdpbusd da 16 byte. Stesso unpack - * 128-bit del ramo AVX2 sotto; 32 elementi/iter come li. */ - const __m128i m4=_mm_set1_epi8(0x0F); const __m128i b8=_mm_set1_epi8(8); - __m128i acc=_mm_setzero_si128(); - for(;i+32<=I;i+=32){ - __m128i by=_mm_loadu_si128((const __m128i*)(w4+(i>>1))); /* 16 byte = 32 nibble */ - __m128i lo=_mm_and_si128(by,m4), hi=_mm_and_si128(_mm_srli_epi16(by,4),m4); - __m128i n0=_mm_unpacklo_epi8(lo,hi), n1=_mm_unpackhi_epi8(lo,hi); /* nibble in ordine */ - __m128i w0=_mm_sub_epi8(n0,b8), w1=_mm_sub_epi8(n1,b8); - __m128i x0=_mm_loadu_si128((const __m128i*)(x+i)); - __m128i x1=_mm_loadu_si128((const __m128i*)(x+i+16)); - acc=_mm_dpbusd_epi32(acc,_mm_abs_epi8(w0),_mm_sign_epi8(x0,w0)); - acc=_mm_dpbusd_epi32(acc,_mm_abs_epi8(w1),_mm_sign_epi8(x1,w1)); - } - sum=hsum128_i32(acc); -#elif defined(__AVX2__) - const __m128i m4=_mm_set1_epi8(0x0F); const __m256i b8=_mm256_set1_epi8(8); - const __m256i ones=_mm256_set1_epi16(1); - __m256i acc=_mm256_setzero_si256(); - for(;i+32<=I;i+=32){ - __m128i by=_mm_loadu_si128((const __m128i*)(w4+(i>>1))); /* 16 byte = 32 nibble */ - __m128i lo=_mm_and_si128(by,m4), hi=_mm_and_si128(_mm_srli_epi16(by,4),m4); - __m128i n0=_mm_unpacklo_epi8(lo,hi), n1=_mm_unpackhi_epi8(lo,hi); /* in ordine */ - __m256i wv=_mm256_sub_epi8(_mm256_set_m128i(n1,n0),b8); - __m256i xv=_mm256_loadu_si256((const __m256i*)(x+i)); - __m256i p=_mm256_maddubs_epi16(_mm256_sign_epi8(wv,wv),_mm256_sign_epi8(xv,wv)); - acc=_mm256_add_epi32(acc,_mm256_madd_epi16(p,ones)); - } - sum=hsum256_i32(acc); -#elif defined(__ARM_NEON) - const uint8x16_t m4q=vdupq_n_u8(0x0F); const int8x16_t b8q=vdupq_n_s8(8); -#if defined(__ARM_FEATURE_DOTPROD) - /* 4 accumulatori indipendenti (vedi dot_i8i8): spezza la catena seriale su acc. - * Misurato su M4: 12.4 -> 29.9 GB/s di pesi per core (2.4x). */ - int32x4_t a0=vdupq_n_s32(0),a1=vdupq_n_s32(0),a2=vdupq_n_s32(0),a3=vdupq_n_s32(0); - for(;i+64<=I;i+=64){ - uint8x16_t byA=vld1q_u8(w4+(i>>1)), byB=vld1q_u8(w4+(i>>1)+16); - uint8x16x2_t zA=vzipq_u8(vandq_u8(byA,m4q), vshrq_n_u8(byA,4)); /* nibble in ordine */ - uint8x16x2_t zB=vzipq_u8(vandq_u8(byB,m4q), vshrq_n_u8(byB,4)); - a0=vdotq_s32(a0,vsubq_s8(vreinterpretq_s8_u8(zA.val[0]),b8q),vld1q_s8(x+i)); - a1=vdotq_s32(a1,vsubq_s8(vreinterpretq_s8_u8(zA.val[1]),b8q),vld1q_s8(x+i+16)); - a2=vdotq_s32(a2,vsubq_s8(vreinterpretq_s8_u8(zB.val[0]),b8q),vld1q_s8(x+i+32)); - a3=vdotq_s32(a3,vsubq_s8(vreinterpretq_s8_u8(zB.val[1]),b8q),vld1q_s8(x+i+48)); - } - int32x4_t acc=vaddq_s32(vaddq_s32(a0,a1),vaddq_s32(a2,a3)); - for(;i+32<=I;i+=32){ - uint8x16_t by=vld1q_u8(w4+(i>>1)); /* 16 byte = 32 nibble */ - uint8x16x2_t z=vzipq_u8(vandq_u8(by,m4q), vshrq_n_u8(by,4)); /* nibble in ordine */ - acc=vdotq_s32(acc,vsubq_s8(vreinterpretq_s8_u8(z.val[0]),b8q),vld1q_s8(x+i)); - acc=vdotq_s32(acc,vsubq_s8(vreinterpretq_s8_u8(z.val[1]),b8q),vld1q_s8(x+i+16)); - } - sum=vaddvq_s32(acc); -#else - int32x4_t acc=vdupq_n_s32(0); - for(;i+32<=I;i+=32){ - uint8x16_t by=vld1q_u8(w4+(i>>1)); /* 16 byte = 32 nibble */ - uint8x16x2_t z=vzipq_u8(vandq_u8(by,m4q), vshrq_n_u8(by,4)); /* nibble in ordine */ - int8x16_t w0=vsubq_s8(vreinterpretq_s8_u8(z.val[0]),b8q); - int8x16_t w1=vsubq_s8(vreinterpretq_s8_u8(z.val[1]),b8q); - int8x16_t x0=vld1q_s8(x+i), x1=vld1q_s8(x+i+16); - int16x8_t p=vmull_s8(vget_low_s8(w0),vget_low_s8(x0)); /* |w|<=8: nessun overflow */ - p=vmlal_s8(p,vget_high_s8(w0),vget_high_s8(x0)); - acc=vpadalq_s16(acc,p); - p=vmull_s8(vget_low_s8(w1),vget_low_s8(x1)); - p=vmlal_s8(p,vget_high_s8(w1),vget_high_s8(x1)); - acc=vpadalq_s16(acc,p); - } - sum=vaddvq_s32(acc); -#endif -#elif defined(__VSX__) - /* 16 byte = 32 nibble. vec_mergeh/vec_mergel su ppc64le (GCC) interallacciano come - * unpacklo/unpackhi x86 (verificato empiricamente su POWER8): i nibble escono in - * ordine di memoria. |w|<=8 dopo il -8, quindi stesso trucco del segno di dot_i8i8. - * EN: vec_mergeh/l on ppc64le interleave like x86 unpacklo/hi (verified on POWER8), - * so nibbles come out in memory order; then the same sign trick as dot_i8i8. */ - const __vector unsigned char m4v=vec_splats((unsigned char)0x0F); - const __vector unsigned char sh4=vec_splats((unsigned char)4); - const __vector signed char b8v=vec_splats((signed char)8); - const __vector signed char vz=vec_splats((signed char)0); - __vector signed int acc=vec_splats(0); - for(;i+32<=I;i+=32){ - __vector unsigned char by=vec_xl(0,w4+(i>>1)); /* 16 byte = 32 nibble */ - __vector unsigned char lo=vec_and(by,m4v), hi=vec_sr(by,sh4); - __vector signed char w0=vec_sub((__vector signed char)vec_mergeh(lo,hi),b8v); - __vector signed char w1=vec_sub((__vector signed char)vec_mergel(lo,hi),b8v); - __vector signed char x0=vec_xl(0,(const signed char*)(x+i)); - __vector signed char x1=vec_xl(0,(const signed char*)(x+i+16)); - __vector __bool char n0=vec_cmplt(w0,vz), n1=vec_cmplt(w1,vz); - acc=vec_msum(vec_sel(x0,vec_sub(vz,x0),n0), - (__vector unsigned char)vec_sel(w0,vec_sub(vz,w0),n0),acc); - acc=vec_msum(vec_sel(x1,vec_sub(vz,x1),n1), - (__vector unsigned char)vec_sel(w1,vec_sub(vz,w1),n1),acc); - } - sum=vec_extract(acc,0)+vec_extract(acc,1)+vec_extract(acc,2)+vec_extract(acc,3); -#endif - for(;i+1>1]; sum+=((int)(b&0xF)-8)*x[i]+((int)(b>>4)-8)*x[i+1]; } - if(i>1]; sum+=((int)(b&0xF)-8)*x[i]; } - return sum; -} -#if defined(__ARM_NEON) && defined(__ARM_FEATURE_MATMUL_INT8) -/* SMMLA (i8mm): vmmlaq_s32 vede ogni int8x16_t come matrice 2x8 row-major (byte 0-7 = - * riga 0, byte 8-15 = riga 1) e accumula C += A*B^T nel 2x2 int32: lane0=a0.b0, - * lane1=a0.b1, lane2=a1.b0, lane3=a1.b1. vcombine di due mezze-righe costruisce la - * matrice: A = due righe di peso (o,o+1), B = due righe di attivazione (s,s+1), quindi - * meta' traffico pesi e doppio lavoro per istruzione a S>=2. EN: vmmlaq_s32 treats each - * int8x16_t as a 2x8 row-major matrix and does C += A*B^T on a 2x2 int32 tile; vcombine - * of vget_low/high halves builds the 2-row register from two weight/activation rows. */ -static inline int32x4_t mm_tile16(int32x4_t acc, int8x16_t wo, int8x16_t wo1, - int8x16_t xs, int8x16_t xs1){ - acc=vmmlaq_s32(acc, vcombine_s8(vget_low_s8(wo), vget_low_s8(wo1)), - vcombine_s8(vget_low_s8(xs), vget_low_s8(xs1))); - return vmmlaq_s32(acc, vcombine_s8(vget_high_s8(wo), vget_high_s8(wo1)), - vcombine_s8(vget_high_s8(xs), vget_high_s8(xs1))); -} -static void matmul_q_idot_mm(float *y, const int8_t *xq, const float *sx, const int8_t *q, - const float *scale, int S, int I, int O){ - #pragma omp parallel for schedule(static) - for(int o=0;o<(O&~1);o+=2){ - const int8_t *wo=q+(int64_t)o*I, *wo1=q+(int64_t)(o+1)*I; - float sc0=scale[o], sc1=scale[o+1]; - for(int s=0;s<(S&~1);s+=2){ - const int8_t *xs=xq+(int64_t)s*I, *xs1=xq+(int64_t)(s+1)*I; - /* 4 accumulatori indipendenti: una sola catena vmmla e' latency-bound. - * EN: 4 independent accumulators; a single vmmla chain is latency-bound. */ - int32x4_t a0=vdupq_n_s32(0),a1=vdupq_n_s32(0),a2=vdupq_n_s32(0),a3=vdupq_n_s32(0); int i=0; - for(;i+64<=I;i+=64){ - a0=mm_tile16(a0,vld1q_s8(wo+i), vld1q_s8(wo1+i), vld1q_s8(xs+i), vld1q_s8(xs1+i)); - a1=mm_tile16(a1,vld1q_s8(wo+i+16),vld1q_s8(wo1+i+16),vld1q_s8(xs+i+16),vld1q_s8(xs1+i+16)); - a2=mm_tile16(a2,vld1q_s8(wo+i+32),vld1q_s8(wo1+i+32),vld1q_s8(xs+i+32),vld1q_s8(xs1+i+32)); - a3=mm_tile16(a3,vld1q_s8(wo+i+48),vld1q_s8(wo1+i+48),vld1q_s8(xs+i+48),vld1q_s8(xs1+i+48)); - } - for(;i+16<=I;i+=16) - a0=mm_tile16(a0,vld1q_s8(wo+i),vld1q_s8(wo1+i),vld1q_s8(xs+i),vld1q_s8(xs1+i)); - int32x4_t acc=vaddq_s32(vaddq_s32(a0,a1),vaddq_s32(a2,a3)); - int32_t d00=vgetq_lane_s32(acc,0), d01=vgetq_lane_s32(acc,1); - int32_t d10=vgetq_lane_s32(acc,2), d11=vgetq_lane_s32(acc,3); - for(;i>1)), byo1=vld1q_u8(wo1+(i>>1)); - uint8x16_t cyo=vld1q_u8(wo+(i>>1)+16), cyo1=vld1q_u8(wo1+(i>>1)+16); - uint8x16x2_t zo =vzipq_u8(vandq_u8(byo, m4q), vshrq_n_u8(byo, 4)); - uint8x16x2_t zo1=vzipq_u8(vandq_u8(byo1,m4q), vshrq_n_u8(byo1,4)); - uint8x16x2_t ko =vzipq_u8(vandq_u8(cyo, m4q), vshrq_n_u8(cyo, 4)); - uint8x16x2_t ko1=vzipq_u8(vandq_u8(cyo1,m4q), vshrq_n_u8(cyo1,4)); - a0=mm_tile16(a0, vsubq_s8(vreinterpretq_s8_u8(zo.val[0]),b8q), - vsubq_s8(vreinterpretq_s8_u8(zo1.val[0]),b8q), - vld1q_s8(xs+i), vld1q_s8(xs1+i)); - a1=mm_tile16(a1, vsubq_s8(vreinterpretq_s8_u8(zo.val[1]),b8q), - vsubq_s8(vreinterpretq_s8_u8(zo1.val[1]),b8q), - vld1q_s8(xs+i+16), vld1q_s8(xs1+i+16)); - a2=mm_tile16(a2, vsubq_s8(vreinterpretq_s8_u8(ko.val[0]),b8q), - vsubq_s8(vreinterpretq_s8_u8(ko1.val[0]),b8q), - vld1q_s8(xs+i+32), vld1q_s8(xs1+i+32)); - a3=mm_tile16(a3, vsubq_s8(vreinterpretq_s8_u8(ko.val[1]),b8q), - vsubq_s8(vreinterpretq_s8_u8(ko1.val[1]),b8q), - vld1q_s8(xs+i+48), vld1q_s8(xs1+i+48)); - } - for(;i+32<=I;i+=32){ - uint8x16_t byo=vld1q_u8(wo+(i>>1)), byo1=vld1q_u8(wo1+(i>>1)); - uint8x16x2_t zo =vzipq_u8(vandq_u8(byo, m4q), vshrq_n_u8(byo, 4)); - uint8x16x2_t zo1=vzipq_u8(vandq_u8(byo1,m4q), vshrq_n_u8(byo1,4)); - a0=mm_tile16(a0, vsubq_s8(vreinterpretq_s8_u8(zo.val[0]),b8q), - vsubq_s8(vreinterpretq_s8_u8(zo1.val[0]),b8q), - vld1q_s8(xs+i), vld1q_s8(xs1+i)); - a1=mm_tile16(a1, vsubq_s8(vreinterpretq_s8_u8(zo.val[1]),b8q), - vsubq_s8(vreinterpretq_s8_u8(zo1.val[1]),b8q), - vld1q_s8(xs+i+16), vld1q_s8(xs1+i+16)); - } - int32x4_t acc=vaddq_s32(vaddq_s32(a0,a1),vaddq_s32(a2,a3)); - int32_t d00=vgetq_lane_s32(acc,0), d01=vgetq_lane_s32(acc,1); - int32_t d10=vgetq_lane_s32(acc,2), d11=vgetq_lane_s32(acc,3); - for(;i+1>1], bo1=wo1[i>>1]; - int a0=(int)(bo&0xF)-8, a1=(int)(bo>>4)-8, b0=(int)(bo1&0xF)-8, b1=(int)(bo1>>4)-8; - int u0=xs[i],u1=xs[i+1],v0=xs1[i],v1=xs1[i+1]; - d00+=a0*u0+a1*u1; d01+=a0*v0+a1*v1; d10+=b0*u0+b1*u1; d11+=b0*v0+b1*v1; } - if(i>1], bo1=wo1[i>>1]; - int a0=(int)(bo&0xF)-8, b0=(int)(bo1&0xF)-8; - d00+=a0*xs[i]; d01+=a0*xs1[i]; d10+=b0*xs[i]; d11+=b0*xs1[i]; } - y[(int64_t)s*O+o] =(float)d00*sc0*sx[s]; - y[(int64_t)s*O+(o+1)] =(float)d10*sc1*sx[s]; - y[(int64_t)(s+1)*O+o] =(float)d01*sc0*sx[s+1]; - y[(int64_t)(s+1)*O+(o+1)]=(float)d11*sc1*sx[s+1]; - } - if(S&1){ int s=S-1; const int8_t *xs=xq+(int64_t)s*I; - y[(int64_t)s*O+o] =(float)dot_i4i8(wo, xs,I)*sc0*sx[s]; - y[(int64_t)s*O+(o+1)]=(float)dot_i4i8(wo1,xs,I)*sc1*sx[s]; } - } - if(O&1){ int o=O-1; const uint8_t *w=q4+(int64_t)o*rb; float sc=scale[o]; - #pragma omp parallel for schedule(static) - for(int s=0;s=2){ matmul_q_idot_mm(y,xq,sx,q,scale,S,I,O); return; } -#endif - #pragma omp parallel for schedule(static) - for(int o=0;o=2){ matmul_i4_idot_mm(y,xq,sx,q4,scale,S,I,O); return; } -#endif - #pragma omp parallel for schedule(static) - for(int o=0;og_qscratch.xq_cap){ - int8_t *p=realloc(g_qscratch.xq,xn); - if(!p){ fprintf(stderr,"OOM quant scratch\n"); exit(1); } - g_qscratch.xq=p; g_qscratch.xq_cap=xn; - } - if(sn>g_qscratch.sx_cap){ - float *p=realloc(g_qscratch.sx,sn*sizeof(float)); - if(!p){ fprintf(stderr,"OOM quant scales\n"); exit(1); } - g_qscratch.sx=p; g_qscratch.sx_cap=sn; - } - *xq=g_qscratch.xq; *sx=g_qscratch.sx; -} - -/* allow_idot=0: forza il kernel int4/int8 ESATTO (attivazioni f32). Serve alle proiezioni di - * attenzione: sono sensibili alla quantizzazione int8 delle attivazioni dell'IDOT. Misurato su - * GLM-5.2 int4, 1023 token, log-lik -5040.33 (esatto) -> -5160.47 (IDOT) = +0.117 nat/token, - * ~+12% perplexity. Gli altri matmul del prefill (o_proj, kv_b, expert) tengono l'IDOT. - * EN: allow_idot=0 forces the EXACT int4/int8 kernel (f32 activations). The attention - * projections need it: IDOT's int8 activation quantization costs +0.117 nats/token there - * (~+12% perplexity), measured. Every other prefill matmul keeps IDOT as before. */ -static void matmul_qt_ex(float *y, const float *x, QT *w, int S, int allow_idot); -static void matmul_qt(float *y, const float *x, QT *w, int S){ matmul_qt_ex(y,x,w,S,1); } static void matmul_qt_ex(float *y, const float *x, QT *w, int S, int allow_idot){ #ifdef COLI_METAL - /* Large row-batches (prefill: kv_b reconstruction, o_proj, dense MLP, step_all logits) - * amortize Metal's ~5ms submit latency; small-S decode matmuls stay on CPU (NEON wins). - * Weights must be registered (all dense QT allocs are, via qalloc). */ if(g_metal_enabled && S>=g_metal_gemm_min && !spec_pinned() && (w->fmt==1||w->fmt==2) && !omp_in_parallel()){ const void *wp = w->fmt==1 ? (const void*)w->q8 : (const void*)w->q4; if(coli_metal_gemm(y,x,wp,w->s,w->fmt,S,w->I,w->O)) return; } #endif #ifdef COLI_CUDA - /* The CUDA backend owns persistent copies only for model-resident tensors. - * Streaming expert slots are reused for different IDs and must never enter - * this cache. Nested OpenMP calls stay on CPU because each device context - * intentionally owns one synchronous scratch stream in this stage. */ if(g_cuda_enabled && w->cuda_eligible && !w->cuda_failed && !omp_in_parallel()){ const void *weights = w->fmt==0 ? (const void*)w->qf : w->fmt==1 ? (const void*)w->q8 : (const void*)w->q4; @@ -1049,21 +344,7 @@ static void matmul_qt_ex(float *y, const float *x, QT *w, int S, int allow_idot) } #endif if(w->fmt==0){ matmul(y,x,w->qf,S,w->I,w->O); return; } - /* fmt=4: grouped int4 — always use the exact grouped kernel (no IDOT approximation, - * since the whole point of grouped scales is better quality). */ if(w->fmt==4){ matmul_i4_grouped(y,x,w->q4,w->s,S,w->I,w->O,w->gs); return; } - /* int8 IDOT vince sempre (1.4-2.5x). int4 IDOT: l'autore su AVX2 trovo' che a S=1 - * non ripaga (soglia S>=2); ma su ARM/SDOT il singolo token CONVIENE (vedi g_i4s / - * PR #9 per il gemello VNNI). Soglia configurabile con I4S. - * EN: int8 IDOT always wins (1.4-2.5x). int4 IDOT: on AVX2 the author found S=1 didn't - * pay (S>=2 gate); on ARM/SDOT single-token DOES pay (see g_i4s / PR #9 for the VNNI - * twin). Threshold configurable via I4S. */ - /* #163: sotto SPEC_PIN il gate int4-IDOT usa la decisione di S=1 per OGNI S, cosi' - * draft e verifica restano sulla stessa famiglia. (CUDA non e' toccato: la sua - * condizione non dipende da S, quindi e' gia' coerente tra draft e verifica.) - * EN: under SPEC_PIN the int4 IDOT gate uses the S=1 decision for EVERY S, so draft - * and verify stay in one family. (CUDA untouched: its condition is S-independent, - * hence already draft/verify-consistent.) */ if(allow_idot && g_idot && (w->fmt==1 || (w->fmt==2 && (spec_pinned() ? g_i4s<=1 : S>=g_i4s)))){ int I=w->I; int8_t *xq; float *sx; if(S<0 || I<0 || (size_t)S>SIZE_MAX/(size_t)(I?I:1)){ fprintf(stderr,"matmul_qt: shape overflow\n"); exit(1); } @@ -1078,49 +359,6 @@ static void matmul_qt_ex(float *y, const float *x, QT *w, int S, int allow_idot) else matmul_i4(y,x,w->q4,w->s,S,w->I,w->O); } -/* quantizza w[O,I] f32 -> int8 q[O,I] + scala[O] simmetrica per riga */ -static void quantize_rows(const float *w, int8_t *q, float *scale, int O, int I, int bits){ - int qmax=(1<<(bits-1))-1; - #pragma omp parallel for schedule(static) - for(int o=0;oamax)amax=a; } - float s=amax/qmax; if(s<1e-8f)s=1e-8f; scale[o]=s; - int8_t *qr=q+(int64_t)o*I; - for(int i=0;iqmax)v=qmax; if(v<-qmax-1)v=-qmax-1; qr[i]=(int8_t)v; } - } -} -/* quantizza w[O,I] f32 -> int4 impacchettato (2/byte) + scala[O]. - * bits<=4: valori in [-qmax-1,qmax] stanno in un nibble [-8,7]; memorizzati come v+8 (0..15). */ -static void pack_int4(const float *w, uint8_t *q4, float *scale, int O, int I, int bits){ - int qmax=(1<<(bits-1))-1, rb=(I+1)/2; - #pragma omp parallel for schedule(static) - for(int o=0;oamax)amax=a; } - float s=amax/qmax; if(s<1e-8f)s=1e-8f; scale[o]=s; - uint8_t *qr=q4+(int64_t)o*rb; - for(int i=0;iqmax)v0=qmax; if(v0<-8)v0=-8; - int v1=0; if(i+1qmax)v1=qmax; if(v1<-8)v1=-8; } - qr[i>>1] = (uint8_t)((v0+8) | ((v1+8)<<4)); - } - } -} - -/* quantizza w[O,I] f32 -> int2 impacchettato (4/byte) + scala[O]. valori nibble 2-bit in [-2,1]. */ -static void pack_int2(const float *w, uint8_t *q2, float *scale, int O, int I, int bits){ - int qmax=(1<<(bits-1))-1, rb=(I+3)/4; - #pragma omp parallel for schedule(static) - for(int o=0;oamax)amax=a; } - float s=amax/qmax; if(s<1e-8f)s=1e-8f; scale[o]=s; - uint8_t *qr=q2+(int64_t)o*rb; - for(int i=0;iqmax)v=qmax; if(v<-2)v=-2; byte|=(uint8_t)((v+2)<<(k*2)); } - qr[i>>2]=byte; - } - } -} - static int g_nopack=0; /* NOPACK=1 -> tiene i valori <=4bit in contenitore int8 (per validare il packing) */ static int g_drop=0; /* DROP=1 -> scarta le pagine expart dopo l'uso. Default 0: le lascia in * page-cache (buff/cache, NON RSS) come L2 gratuito -> sfrutta lo @@ -1206,6 +444,11 @@ static int g_disk_split=0; /* DISK_SPLIT=1: contatori che spezzano i DISK LOAD ( * (int4), con i byte letti. Default OFF: a flag spento gli atomic * non vengono MAI toccati (zero overhead), le righe extra di stats * non vengono stampate. Solo misura: nessun effetto sull'output. */ + +#include "sample.h" +#include "kv_persist.h" +#include "telemetry.h" + /* Aligned allocator for dense QT weights/scales: under METAL, page-align + register so the * GPU reads them zero-copy (no upload duplicate). Plain malloc otherwise. */ /* ---- COLI_NUMA=1 (#82): interleave the expert slabs across NUMA nodes ---- @@ -4239,13 +3482,14 @@ static void mtp_absorb(Model *m, const int *next_ids, const float *x, int S, int } static inline int argmax_v(const float *lo, int V){ - /* skip NaN (x==x is false for NaN) so a poisoned logit can't pin the argmax - * to index 0 — pick the max finite/+Inf entry instead. */ int b=-1; float bv=-INFINITY; for(int i=0;ibv){ bv=x; b=i; } } return b<0?0:b; } +static void repin_pass_limit(Model *m,int limit); +static void repin_pass(Model *m){ repin_pass_limit(m,16); } + /* ---- METODO F: draft grammaticale (#48) ---- * gr_feed consuma i byte di ogni token EMESSO e tiene il walker in sync con l'output; * grammar_draft propone lo span FORZATO successivo (un solo byte legale per posizione) @@ -4326,113 +3570,6 @@ static int grammar_draft(int *draft, int cap){ * Rejection sampling di Leviathan: accetta il draft x_d con prob p(x_d); al rifiuto * ricampiona da p con x_d azzerato e rinormalizzato. La distribuzione risultante e' * ESATTAMENTE p: la speculazione resta invisibile all'output anche col sampling. */ -static uint64_t g_rng=0x9E3779B97F4A7C15ULL; -static inline double rndu(void){ g_rng^=g_rng<<13; g_rng^=g_rng>>7; g_rng^=g_rng<<17; - return (double)(g_rng>>11)*(1.0/9007199254740992.0); } -static float *g_pbuf=NULL; static int *g_pidx=NULL; /* buffer riusati (decode single-thread) */ -/* sift-down su max-heap in h[0..n), chiave = g_pbuf[h[i]] (#335: partial top-p select). - * Versione "a buco": porta il valore di radice e lo deposita solo alla fine, cosi' - * heapify e' O(V) e ogni pop e' O(log n) senza qsort sull'intero vocabolario. */ -static void topp_siftdown(int *h, int n, int i){ - int iv=h[i]; float kv=g_pbuf[iv]; - for(;;){ int l=2*i+1; - if(l>=n) break; /* foglia */ - int b=l; if(l+1g_pbuf[h[l]]) b=l+1; /* figlio maggiore */ - if(g_pbuf[h[b]]<=kv) break; /* nessun figlio supera la radice -> ferma */ - h[i]=h[b]; i=b; } - h[i]=iv; -} -/* costruisce in g_pbuf la distribuzione target: softmax(lo/temp) troncata a top-p g_nuc. - * Invariante per dist_sample: g_pbuf resta INDICIZZATO per token-id (mai riordinato); - * la coda troncata va AZZERATA in g_pbuf (dist_sample la legge direttamente per id). */ -static void dist_build(const float *lo, int V){ - if(!g_pbuf){ g_pbuf=falloc(V); g_pidx=malloc(V*sizeof(int)); } - /* Un solo logit NaN/+Inf avvelenava tutto (#369): +Inf diventava mx, NaN/Inf-mx - * -> expf NaN -> s NaN -> ogni prob NaN -> dist_sample cade sul fallback - * `g_pbuf[i]>0` (NaN>0 e' falso ovunque) e ritorna 0 PER SEMPRE, in silenzio. - * Difesa: mx solo sui finiti; un logit non finito contribuisce prob 0; - * se la distribuzione degenera (tutti non finiti / somma non valida) si - * ripiega sull'argmax dei finiti e si avvisa UNA volta, mai in silenzio. */ - int mxi=-1; float mx=0; - for(int i=0;imx)){ mx=lo[i]; mxi=i; } - double s=0; float invt=1.f/(g_temp>1e-4f?g_temp:1e-4f); - if(mxi>=0){ - for(int i=0;i=0)?mxi:0; /* mxi = argmax dei logit FINITI (robusto - * anche se lo[0] e' NaN, dove argmax_v fallirebbe) */ - for(int i=0;i0 && g_nuc<1.f){ - for(int i=0;i=0;i--) topp_siftdown(g_pidx,V,i); /* heapify O(V) */ - /* pop verso la coda: i vincitori (testa top-p) cadono in g_pidx[out..V-1] in ordine - * DECRESCENTE, come il vecchio qsort, quindi s2 accumula nello stesso ordine -> - * head bit-identical sui casi senza pareggi (i pareggi erano gia' non specificati - * sotto il qsort instabile e restano tali). Il prefisso g_pidx[0..out-1) e' la coda. */ - double s2=0, cum=0; int out=V; - do{ int root=g_pidx[0]; /* massimo corrente */ - g_pidx[0]=g_pidx[--out]; g_pidx[out]=root; /* sposta il max in coda */ - s2+=g_pbuf[root]; cum+=g_pbuf[root]; - if(out>0) topp_siftdown(g_pidx,out,0); - } while(cum0); - for(int i=0;i=0 -> quel token e' escluso (rinormalizzando al volo) */ -static int dist_sample(int V, int ban){ - double z = 1.0 - (ban>=0 ? g_pbuf[ban] : 0.0); if(z<=1e-12) z=1e-12; - double u = rndu()*z, cum=0; - for(int i=0;i=u) return i; } - for(int i=V-1;i>=0;i--) if(i!=ban && g_pbuf[i]>0) return i; - return 0; -} -/* prossimo token dai logits: greedy se g_temp<=0, altrimenti sampling. - * ban = token escluso perche' rifiutato dalla verifica speculativa precedente. */ -static int pick_tok(const float *lo, int V, int ban){ - if(g_temp<=0) return argmax_v(lo,V); - dist_build(lo,V); - return dist_sample(V,ban); -} - -/* stop-set attivo (popolato da run_text/run_serve dal config; vuoto in validazione, - * dove si genera un numero fisso di token da confrontare con l'oracolo) */ -static int g_stop[64], g_nstop=0; /* config eos + ogni added-token "special" del tokenizer */ -static void repin_pass_limit(Model *m,int limit); -static void repin_pass(Model *m){ repin_pass_limit(m,16); } -static inline int is_stop(int t){ for(int i=0;i solo gli stop del config (validazione/oracolo, dove il tokenizer non serve). */ -static void stops_arm_tok(const Cfg *c, int tok_eos, Tok *T){ - g_nstop=0; - for(int i=0;in_stop && g_nstop<64;i++) g_stop[g_nstop++]=c->stop_ids[i]; - if(tok_eos>=0 && !is_stop(tok_eos) && g_nstop<64) g_stop[g_nstop++]=tok_eos; - int nsp=0; - /* DIFESA IN PROFONDITA' (woolcoxm, #298): il tokenizer marca "special":true i token di - * CONTROLLO -- <|user|>, <|assistant|>, <|observation|>, , [gMASK], i marker - * image/video/audio. Nessuno di questi e' contenuto legittimo di una risposta: se il - * modello ne emette uno, il turno e' finito (infatti GLM ne elenca tre fra gli eos - * ufficiali). Senza questo, uno di quei token non elencato nel config veniva - * DETOKENIZZATO E STAMPATO IN CHAT come testo, e la generazione proseguiva oltre la - * fine reale -- l'"added stuff on the end" riportato su un checkpoint convertito. - * Fidarsi del config di pesi convertiti da terzi e' precisamente cio' che non - * possiamo controllare; il flag del tokenizer lo possiamo leggere. - * NB: // hanno "special":false e restano contenuto vero. */ - if(T) for(int id=0; idn_ids && g_nstop<64; id++) - if(T->id_special[id] && !is_stop(id)){ g_stop[g_nstop++]=id; nsp++; } - fprintf(stderr,"[stop] %d stop tokens:",g_nstop); - for(int i=0;imx){mx=lo[i];best=i;} } - double se=0; for(int i=0;i .. " (T=ctxlen+contlen) * output: riga " " per richiesta. @@ -5101,117 +4226,6 @@ static void repin_pass_limit(Model *m,int limit){ * si appendono SOLO le posizioni nuove e si riscrive nrec per ultimo: un crash a meta' * append lascia nrec vecchio = file coerente. La riga KV del layer MTP non si salva: * al resume kv_start=-1 e la finestra di draft riparte da sola. */ -static int g_kvsave=1; -#define KV_MAGIC "COLIKV1\0" -static void kv_hdr(Model *m, int32_t *h, int nrec){ - Cfg *c=&m->c; int nic=0; - for(int i=0;in_layers;i++) if(m->Ic && m->Ic[i]) nic++; - h[0]=c->n_layers; h[1]=c->kv_lora; h[2]=c->qk_rope; - h[3]=m->has_dsa?c->index_hd:0; h[4]=nic; h[5]=c->vocab; h[6]=nrec; h[7]=0; -} -/* Bytes of one on-disk record: [tok i32][Lc+Rc per layer][Ic per DSA layer]. - * Layout matches what kv_disk_append writes and kv_disk_load reads. */ -static int64_t kv_rec_bytes(Model *m){ - Cfg *c=&m->c; - int64_t rec = 4 + (int64_t)c->n_layers*(c->kv_lora+c->qk_rope)*4; - if(m->has_dsa) for(int i=0;in_layers;i++) if(m->Ic[i]) rec+=(int64_t)c->index_hd*4; - return rec; -} -/* Open the persistent handle lazily; write the header if the file is new. After - * this returns successfully, k->disk_fp is valid for the engine's lifetime and - * positioned at end-of-header (nrec==0 case) or wherever the caller seeks. */ -static int kv_disk_open(Model *m){ - KVState *k=m->kv; - if(k->disk_fp) return 1; - k->disk_fp=fopen(k->disk_path,"r+b"); - if(!k->disk_fp){ /* not there yet -> create + header */ - k->disk_fp=fopen(k->disk_path,"wb"); - if(!k->disk_fp) return 0; - int32_t h[8]; kv_hdr(m,h,0); - fwrite(KV_MAGIC,1,8,k->disk_fp); fwrite(h,4,8,k->disk_fp); - fflush(k->disk_fp); - fclose(k->disk_fp); - k->disk_fp=fopen(k->disk_path,"r+b"); /* reopen r+b for append */ - if(!k->disk_fp) return 0; - } - return 1; -} -static void kv_disk_truncate(Model *m, int nrec){ - if(!g_kvsave) return; - KVState *k=m->kv; - if(k->disk_fp){ fclose(k->disk_fp); k->disk_fp=NULL; } /* drop to shrink on disc */ - FILE *f=fopen(k->disk_path,"r+b"); - if(!f){ k->disk_nrec=0; return; } - k->disk_nrec=nrec; - int32_t nr=nrec; fseek(f,8+6*4,SEEK_SET); fwrite(&nr,4,1,f); - fflush(f); fclose(f); -} -static void kv_disk_reset(Model *m){ kv_disk_truncate(m,0); } -static void kv_disk_append(Model *m, const int *hist, int len){ - KVState *k=m->kv; - if(!g_kvsave || len<=k->disk_nrec) return; - Cfg *c=&m->c; - if(!kv_disk_open(m)) return; - FILE *f=k->disk_fp; - int64_t rec = kv_rec_bytes(m); - /* grow the contiguous staging buffer if the record is larger (#1 batching) */ - if(rec > k->disk_buf_cap){ - uint8_t *nb=realloc(k->disk_buf, rec); - if(!nb) return; /* OOM: skip this turn, retry next */ - k->disk_buf=nb; k->disk_buf_cap=rec; - } - fseek(f, 8+8*4 + (int64_t)k->disk_nrec*rec, SEEK_SET); - for(int p=k->disk_nrec;pdisk_buf; /* pack token + every layer into one record */ - *(int32_t*)b = hist[p]; b+=4; - for(int i=0;in_layers;i++){ - memcpy(b, m->Lc[i]+(int64_t)p*c->kv_lora, (size_t)c->kv_lora*4); b+=c->kv_lora*4; - memcpy(b, m->Rc[i]+(int64_t)p*c->qk_rope,(size_t)c->qk_rope*4); b+=c->qk_rope*4; - } - if(m->has_dsa) for(int i=0;in_layers;i++) if(m->Ic[i]){ - memcpy(b, m->Ic[i]+(int64_t)p*c->index_hd, (size_t)c->index_hd*4); b+=c->index_hd*4; - } - fwrite(k->disk_buf, 1, (size_t)rec, f); /* one fwrite per position (was ~157) */ - } - fflush(f); /* dati prima, contatore poi */ - int32_t nr=len; fseek(f,8+6*4,SEEK_SET); fwrite(&nr,4,1,f); - fflush(f); /* persist the counter too */ - k->disk_nrec=len; -} -static int kv_disk_load(Model *m, int *hist, int maxctx){ - if(!g_kvsave) return 0; - KVState *k=m->kv; - Cfg *c=&m->c; - FILE *f=fopen(k->disk_path,"rb"); if(!f) return 0; - char mg[8]; int32_t h[8], w[8]; kv_hdr(m,w,0); - if(fread(mg,1,8,f)!=8 || memcmp(mg,KV_MAGIC,8) || fread(h,4,8,f)!=8 || - h[0]!=w[0]||h[1]!=w[1]||h[2]!=w[2]||h[3]!=w[3]||h[4]!=w[4]||h[5]!=w[5]){ - fprintf(stderr,"[KV] ignoring .coli_kv from a different model or version\n"); fclose(f); return 0; } - int nrec=h[6]; - if(nrec<1){ fclose(f); return 0; } - if(nrec>=maxctx-8-g_draft){ - fprintf(stderr,"[KV] saved conversation (%d tokens) exceeds the context: starting over\n",nrec); - fclose(f); return 0; } - double t0=now_s(); - for(int p=0;pn_layers;i++){ - if(fread(m->Lc[i]+(int64_t)p*c->kv_lora, 4, c->kv_lora, f)!=(size_t)c->kv_lora || - fread(m->Rc[i]+(int64_t)p*c->qk_rope, 4, c->qk_rope, f)!=(size_t)c->qk_rope){ nrec=p; goto out; } - } - if(m->has_dsa) for(int i=0;in_layers;i++) if(m->Ic[i]) - if(fread(m->Ic[i]+(int64_t)p*c->index_hd, 4, c->index_hd, f)!=(size_t)c->index_hd){ nrec=p; goto out; } - } -out: - fclose(f); - if(nrec>0){ - if(m->has_mtp) m->kv_start[c->n_layers]=-1; /* la finestra MTP riparte da sola */ - fprintf(stderr,"[KV] resumed conversation from disk: %d tokens in %.1fs (no re-prefill)\n", - nrec, now_s()-t0); - } - k->disk_nrec=nrec; - return nrec; -} typedef struct { KVState kv; int *hist, len, first; } ServeCtx; static double kv_pool_bytes(Model *m, int max_ctx); @@ -5622,212 +4636,7 @@ static int *read_arr(jval*o,const char*k,int*n){ if(!r){ fprintf(stderr,"OOM read_arr\n"); exit(1); } for(int i=0;ilen;i++) r[i]=(int)a->kids[i]->num; *n=a->len; return r; } -/* byte residenti di un tensore [O,I] al numero di bit dato (specchio di qt_bytes) */ -static int64_t tbytes(int O,int I,int bits){ - if(bits>=16) return (int64_t)O*I*4; - if(bits>=5) return (int64_t)O*I + (int64_t)O*4; - return (int64_t)O*((I+1)/2) + (int64_t)O*4; -} -/* byte VERI di un expert: dal container se pre-quantizzato, altrimenti stima da ebits */ -static int64_t expert_bytes_probe(Model *m, int ebits){ - Cfg *c=&m->c; int64_t eb=0; char nm[256]; - snprintf(nm,sizeof(nm),"model.layers.%d.mlp.experts.0.gate_proj.weight",c->first_dense); - if(st_nbytes(&m->S,nm)>0){ - const char *suf[3]={"gate_proj","up_proj","down_proj"}; - for(int k=0;k<3;k++){ - snprintf(nm,sizeof(nm),"model.layers.%d.mlp.experts.0.%s.weight",c->first_dense,suf[k]); - eb+=st_nbytes(&m->S,nm); - snprintf(nm,sizeof(nm),"model.layers.%d.mlp.experts.0.%s.weight.qs",c->first_dense,suf[k]); - int64_t q=st_nbytes(&m->S,nm); if(q>0) eb+=q; - } - } - if(eb<=0) eb = tbytes(c->moe_inter,c->hidden,ebits)*2 + tbytes(c->hidden,c->moe_inter,ebits); - return eb; -} - -/* TIERS: fotografia della piramide expert per la dashboard web — - * "TIERS " sul canale di protocollo. - * ram = pinnati non-VRAM + LRU corrente; disk = tutto il resto. */ -/* BRAIN MAP (dashboard): per-turn expert hit bitmap + full residency/heat map. - * g_ehit[layer][eid]=1 quando l'expert viene instradato in questo turno; - * hits_emit lo serializza e lo azzera. emap_emit fotografa tier+heat di TUTTI. */ -static uint8_t **g_ehit; -static void ehit_mark(Model *m, int layer, int eid){ - if(!g_ehit){ Cfg *c=&m->c; - g_ehit=calloc(c->n_layers+1,sizeof(uint8_t*)); - for(int i=0;i<=c->n_layers;i++) g_ehit[i]=calloc(c->n_experts,1); - } - g_ehit[layer][eid]=1; -} - -/* HWINFO: hardware snapshot for the web dashboard — emitted once at READY. */ -/* CPU model + cores + RAM (GB); empty/zero where unavailable. - * Shared by the dashboard HWINFO line and the PROF=1 header. */ -static void hw_probe(char *cpu, size_t cn, int *cores, double *ram_total, double *ram_avail){ - cpu[0]=0; -#ifdef _WIN32 - /* niente /proc su Windows: brand string via CPUID (0x80000002..4), zero - * dipendenze extra. La dashboard mostrava "0 GB RAM / 0 cores" perche' - * tutto questo blocco era solo-Linux mentre il ramo CUDA funzionava. */ -#if defined(__x86_64__) || defined(__i386__) - { unsigned int r[12]={0}; unsigned int *w=r; - for(unsigned int f=0x80000002u; f<=0x80000004u; f++,w+=4) - __get_cpuid(f,&w[0],&w[1],&w[2],&w[3]); - char *b=(char*)r; b[47]=0; while(*b==' ')b++; - snprintf(cpu,cn,"%s",b); } -#endif -#else - FILE *ci=fopen("/proc/cpuinfo","r"); - if(ci){ char ln[256]; - while(fgets(ln,sizeof(ln),ci)) if(!strncmp(ln,"model name",10)){ - char *p=strchr(ln,':'); if(p){ p++; while(*p==' ')p++; - int n=(int)strlen(p); if(n>0&&p[n-1]=='\n')p[--n]=0; - snprintf(cpu,cn,"%s",p); } break; } - fclose(ci); } -#endif - *cores=0; -#ifdef _WIN32 - { SYSTEM_INFO si; GetSystemInfo(&si); *cores=(int)si.dwNumberOfProcessors; } -#elif defined(_SC_NPROCESSORS_ONLN) - *cores=(int)sysconf(_SC_NPROCESSORS_ONLN); -#endif - *ram_total=*ram_avail=0; -#ifdef _WIN32 - compat_meminfo(ram_total,ram_avail); /* GlobalMemoryStatusEx, gia' in compat.h */ -#else - FILE *mi=fopen("/proc/meminfo","r"); - if(mi){ char ln[256]; double mt=0,ma=0; - while(fgets(ln,sizeof(ln),mi)){ - if(sscanf(ln,"MemTotal: %lf",&mt)==1) *ram_total=mt/1e6; - if(sscanf(ln,"MemAvailable: %lf",&ma)==1) *ram_avail=ma/1e6; - } fclose(mi); } -#endif -} - -static void hwinfo_emit(Model *m){ - Cfg *c=&m->c; (void)c; /* silence -Wunused on builds without /proc (#148 report) */ - char cpu[256]; int cores; double ram_total,ram_avail; - hw_probe(cpu,sizeof(cpu),&cores,&ram_total,&ram_avail); - /* GPU */ - int ngpu=0; double vram_total=0; - char gpu_name[128]=""; -#ifdef COLI_CUDA - ngpu=g_cuda_ndev; vram_total=m->gpu_expert_bytes/1e9; - for(int i=0;i0){ - /* We don't have a device-name API; parse from the init log line stored in stderr. - * Simpler: just read the nvidia driver sysfs or use a fixed label. */ - snprintf(gpu_name,sizeof(gpu_name),"CUDA device x%d",g_cuda_ndev); - } -#endif - printf("HWINFO %d %.1f %.1f %d %.1f %s|%s\n", - cores,ram_total,ram_avail,ngpu,vram_total,cpu,gpu_name); - fflush(stdout); -} - -static void tiers_emit(Model *m){ - Cfg *c=&m->c; int nsp=0; - for(int i=0;in_layers;i++) if(m->L[i].sparse) nsp++; - int total=(nsp+(m->has_mtp?1:0))*c->n_experts; - int pinned=0,lru=0; - for(int i=0;i<=c->n_layers;i++){ pinned+=m->npin?m->npin[i]:0; lru+=m->ecn?m->ecn[i]:0; } - int vram=0; double vram_gb=0; -#ifdef COLI_CUDA - vram=m->gpu_expert_count; vram_gb=m->gpu_expert_bytes/1e9; -#endif - int ram=pinned-vram+lru; if(ram<0) ram=0; - int disk=total-vram-ram; if(disk<0) disk=0; - double eb=(double)expert_bytes_probe(m,m->ebits); - printf("TIERS %d %d %d %.2f %.2f\n",vram,ram,disk,vram_gb,ram*eb/1e9); - fflush(stdout); -} - -/* EMAP: 1 byte/expert (2bit tier: 0=disk 1=RAM 2=VRAM | 6bit heat log2-bucket), - * righe = layer sparsi in ordine (+MTP se presente), colonne = n_experts. Hex. */ -static void emap_emit(Model *m){ - Cfg *c=&m->c; - int rows=0; - for(int i=0;in_layers;i++) if(m->L[i].sparse) rows++; - int has_mtp = m->has_mtp && m->eusage[c->n_layers]; - if(has_mtp) rows++; - int cols=c->n_experts; - char *hex=malloc((size_t)rows*cols*2+1); int w=0; - for(int i=0;i<=c->n_layers;i++){ - int is_row = (in_layers && m->L[i].sparse) || (i==c->n_layers && has_mtp); - if(!is_row) continue; - for(int e=0;epin[i]; - for(int z=0;znpin[i];z++) if(P[z].eid==e){ -#ifdef COLI_CUDA - tier = P[z].g.cuda?2:1; -#else - tier = 1; -#endif - break; } - if(!tier && m->ecache && m->ecache[i]) - for(int z=0;zecn[i];z++) if(m->ecache[i][z].eid==e){ tier=1; break; } - uint32_t u = m->eusage[i]?m->eusage[i][e]:0; - int heat=0; while(u){ heat++; u>>=1; } if(heat>63) heat=63; - int b=(tier<<6)|heat; - hex[w++]="0123456789abcdef"[b>>4]; hex[w++]="0123456789abcdef"[b&15]; - } - } - hex[w]=0; - printf("EMAP %d %d %s\n",rows,cols,hex); fflush(stdout); free(hex); -} - -/* HITS: bitmap 1bit/expert (stesso ordine di EMAP), poi azzera per il turno dopo. */ -static void hits_emit(Model *m){ - Cfg *c=&m->c; if(!g_ehit) return; - int rows=0; - for(int i=0;in_layers;i++) if(m->L[i].sparse) rows++; - int has_mtp = m->has_mtp && m->eusage[c->n_layers]; - if(has_mtp) rows++; - int cols=c->n_experts, nb=(rows*cols+7)/8; - uint8_t *bm=calloc(nb,1); int bit=0; - for(int i=0;i<=c->n_layers;i++){ - int is_row = (in_layers && m->L[i].sparse) || (i==c->n_layers && has_mtp); - if(!is_row) continue; - for(int e=0;e>3]|=1<<(bit&7); g_ehit[i][e]=0; } - } - char *hex=malloc((size_t)nb*2+1); int w=0; - for(int b=0;b>4]; hex[w++]="0123456789abcdef"[bm[b]&15]; } - hex[w]=0; - printf("HITS %d %d %s\n",rows,cols,hex); fflush(stdout); free(hex); free(bm); -} - -/* scarica su file l'istogramma d'uso degli expert: righe "layer eid count" (per PIN). - * Include la riga MTP (layer n_layers). Scrittura atomica (tmp+rename): viene chiamata - * anche a ogni turno di serve e il processo puo' morire in qualsiasi momento. */ -static void stats_dump_q(Model *m, const char *path, int quiet){ - char tmp[2100]; snprintf(tmp,sizeof(tmp),"%s.tmp",path); - FILE *f=fopen(tmp,"w"); if(!f){ if(!quiet) perror(tmp); return; } - Cfg *c=&m->c; int64_t tot=0, nz=0; - for(int i=0;i<=c->n_layers;i++){ if(!m->eusage[i]) continue; - for(int e=0;en_experts;e++) if(m->eusage[i][e]){ fprintf(f,"%d %d %u\n",i,e,m->eusage[i][e]); tot+=m->eusage[i][e]; nz++; } } - fclose(f); rename(tmp,path); - if(!quiet) fprintf(stderr,"[STATS] %lld selections across %lld distinct experts -> %s\n",(long long)tot,(long long)nz,path); -} -static void stats_dump(Model *m, const char *path){ stats_dump_q(m,path,0); } - -/* CACHE CHE IMPARA: istogramma d'uso PERSISTENTE in /.coli_usage. - * Caricato all'avvio (i contatori ripartono dalla storia), salvato a ogni turno: - * piu' usi colibri', meglio l'auto-pin conosce i TUOI expert caldi. */ -static char g_usage_path[2100]=""; -static int64_t usage_load(Model *m, const char *path){ - FILE *f=fopen(path,"r"); if(!f) return 0; - Cfg *c=&m->c; int l,e; uint32_t cnt; int64_t tot=0; - while(fscanf(f,"%d %d %u",&l,&e,&cnt)==3) - if(l>=0&&l<=c->n_layers&&e>=0&&en_experts&&m->eusage[l]){ m->eusage[l][e]+=cnt; tot+=cnt; } - fclose(f); return tot; -} -static void usage_save(Model *m){ if(g_usage_path[0]) stats_dump_q(m,g_usage_path,1); } +/* telemetry, stats, usage persistence — moved to telemetry.h */ /* HOT-STORE ("il redis del colibri'"): carica in RAM, UNA VOLTA e per sempre, i top expert * per frequenza d'uso misurata (file STATS di un run precedente), entro un budget in GB. diff --git a/c/kv_persist.h b/c/kv_persist.h new file mode 100644 index 0000000..51289a8 --- /dev/null +++ b/c/kv_persist.h @@ -0,0 +1,121 @@ +/* kv_persist.h — .coli_kv on-disk KV cache persistence. + * Conversations reopen warm across engine restarts: the compressed MLA KV-cache + * is appended incrementally after every turn, crash-safe (nrec written last). + * Include after Model/KVState/Cfg are defined; requires now_s() and g_draft. */ +#ifndef KV_PERSIST_H +#define KV_PERSIST_H + +static int g_kvsave=1; +#define KV_MAGIC "COLIKV1\0" + +static void kv_hdr(Model *m, int32_t *h, int nrec){ + Cfg *c=&m->c; int nic=0; + for(int i=0;in_layers;i++) if(m->Ic && m->Ic[i]) nic++; + h[0]=c->n_layers; h[1]=c->kv_lora; h[2]=c->qk_rope; + h[3]=m->has_dsa?c->index_hd:0; h[4]=nic; h[5]=c->vocab; h[6]=nrec; h[7]=0; +} + +static int64_t kv_rec_bytes(Model *m){ + Cfg *c=&m->c; + int64_t rec = 4 + (int64_t)c->n_layers*(c->kv_lora+c->qk_rope)*4; + if(m->has_dsa) for(int i=0;in_layers;i++) if(m->Ic[i]) rec+=(int64_t)c->index_hd*4; + return rec; +} + +static int kv_disk_open(Model *m){ + KVState *k=m->kv; + if(k->disk_fp) return 1; + k->disk_fp=fopen(k->disk_path,"r+b"); + if(!k->disk_fp){ + k->disk_fp=fopen(k->disk_path,"wb"); + if(!k->disk_fp) return 0; + int32_t h[8]; kv_hdr(m,h,0); + fwrite(KV_MAGIC,1,8,k->disk_fp); fwrite(h,4,8,k->disk_fp); + fflush(k->disk_fp); + fclose(k->disk_fp); + k->disk_fp=fopen(k->disk_path,"r+b"); + if(!k->disk_fp) return 0; + } + return 1; +} + +static void kv_disk_truncate(Model *m, int nrec){ + if(!g_kvsave) return; + KVState *k=m->kv; + if(k->disk_fp){ fclose(k->disk_fp); k->disk_fp=NULL; } + FILE *f=fopen(k->disk_path,"r+b"); + if(!f){ k->disk_nrec=0; return; } + k->disk_nrec=nrec; + int32_t nr=nrec; fseek(f,8+6*4,SEEK_SET); fwrite(&nr,4,1,f); + fflush(f); fclose(f); +} + +static void kv_disk_reset(Model *m){ kv_disk_truncate(m,0); } + +static void kv_disk_append(Model *m, const int *hist, int len){ + KVState *k=m->kv; + if(!g_kvsave || len<=k->disk_nrec) return; + Cfg *c=&m->c; + if(!kv_disk_open(m)) return; + FILE *f=k->disk_fp; + int64_t rec = kv_rec_bytes(m); + if(rec > k->disk_buf_cap){ + uint8_t *nb=realloc(k->disk_buf, rec); + if(!nb) return; + k->disk_buf=nb; k->disk_buf_cap=rec; + } + fseek(f, 8+8*4 + (int64_t)k->disk_nrec*rec, SEEK_SET); + for(int p=k->disk_nrec;pdisk_buf; + *(int32_t*)b = hist[p]; b+=4; + for(int i=0;in_layers;i++){ + memcpy(b, m->Lc[i]+(int64_t)p*c->kv_lora, (size_t)c->kv_lora*4); b+=c->kv_lora*4; + memcpy(b, m->Rc[i]+(int64_t)p*c->qk_rope,(size_t)c->qk_rope*4); b+=c->qk_rope*4; + } + if(m->has_dsa) for(int i=0;in_layers;i++) if(m->Ic[i]){ + memcpy(b, m->Ic[i]+(int64_t)p*c->index_hd, (size_t)c->index_hd*4); b+=c->index_hd*4; + } + fwrite(k->disk_buf, 1, (size_t)rec, f); + } + fflush(f); + int32_t nr=len; fseek(f,8+6*4,SEEK_SET); fwrite(&nr,4,1,f); + fflush(f); + k->disk_nrec=len; +} + +static int kv_disk_load(Model *m, int *hist, int maxctx){ + if(!g_kvsave) return 0; + KVState *k=m->kv; + Cfg *c=&m->c; + FILE *f=fopen(k->disk_path,"rb"); if(!f) return 0; + char mg[8]; int32_t h[8], w[8]; kv_hdr(m,w,0); + if(fread(mg,1,8,f)!=8 || memcmp(mg,KV_MAGIC,8) || fread(h,4,8,f)!=8 || + h[0]!=w[0]||h[1]!=w[1]||h[2]!=w[2]||h[3]!=w[3]||h[4]!=w[4]||h[5]!=w[5]){ + fprintf(stderr,"[KV] ignoring .coli_kv from a different model or version\n"); fclose(f); return 0; } + int nrec=h[6]; + if(nrec<1){ fclose(f); return 0; } + if(nrec>=maxctx-8-g_draft){ + fprintf(stderr,"[KV] saved conversation (%d tokens) exceeds the context: starting over\n",nrec); + fclose(f); return 0; } + double t0=now_s(); + for(int p=0;pn_layers;i++){ + if(fread(m->Lc[i]+(int64_t)p*c->kv_lora, 4, c->kv_lora, f)!=(size_t)c->kv_lora || + fread(m->Rc[i]+(int64_t)p*c->qk_rope, 4, c->qk_rope, f)!=(size_t)c->qk_rope){ nrec=p; goto out; } + } + if(m->has_dsa) for(int i=0;in_layers;i++) if(m->Ic[i]) + if(fread(m->Ic[i]+(int64_t)p*c->index_hd, 4, c->index_hd, f)!=(size_t)c->index_hd){ nrec=p; goto out; } + } +out: + fclose(f); + if(nrec>0){ + if(m->has_mtp) m->kv_start[c->n_layers]=-1; + fprintf(stderr,"[KV] resumed conversation from disk: %d tokens in %.1fs (no re-prefill)\n", + nrec, now_s()-t0); + } + k->disk_nrec=nrec; + return nrec; +} + +#endif /* KV_PERSIST_H */ diff --git a/c/quant.h b/c/quant.h new file mode 100644 index 0000000..928de96 --- /dev/null +++ b/c/quant.h @@ -0,0 +1,672 @@ +/* quant.h — quantized matmul kernels (header-only, all functions static). + * Multi-architecture SIMD: AVX2 / AVX-512 / AVX-VNNI / ARM NEON / NEON-SDOT / + * NEON-i8mm / POWER VSX. Pure compute — no Model or QT dependency. */ +#ifndef COLI_QUANT_H +#define COLI_QUANT_H + +#include +#include +#include +#include +#include + +#ifdef _OPENMP +#include +#endif + +/* ---- SIMD includes -------------------------------------------------------- */ +#ifdef __AVX2__ +#include +static inline float hsum256(__m256 v){ + __m128 lo=_mm256_castps256_ps128(v), hi=_mm256_extractf128_ps(v,1); + lo=_mm_add_ps(lo,hi); __m128 sh=_mm_movehl_ps(lo,lo); lo=_mm_add_ps(lo,sh); + sh=_mm_shuffle_ps(lo,lo,1); lo=_mm_add_ss(lo,sh); return _mm_cvtss_f32(lo); +} +static inline int hsum256_i32(__m256i v){ + __m128i lo=_mm256_castsi256_si128(v), hi=_mm256_extracti128_si256(v,1); + lo=_mm_add_epi32(lo,hi); lo=_mm_hadd_epi32(lo,lo); lo=_mm_hadd_epi32(lo,lo); + return _mm_cvtsi128_si32(lo); +} +#endif +#if defined(__AVXVNNI__) && defined(__AVX2__) +static inline int hsum128_i32(__m128i v){ + v=_mm_hadd_epi32(v,v); v=_mm_hadd_epi32(v,v); return _mm_cvtsi128_si32(v); +} +#endif +#ifdef __ARM_NEON +#include +#endif +#ifdef __VSX__ +#include +#undef vector +#undef pixel +#undef bool +#endif + +/* ---- AVX-512 int4->float accumulator -------------------------------------- */ +#if defined(__AVX512F__) && defined(__AVX512BW__) +static int g_i4_acc512=1; +static inline float dot_i4f_avx512(const uint8_t *w,const float *x,int I){ + const __m128i m4=_mm_set1_epi8(0x0F); const __m512i b8=_mm512_set1_epi32(8); + __m512 acc0=_mm512_setzero_ps(),acc1=_mm512_setzero_ps(); int i=0; + for(;i+32<=I;i+=32){ __m128i by=_mm_loadu_si128((const __m128i*)(w+(i>>1))); + __m128i lo=_mm_and_si128(by,m4),hi=_mm_and_si128(_mm_srli_epi16(by,4),m4); + __m128i n0=_mm_unpacklo_epi8(lo,hi),n1=_mm_unpackhi_epi8(lo,hi); + __m512 w0=_mm512_cvtepi32_ps(_mm512_sub_epi32(_mm512_cvtepu8_epi32(n0),b8)); + __m512 w1=_mm512_cvtepi32_ps(_mm512_sub_epi32(_mm512_cvtepu8_epi32(n1),b8)); + acc0=_mm512_fmadd_ps(_mm512_loadu_ps(x+i),w0,acc0); + acc1=_mm512_fmadd_ps(_mm512_loadu_ps(x+i+16),w1,acc1); + } + return _mm512_reduce_add_ps(_mm512_add_ps(acc0,acc1)); +} +static int i4_acc512_selftest(void){ + enum { N=224 }; uint8_t w[(N+1)/2]; float x[N]; + for(int i=0;i>1]=(uint8_t)(q+8); + else w[i>>1]|=(uint8_t)((q+8)<<4); + x[i]=(float)(((i*29+7)%101)-50)/37.f; + } + for(int n=32;n<=N;n+=32){ + float ref=0; for(int i=0;i>1]>>((i&1)*4))&15)-8); + float got=dot_i4f_avx512(w,x,n),tol=2e-5f*(1.f+fabsf(ref)); + if(fabsf(got-ref)>tol){ fprintf(stderr,"AVX512 i4 selftest n=%d: %.9g != %.9g\n",n,got,ref); return 0; } + } + return 1; +} +#endif + +/* ---- y[S,O] = x[S,I] @ W^T, W[O,I] f32 ---------------------------------- */ +static void matmul(float *y, const float *x, const float *W, int S, int I, int O){ + #pragma omp parallel for schedule(static) + for (int o=0;o>1))); + __m128i lo=_mm_and_si128(by,m4), hi=_mm_and_si128(_mm_srli_epi16(by,4),m4); + __m128i nib=_mm_unpacklo_epi8(lo,hi); + __m256 w0=_mm256_cvtepi32_ps(_mm256_sub_epi32(_mm256_cvtepu8_epi32(nib),b8)); + __m256 w1=_mm256_cvtepi32_ps(_mm256_sub_epi32(_mm256_cvtepu8_epi32(_mm_srli_si128(nib,8)),b8)); + acc=_mm256_fmadd_ps(_mm256_loadu_ps(xs+i), w0, acc); + acc=_mm256_fmadd_ps(_mm256_loadu_ps(xs+i+8), w1, acc); } + a=hsum256(acc); +#elif defined(__ARM_NEON) + const uint8x8_t m4=vdup_n_u8(0x0F); const int8x8_t b8=vdup_n_s8(8); + float32x4_t ac0=vdupq_n_f32(0), ac1=vdupq_n_f32(0); + for(;i+16<=I;i+=16){ uint8x8_t by=vld1_u8(w+(i>>1)); + uint8x8x2_t z=vzip_u8(vand_u8(by,m4), vshr_n_u8(by,4)); + int16x8_t w0=vmovl_s8(vsub_s8(vreinterpret_s8_u8(z.val[0]),b8)); + int16x8_t w1=vmovl_s8(vsub_s8(vreinterpret_s8_u8(z.val[1]),b8)); + ac0=vfmaq_f32(ac0, vld1q_f32(xs+i), vcvtq_f32_s32(vmovl_s16(vget_low_s16(w0)))); + ac1=vfmaq_f32(ac1, vld1q_f32(xs+i+4), vcvtq_f32_s32(vmovl_s16(vget_high_s16(w0)))); + ac0=vfmaq_f32(ac0, vld1q_f32(xs+i+8), vcvtq_f32_s32(vmovl_s16(vget_low_s16(w1)))); + ac1=vfmaq_f32(ac1, vld1q_f32(xs+i+12), vcvtq_f32_s32(vmovl_s16(vget_high_s16(w1)))); } + a=vaddvq_f32(vaddq_f32(ac0,ac1)); +#endif +#if defined(__AVX512F__) && defined(__AVX512BW__) + } +#endif + for(;i+1>1]; int lo=(int)(byte&0xF)-8, hi=(int)(byte>>4)-8; + a += xs[i]*(float)lo + xs[i+1]*(float)hi; } + if(i>1]; int lo=(int)(byte&0xF)-8; a += xs[i]*(float)lo; } + y[(int64_t)s*O+o]=a*sc; } } +} + +/* ---- y[S,O] = x[S,I] @ W^T, W int4 packed + per-GROUP scales (fmt=4) ----- */ +static void matmul_i4_grouped(float *y, const float *x, const uint8_t *q4, const float *scale, + int S, int I, int O, int gs){ + int rb=(I+1)/2; int ng=(I+gs-1)/gs; + #pragma omp parallel for schedule(static) + for(int o=0;oI) glen=I-base; + float sc=scl[g]; + int i=base; +#ifdef __AVX2__ + const __m128i m4=_mm_set1_epi8(0x0F); const __m256i b8=_mm256_set1_epi32(8); + __m256 acc=_mm256_setzero_ps(); + for(; i+16<=base+glen; i+=16){ __m128i by=_mm_loadl_epi64((const __m128i*)(w+(i>>1))); + __m128i lo=_mm_and_si128(by,m4),hi=_mm_and_si128(_mm_srli_epi16(by,4),m4); + __m128i nib=_mm_unpacklo_epi8(lo,hi); + __m256 w0=_mm256_cvtepi32_ps(_mm256_sub_epi32(_mm256_cvtepu8_epi32(nib),b8)); + __m256 w1=_mm256_cvtepi32_ps(_mm256_sub_epi32(_mm256_cvtepu8_epi32(_mm_srli_si128(nib,8)),b8)); + acc=_mm256_fmadd_ps(_mm256_loadu_ps(xs+i), w0, acc); + acc=_mm256_fmadd_ps(_mm256_loadu_ps(xs+i+8), w1, acc); } + a+=hsum256(acc)*sc; +#endif + for(; i>1]; + a+=(xs[i]*(float)((int)(byte&0xF)-8)+xs[i+1]*(float)((int)(byte>>4)-8))*sc; } + else { uint8_t byte=w[i>>1]; a+=xs[i]*(float)((int)(byte&0xF)-8)*sc; } + } + } + y[(int64_t)s*O+o]=a; + } + } +} + +/* ---- fused gate+up: one OMP dispatch for both matrices -------------------- */ +static void matmul_i4_pair(float *yg, float *yu, const float *x, + const uint8_t *qg, const float *sg, + const uint8_t *qu, const float *su, int I, int O){ + int rb=(I+1)/2; + #pragma omp parallel for schedule(static) + for(int z=0;z<2*O;z++){ + int o=z>1))); + __m128i lo=_mm_and_si128(by,m4),hi=_mm_and_si128(_mm_srli_epi16(by,4),m4); + __m128i nib=_mm_unpacklo_epi8(lo,hi); + __m256 w0=_mm256_cvtepi32_ps(_mm256_sub_epi32(_mm256_cvtepu8_epi32(nib),b8)); + __m256 w1=_mm256_cvtepi32_ps(_mm256_sub_epi32(_mm256_cvtepu8_epi32(_mm_srli_si128(nib,8)),b8)); + acc=_mm256_fmadd_ps(_mm256_loadu_ps(x+i),w0,acc); + acc=_mm256_fmadd_ps(_mm256_loadu_ps(x+i+8),w1,acc); } + a=hsum256(acc); +#elif defined(__ARM_NEON) + const uint8x8_t m4=vdup_n_u8(0x0F); const int8x8_t b8=vdup_n_s8(8); + float32x4_t ac0=vdupq_n_f32(0),ac1=vdupq_n_f32(0); + for(;i+16<=I;i+=16){ uint8x8_t by=vld1_u8(w+(i>>1)); + uint8x8x2_t n=vzip_u8(vand_u8(by,m4),vshr_n_u8(by,4)); + int16x8_t w0=vmovl_s8(vsub_s8(vreinterpret_s8_u8(n.val[0]),b8)); + int16x8_t w1=vmovl_s8(vsub_s8(vreinterpret_s8_u8(n.val[1]),b8)); + ac0=vfmaq_f32(ac0,vld1q_f32(x+i),vcvtq_f32_s32(vmovl_s16(vget_low_s16(w0)))); + ac1=vfmaq_f32(ac1,vld1q_f32(x+i+4),vcvtq_f32_s32(vmovl_s16(vget_high_s16(w0)))); + ac0=vfmaq_f32(ac0,vld1q_f32(x+i+8),vcvtq_f32_s32(vmovl_s16(vget_low_s16(w1)))); + ac1=vfmaq_f32(ac1,vld1q_f32(x+i+12),vcvtq_f32_s32(vmovl_s16(vget_high_s16(w1)))); } + a=vaddvq_f32(vaddq_f32(ac0,ac1)); +#endif +#if defined(__AVX512F__) && defined(__AVX512BW__) + } +#endif + for(;i+1>1]; a+=x[i]*(float)((b&15)-8)+x[i+1]*(float)((b>>4)-8); } + if(i>1]&15)-8); + (z>2))); + __m128i p0=_mm_and_si128(by,m2), p1=_mm_and_si128(_mm_srli_epi16(by,2),m2); + __m128i p2=_mm_and_si128(_mm_srli_epi16(by,4),m2), p3=_mm_and_si128(_mm_srli_epi16(by,6),m2); + __m128i lo=_mm_unpacklo_epi8(p0,p1), hi=_mm_unpacklo_epi8(p2,p3); + __m128i nib=_mm_unpacklo_epi16(lo,hi); + __m256 w0=_mm256_cvtepi32_ps(_mm256_sub_epi32(_mm256_cvtepu8_epi32(nib),b2)); + __m256 w1=_mm256_cvtepi32_ps(_mm256_sub_epi32(_mm256_cvtepu8_epi32(_mm_srli_si128(nib,8)),b2)); + acc=_mm256_fmadd_ps(_mm256_loadu_ps(xs+i), w0, acc); + acc=_mm256_fmadd_ps(_mm256_loadu_ps(xs+i+8), w1, acc); } + a=hsum256(acc); +#elif defined(__ARM_NEON) + const uint8x8_t m2v=vdup_n_u8(3); const int8x8_t b2v=vdup_n_s8(2); + float32x4_t ac0=vdupq_n_f32(0), ac1=vdupq_n_f32(0); + for(;i+16<=I;i+=16){ uint32_t wd; memcpy(&wd, w+(i>>2), 4); + uint8x8_t by=vreinterpret_u8_u32(vdup_n_u32(wd)); + uint8x8x2_t z01=vzip_u8(vand_u8(by,m2v), vand_u8(vshr_n_u8(by,2),m2v)); + uint8x8x2_t z23=vzip_u8(vand_u8(vshr_n_u8(by,4),m2v), vshr_n_u8(by,6)); + uint16x4x2_t zz=vzip_u16(vreinterpret_u16_u8(z01.val[0]), vreinterpret_u16_u8(z23.val[0])); + int16x8_t w0=vmovl_s8(vsub_s8(vreinterpret_s8_u16(zz.val[0]),b2v)); + int16x8_t w1=vmovl_s8(vsub_s8(vreinterpret_s8_u16(zz.val[1]),b2v)); + ac0=vfmaq_f32(ac0, vld1q_f32(xs+i), vcvtq_f32_s32(vmovl_s16(vget_low_s16(w0)))); + ac1=vfmaq_f32(ac1, vld1q_f32(xs+i+4), vcvtq_f32_s32(vmovl_s16(vget_high_s16(w0)))); + ac0=vfmaq_f32(ac0, vld1q_f32(xs+i+8), vcvtq_f32_s32(vmovl_s16(vget_low_s16(w1)))); + ac1=vfmaq_f32(ac1, vld1q_f32(xs+i+12), vcvtq_f32_s32(vmovl_s16(vget_high_s16(w1)))); } + a=vaddvq_f32(vaddq_f32(ac0,ac1)); +#endif + for(;i>2]; int sh=(i&3)*2; a += xs[i]*(float)((int)((byte>>sh)&3)-2); } + y[(int64_t)s*O+o]=a*sc; } } +} + +/* ---- IDOT: integer dot kernels (int8-quantized activations) --------------- */ +#if defined(__AVX512VNNI__) && defined(__AVX512BW__) +#define IDOT_KERNEL "avx512-vnni" +#elif defined(__AVXVNNI__) && defined(__AVX2__) +#define IDOT_KERNEL "avx-vnni" +#elif defined(__AVX2__) +#define IDOT_KERNEL "avx2" +#elif defined(__ARM_NEON) && defined(__ARM_FEATURE_MATMUL_INT8) +#define IDOT_KERNEL "neon-i8mm" +#elif defined(__ARM_NEON) +#define IDOT_KERNEL "neon" +#elif defined(__VSX__) +#define IDOT_KERNEL "vsx" +#else +#define IDOT_KERNEL "scalar" +#endif +static int g_idot=1; +#if defined(__ARM_NEON) && defined(__ARM_FEATURE_DOTPROD) +static int g_i4s=1; +#elif defined(__VSX__) +static int g_i4s=1; +#else +static int g_i4s=2; +#endif + +static inline float qrow_i8(const float *x, int8_t *q, int I){ + float amax=0; for(int i=0;iamax)amax=a; } + float s=amax/127.f; if(s<1e-12f) s=1e-12f; float inv=1.f/s; + for(int i=0;i>1))); + __m256i lo=_mm256_and_si256(by,m4v), hi=_mm256_and_si256(_mm256_srli_epi16(by,4),m4v); + __m256i z0=_mm256_unpacklo_epi8(lo,hi), z1=_mm256_unpackhi_epi8(lo,hi); + __m512i wv=_mm512_sub_epi8(_mm512_inserti64x4(_mm512_castsi256_si512(z0),z1,1),b8v); + __m512i xv=_mm512_permutexvar_epi64(xidx,_mm512_loadu_si512((const void*)(x+i))); + __mmask64 neg=_mm512_movepi8_mask(wv); + __m512i xs=_mm512_mask_sub_epi8(xv,neg,_mm512_setzero_si512(),xv); + acc=_mm512_dpbusd_epi32(acc,_mm512_abs_epi8(wv),xs); + } + sum=_mm512_reduce_add_epi32(acc); +#elif defined(__AVXVNNI__) && defined(__AVX2__) + const __m128i m4=_mm_set1_epi8(0x0F); const __m128i b8=_mm_set1_epi8(8); + __m128i acc=_mm_setzero_si128(); + for(;i+32<=I;i+=32){ + __m128i by=_mm_loadu_si128((const __m128i*)(w4+(i>>1))); + __m128i lo=_mm_and_si128(by,m4), hi=_mm_and_si128(_mm_srli_epi16(by,4),m4); + __m128i n0=_mm_unpacklo_epi8(lo,hi), n1=_mm_unpackhi_epi8(lo,hi); + __m128i w0=_mm_sub_epi8(n0,b8), w1=_mm_sub_epi8(n1,b8); + __m128i x0=_mm_loadu_si128((const __m128i*)(x+i)); + __m128i x1=_mm_loadu_si128((const __m128i*)(x+i+16)); + acc=_mm_dpbusd_epi32(acc,_mm_abs_epi8(w0),_mm_sign_epi8(x0,w0)); + acc=_mm_dpbusd_epi32(acc,_mm_abs_epi8(w1),_mm_sign_epi8(x1,w1)); + } + sum=hsum128_i32(acc); +#elif defined(__AVX2__) + const __m128i m4=_mm_set1_epi8(0x0F); const __m256i b8=_mm256_set1_epi8(8); + const __m256i ones=_mm256_set1_epi16(1); + __m256i acc=_mm256_setzero_si256(); + for(;i+32<=I;i+=32){ + __m128i by=_mm_loadu_si128((const __m128i*)(w4+(i>>1))); + __m128i lo=_mm_and_si128(by,m4), hi=_mm_and_si128(_mm_srli_epi16(by,4),m4); + __m128i n0=_mm_unpacklo_epi8(lo,hi), n1=_mm_unpackhi_epi8(lo,hi); + __m256i wv=_mm256_sub_epi8(_mm256_set_m128i(n1,n0),b8); + __m256i xv=_mm256_loadu_si256((const __m256i*)(x+i)); + __m256i p=_mm256_maddubs_epi16(_mm256_sign_epi8(wv,wv),_mm256_sign_epi8(xv,wv)); + acc=_mm256_add_epi32(acc,_mm256_madd_epi16(p,ones)); + } + sum=hsum256_i32(acc); +#elif defined(__ARM_NEON) + const uint8x16_t m4q=vdupq_n_u8(0x0F); const int8x16_t b8q=vdupq_n_s8(8); +#if defined(__ARM_FEATURE_DOTPROD) + int32x4_t a0=vdupq_n_s32(0),a1=vdupq_n_s32(0),a2=vdupq_n_s32(0),a3=vdupq_n_s32(0); + for(;i+64<=I;i+=64){ + uint8x16_t byA=vld1q_u8(w4+(i>>1)), byB=vld1q_u8(w4+(i>>1)+16); + uint8x16x2_t zA=vzipq_u8(vandq_u8(byA,m4q), vshrq_n_u8(byA,4)); + uint8x16x2_t zB=vzipq_u8(vandq_u8(byB,m4q), vshrq_n_u8(byB,4)); + a0=vdotq_s32(a0,vsubq_s8(vreinterpretq_s8_u8(zA.val[0]),b8q),vld1q_s8(x+i)); + a1=vdotq_s32(a1,vsubq_s8(vreinterpretq_s8_u8(zA.val[1]),b8q),vld1q_s8(x+i+16)); + a2=vdotq_s32(a2,vsubq_s8(vreinterpretq_s8_u8(zB.val[0]),b8q),vld1q_s8(x+i+32)); + a3=vdotq_s32(a3,vsubq_s8(vreinterpretq_s8_u8(zB.val[1]),b8q),vld1q_s8(x+i+48)); + } + int32x4_t acc=vaddq_s32(vaddq_s32(a0,a1),vaddq_s32(a2,a3)); + for(;i+32<=I;i+=32){ + uint8x16_t by=vld1q_u8(w4+(i>>1)); + uint8x16x2_t z=vzipq_u8(vandq_u8(by,m4q), vshrq_n_u8(by,4)); + acc=vdotq_s32(acc,vsubq_s8(vreinterpretq_s8_u8(z.val[0]),b8q),vld1q_s8(x+i)); + acc=vdotq_s32(acc,vsubq_s8(vreinterpretq_s8_u8(z.val[1]),b8q),vld1q_s8(x+i+16)); + } + sum=vaddvq_s32(acc); +#else + int32x4_t acc=vdupq_n_s32(0); + for(;i+32<=I;i+=32){ + uint8x16_t by=vld1q_u8(w4+(i>>1)); + uint8x16x2_t z=vzipq_u8(vandq_u8(by,m4q), vshrq_n_u8(by,4)); + int8x16_t w0=vsubq_s8(vreinterpretq_s8_u8(z.val[0]),b8q); + int8x16_t w1=vsubq_s8(vreinterpretq_s8_u8(z.val[1]),b8q); + int8x16_t x0=vld1q_s8(x+i), x1=vld1q_s8(x+i+16); + int16x8_t p=vmull_s8(vget_low_s8(w0),vget_low_s8(x0)); + p=vmlal_s8(p,vget_high_s8(w0),vget_high_s8(x0)); + acc=vpadalq_s16(acc,p); + p=vmull_s8(vget_low_s8(w1),vget_low_s8(x1)); + p=vmlal_s8(p,vget_high_s8(w1),vget_high_s8(x1)); + acc=vpadalq_s16(acc,p); + } + sum=vaddvq_s32(acc); +#endif +#elif defined(__VSX__) + const __vector unsigned char m4v=vec_splats((unsigned char)0x0F); + const __vector unsigned char sh4=vec_splats((unsigned char)4); + const __vector signed char b8v=vec_splats((signed char)8); + const __vector signed char vz=vec_splats((signed char)0); + __vector signed int acc=vec_splats(0); + for(;i+32<=I;i+=32){ + __vector unsigned char by=vec_xl(0,w4+(i>>1)); + __vector unsigned char lo=vec_and(by,m4v), hi=vec_sr(by,sh4); + __vector signed char w0=vec_sub((__vector signed char)vec_mergeh(lo,hi),b8v); + __vector signed char w1=vec_sub((__vector signed char)vec_mergel(lo,hi),b8v); + __vector signed char x0=vec_xl(0,(const signed char*)(x+i)); + __vector signed char x1=vec_xl(0,(const signed char*)(x+i+16)); + __vector __bool char n0=vec_cmplt(w0,vz), n1=vec_cmplt(w1,vz); + acc=vec_msum(vec_sel(x0,vec_sub(vz,x0),n0), + (__vector unsigned char)vec_sel(w0,vec_sub(vz,w0),n0),acc); + acc=vec_msum(vec_sel(x1,vec_sub(vz,x1),n1), + (__vector unsigned char)vec_sel(w1,vec_sub(vz,w1),n1),acc); + } + sum=vec_extract(acc,0)+vec_extract(acc,1)+vec_extract(acc,2)+vec_extract(acc,3); +#endif + for(;i+1>1]; sum+=((int)(b&0xF)-8)*x[i]+((int)(b>>4)-8)*x[i+1]; } + if(i>1]; sum+=((int)(b&0xF)-8)*x[i]; } + return sum; +} + +/* ---- ARM i8mm SMMLA tiled kernels ---------------------------------------- */ +#if defined(__ARM_NEON) && defined(__ARM_FEATURE_MATMUL_INT8) +static inline int32x4_t mm_tile16(int32x4_t acc, int8x16_t wo, int8x16_t wo1, + int8x16_t xs, int8x16_t xs1){ + acc=vmmlaq_s32(acc, vcombine_s8(vget_low_s8(wo), vget_low_s8(wo1)), + vcombine_s8(vget_low_s8(xs), vget_low_s8(xs1))); + return vmmlaq_s32(acc, vcombine_s8(vget_high_s8(wo), vget_high_s8(wo1)), + vcombine_s8(vget_high_s8(xs), vget_high_s8(xs1))); +} +static void matmul_q_idot_mm(float *y, const int8_t *xq, const float *sx, const int8_t *q, + const float *scale, int S, int I, int O){ + #pragma omp parallel for schedule(static) + for(int o=0;o<(O&~1);o+=2){ + const int8_t *wo=q+(int64_t)o*I, *wo1=q+(int64_t)(o+1)*I; + float sc0=scale[o], sc1=scale[o+1]; + for(int s=0;s<(S&~1);s+=2){ + const int8_t *xs=xq+(int64_t)s*I, *xs1=xq+(int64_t)(s+1)*I; + int32x4_t a0=vdupq_n_s32(0),a1=vdupq_n_s32(0),a2=vdupq_n_s32(0),a3=vdupq_n_s32(0); int i=0; + for(;i+64<=I;i+=64){ + a0=mm_tile16(a0,vld1q_s8(wo+i), vld1q_s8(wo1+i), vld1q_s8(xs+i), vld1q_s8(xs1+i)); + a1=mm_tile16(a1,vld1q_s8(wo+i+16),vld1q_s8(wo1+i+16),vld1q_s8(xs+i+16),vld1q_s8(xs1+i+16)); + a2=mm_tile16(a2,vld1q_s8(wo+i+32),vld1q_s8(wo1+i+32),vld1q_s8(xs+i+32),vld1q_s8(xs1+i+32)); + a3=mm_tile16(a3,vld1q_s8(wo+i+48),vld1q_s8(wo1+i+48),vld1q_s8(xs+i+48),vld1q_s8(xs1+i+48)); + } + for(;i+16<=I;i+=16) + a0=mm_tile16(a0,vld1q_s8(wo+i),vld1q_s8(wo1+i),vld1q_s8(xs+i),vld1q_s8(xs1+i)); + int32x4_t acc=vaddq_s32(vaddq_s32(a0,a1),vaddq_s32(a2,a3)); + int32_t d00=vgetq_lane_s32(acc,0), d01=vgetq_lane_s32(acc,1); + int32_t d10=vgetq_lane_s32(acc,2), d11=vgetq_lane_s32(acc,3); + for(;i>1)), byo1=vld1q_u8(wo1+(i>>1)); + uint8x16_t cyo=vld1q_u8(wo+(i>>1)+16), cyo1=vld1q_u8(wo1+(i>>1)+16); + uint8x16x2_t zo =vzipq_u8(vandq_u8(byo, m4q), vshrq_n_u8(byo, 4)); + uint8x16x2_t zo1=vzipq_u8(vandq_u8(byo1,m4q), vshrq_n_u8(byo1,4)); + uint8x16x2_t ko =vzipq_u8(vandq_u8(cyo, m4q), vshrq_n_u8(cyo, 4)); + uint8x16x2_t ko1=vzipq_u8(vandq_u8(cyo1,m4q), vshrq_n_u8(cyo1,4)); + a0=mm_tile16(a0, vsubq_s8(vreinterpretq_s8_u8(zo.val[0]),b8q), + vsubq_s8(vreinterpretq_s8_u8(zo1.val[0]),b8q), + vld1q_s8(xs+i), vld1q_s8(xs1+i)); + a1=mm_tile16(a1, vsubq_s8(vreinterpretq_s8_u8(zo.val[1]),b8q), + vsubq_s8(vreinterpretq_s8_u8(zo1.val[1]),b8q), + vld1q_s8(xs+i+16), vld1q_s8(xs1+i+16)); + a2=mm_tile16(a2, vsubq_s8(vreinterpretq_s8_u8(ko.val[0]),b8q), + vsubq_s8(vreinterpretq_s8_u8(ko1.val[0]),b8q), + vld1q_s8(xs+i+32), vld1q_s8(xs1+i+32)); + a3=mm_tile16(a3, vsubq_s8(vreinterpretq_s8_u8(ko.val[1]),b8q), + vsubq_s8(vreinterpretq_s8_u8(ko1.val[1]),b8q), + vld1q_s8(xs+i+48), vld1q_s8(xs1+i+48)); + } + for(;i+32<=I;i+=32){ + uint8x16_t byo=vld1q_u8(wo+(i>>1)), byo1=vld1q_u8(wo1+(i>>1)); + uint8x16x2_t zo =vzipq_u8(vandq_u8(byo, m4q), vshrq_n_u8(byo, 4)); + uint8x16x2_t zo1=vzipq_u8(vandq_u8(byo1,m4q), vshrq_n_u8(byo1,4)); + a0=mm_tile16(a0, vsubq_s8(vreinterpretq_s8_u8(zo.val[0]),b8q), + vsubq_s8(vreinterpretq_s8_u8(zo1.val[0]),b8q), + vld1q_s8(xs+i), vld1q_s8(xs1+i)); + a1=mm_tile16(a1, vsubq_s8(vreinterpretq_s8_u8(zo.val[1]),b8q), + vsubq_s8(vreinterpretq_s8_u8(zo1.val[1]),b8q), + vld1q_s8(xs+i+16), vld1q_s8(xs1+i+16)); + } + int32x4_t acc=vaddq_s32(vaddq_s32(a0,a1),vaddq_s32(a2,a3)); + int32_t d00=vgetq_lane_s32(acc,0), d01=vgetq_lane_s32(acc,1); + int32_t d10=vgetq_lane_s32(acc,2), d11=vgetq_lane_s32(acc,3); + for(;i+1>1], bo1=wo1[i>>1]; + int a0=(int)(bo&0xF)-8, a1=(int)(bo>>4)-8, b0=(int)(bo1&0xF)-8, b1=(int)(bo1>>4)-8; + int u0=xs[i],u1=xs[i+1],v0=xs1[i],v1=xs1[i+1]; + d00+=a0*u0+a1*u1; d01+=a0*v0+a1*v1; d10+=b0*u0+b1*u1; d11+=b0*v0+b1*v1; } + if(i>1], bo1=wo1[i>>1]; + int a0=(int)(bo&0xF)-8, b0=(int)(bo1&0xF)-8; + d00+=a0*xs[i]; d01+=a0*xs1[i]; d10+=b0*xs[i]; d11+=b0*xs1[i]; } + y[(int64_t)s*O+o] =(float)d00*sc0*sx[s]; + y[(int64_t)s*O+(o+1)] =(float)d10*sc1*sx[s]; + y[(int64_t)(s+1)*O+o] =(float)d01*sc0*sx[s+1]; + y[(int64_t)(s+1)*O+(o+1)]=(float)d11*sc1*sx[s+1]; + } + if(S&1){ int s=S-1; const int8_t *xs=xq+(int64_t)s*I; + y[(int64_t)s*O+o] =(float)dot_i4i8(wo, xs,I)*sc0*sx[s]; + y[(int64_t)s*O+(o+1)]=(float)dot_i4i8(wo1,xs,I)*sc1*sx[s]; } + } + if(O&1){ int o=O-1; const uint8_t *w=q4+(int64_t)o*rb; float sc=scale[o]; + #pragma omp parallel for schedule(static) + for(int s=0;s=2){ matmul_q_idot_mm(y,xq,sx,q,scale,S,I,O); return; } +#endif + #pragma omp parallel for schedule(static) + for(int o=0;o=2){ matmul_i4_idot_mm(y,xq,sx,q4,scale,S,I,O); return; } +#endif + #pragma omp parallel for schedule(static) + for(int o=0;og_qscratch.xq_cap){ + int8_t *p=realloc(g_qscratch.xq,xn); + if(!p){ fprintf(stderr,"OOM quant scratch\n"); exit(1); } + g_qscratch.xq=p; g_qscratch.xq_cap=xn; + } + if(sn>g_qscratch.sx_cap){ + float *p=realloc(g_qscratch.sx,sn*sizeof(float)); + if(!p){ fprintf(stderr,"OOM quant scales\n"); exit(1); } + g_qscratch.sx=p; g_qscratch.sx_cap=sn; + } + *xq=g_qscratch.xq; *sx=g_qscratch.sx; +} + +/* ---- f32 -> quantized packing --------------------------------------------- */ +static void quantize_rows(const float *w, int8_t *q, float *scale, int O, int I, int bits){ + int qmax=(1<<(bits-1))-1; + #pragma omp parallel for schedule(static) + for(int o=0;oamax)amax=a; } + float s=amax/qmax; if(s<1e-8f)s=1e-8f; scale[o]=s; + int8_t *qr=q+(int64_t)o*I; + for(int i=0;iqmax)v=qmax; if(v<-qmax-1)v=-qmax-1; qr[i]=(int8_t)v; } + } +} +static void pack_int4(const float *w, uint8_t *q4, float *scale, int O, int I, int bits){ + int qmax=(1<<(bits-1))-1, rb=(I+1)/2; + #pragma omp parallel for schedule(static) + for(int o=0;oamax)amax=a; } + float s=amax/qmax; if(s<1e-8f)s=1e-8f; scale[o]=s; + uint8_t *qr=q4+(int64_t)o*rb; + for(int i=0;iqmax)v0=qmax; if(v0<-8)v0=-8; + int v1=0; if(i+1qmax)v1=qmax; if(v1<-8)v1=-8; } + qr[i>>1] = (uint8_t)((v0+8) | ((v1+8)<<4)); + } + } +} +static void pack_int2(const float *w, uint8_t *q2, float *scale, int O, int I, int bits){ + int qmax=(1<<(bits-1))-1, rb=(I+3)/4; + #pragma omp parallel for schedule(static) + for(int o=0;oamax)amax=a; } + float s=amax/qmax; if(s<1e-8f)s=1e-8f; scale[o]=s; + uint8_t *qr=q2+(int64_t)o*rb; + for(int i=0;iqmax)v=qmax; if(v<-2)v=-2; byte|=(uint8_t)((v+2)<<(k*2)); } + qr[i>>2]=byte; + } + } +} + +#endif /* COLI_QUANT_H */ diff --git a/c/sample.h b/c/sample.h new file mode 100644 index 0000000..abce88c --- /dev/null +++ b/c/sample.h @@ -0,0 +1,143 @@ +/* sample.h — sampling (temperature + nucleus) and stop-set management. + * Header-only: all functions are static — include from the main engine file. */ +#ifndef SAMPLE_H +#define SAMPLE_H + +#include +#include +#include +#include "tok.h" + +/* ---- RNG (xorshift64*) -------------------------------------------------- */ +static uint64_t g_rng = 0x9E3779B97F4A7C15ULL; +static inline double rndu(void){ + g_rng ^= g_rng << 13; g_rng ^= g_rng >> 7; g_rng ^= g_rng << 17; + return (double)(g_rng >> 11) * (1.0 / 9007199254740992.0); +} + +/* ---- argmax over a float vector ----------------------------------------- */ +static inline int argmax_v(const float *lo, int V){ + int b = 0; float bv = lo[0]; + for (int i = 1; i < V; i++) if (lo[i] > bv) { bv = lo[i]; b = i; } + return b; +} + +/* ---- distribution buffers (reused, single-threaded decode) --------------- */ +static float *g_pbuf = NULL; +static int *g_pidx = NULL; + +/* sift-down on max-heap in h[0..n), key = g_pbuf[h[i]] (#335: partial top-p). + * "hole" variant: carries the root value and deposits only at the end, so + * heapify is O(V) and each pop is O(log n) without qsort on the full vocab. */ +static void topp_siftdown(int *h, int n, int i){ + int iv = h[i]; float kv = g_pbuf[iv]; + for (;;) { + int l = 2*i + 1; + if (l >= n) break; + int b = l; if (l+1 < n && g_pbuf[h[l+1]] > g_pbuf[h[l]]) b = l+1; + if (g_pbuf[h[b]] <= kv) break; + h[i] = h[b]; i = b; + } + h[i] = iv; +} + +/* build the target distribution in g_pbuf: softmax(lo/temp) truncated to + * top-p g_nuc. Invariant: g_pbuf stays indexed by token-id (never reordered); + * the truncated tail is zeroed (dist_sample reads by id directly). + * Requires: g_temp, g_nuc, falloc() — declared in the main engine file. */ +static void dist_build(const float *lo, int V){ + if (!g_pbuf) { g_pbuf = falloc(V); g_pidx = malloc(V * sizeof(int)); } + int mxi = -1; float mx = 0; + for (int i = 0; i < V; i++) + if (isfinite(lo[i]) && (mxi < 0 || lo[i] > mx)) { mx = lo[i]; mxi = i; } + double s = 0; float invt = 1.f / (g_temp > 1e-4f ? g_temp : 1e-4f); + if (mxi >= 0) { + for (int i = 0; i < V; i++) { + g_pbuf[i] = isfinite(lo[i]) ? expf((lo[i] - mx) * invt) : 0.f; + s += g_pbuf[i]; + } + } + if (mxi < 0 || !isfinite(s) || s <= 0.0) { + static int warned = 0; + if (!warned) { warned = 1; fprintf(stderr, + "[SAMPLE] warning: non-finite logits (NaN/Inf) — falling back to argmax; " + "output may be degraded. This usually means a numerical blow-up upstream.\n"); } + int a = (mxi >= 0) ? mxi : 0; + for (int i = 0; i < V; i++) g_pbuf[i] = 0.f; + g_pbuf[a] = 1.f; + return; + } + for (int i = 0; i < V; i++) g_pbuf[i] /= (float)s; + if (g_nuc > 0 && g_nuc < 1.f) { + for (int i = 0; i < V; i++) g_pidx[i] = i; + for (int i = V/2-1; i >= 0; i--) topp_siftdown(g_pidx, V, i); + double s2 = 0, cum = 0; int out = V; + do { + int root = g_pidx[0]; + g_pidx[0] = g_pidx[--out]; g_pidx[out] = root; + s2 += g_pbuf[root]; cum += g_pbuf[root]; + if (out > 0) topp_siftdown(g_pidx, out, 0); + } while (cum < g_nuc && out > 0); + for (int i = 0; i < out; i++) g_pbuf[g_pidx[i]] = 0; + float s2f = (float)s2; + for (int i = out; i < V; i++) g_pbuf[g_pidx[i]] /= s2f; + } +} + +/* sample from g_pbuf; ban>=0 excludes that token (renormalizing on the fly) */ +static int dist_sample(int V, int ban){ + double z = 1.0 - (ban >= 0 ? g_pbuf[ban] : 0.0); + if (z <= 1e-12) z = 1e-12; + double u = rndu() * z, cum = 0; + for (int i = 0; i < V; i++) { if (i == ban) continue; cum += g_pbuf[i]; if (cum >= u) return i; } + for (int i = V-1; i >= 0; i--) if (i != ban && g_pbuf[i] > 0) return i; + return 0; +} + +/* next token from logits: greedy if g_temp<=0, sampling otherwise. + * ban = token excluded because it was rejected by speculative verification. */ +static int pick_tok(const float *lo, int V, int ban){ + if (g_temp <= 0) return argmax_v(lo, V); + dist_build(lo, V); + return dist_sample(V, ban); +} + +/* ---- stop set ----------------------------------------------------------- */ +static int g_stop[64], g_nstop = 0; +static inline int is_stop(int t){ + for (int i = 0; i < g_nstop; i++) if (t == g_stop[i]) return 1; + return 0; +} +/* T=NULL -> config stops only (validation/oracle, where the tokenizer is not needed). */ +static void stops_arm_tok(const Cfg *c, int tok_eos, Tok *T){ + g_nstop = 0; + for (int i = 0; i < c->n_stop && g_nstop < 64; i++) g_stop[g_nstop++] = c->stop_ids[i]; + if (tok_eos >= 0 && !is_stop(tok_eos) && g_nstop < 64) g_stop[g_nstop++] = tok_eos; + int nsp = 0; + if (T) for (int id = 0; id < T->n_ids && g_nstop < 64; id++) + if (T->id_special[id] && !is_stop(id)) { g_stop[g_nstop++] = id; nsp++; } + fprintf(stderr, "[stop] %d stop tokens:", g_nstop); + for (int i = 0; i < g_nstop; i++) fprintf(stderr, " %d", g_stop[i]); + if (nsp) fprintf(stderr, " (%d from the tokenizer's special set)", nsp); + fprintf(stderr, "\n"); +} +static void stops_arm(const Cfg *c, int tok_eos){ stops_arm_tok(c, tok_eos, NULL); } + +/* ---- log-prob of a target token given the logit vector ------------------- */ +static double logprob_target(const float *lo, int V, int target, int *am){ + float mx = lo[0]; int best = 0; + for (int i = 1; i < V; i++) if (lo[i] > mx) { mx = lo[i]; best = i; } + double se = 0; + for (int i = 0; i < V; i++) se += exp((double)lo[i] - mx); + if (am) *am = (best == target); + return (double)(lo[target] - mx) - log(se); +} + +/* "glm" in model_type, case-insensitive */ +static int mt_is_glm(const char *s){ + if (s) for (; *s; s++) + if ((s[0]|32) == 'g' && (s[1]|32) == 'l' && (s[2]|32) == 'm') return 1; + return 0; +} + +#endif /* SAMPLE_H */ diff --git a/c/setup.sh b/c/setup.sh index e31e60a..dcb851d 100755 --- a/c/setup.sh +++ b/c/setup.sh @@ -32,11 +32,11 @@ esac # 2) build: nativa (veloce, per QUESTA macchina). Per un binario da distribuire: make portable echo " building (ARCH=${ARCH:-native})…" -make -s glm ARCH="${ARCH:-native}" +make -s colibri ARCH="${ARCH:-native}" # 3) self-test sull'oracolo tiny, se presente if [ -d glm_tiny ] && [ -f ref_glm.json ]; then - r=$(SNAP=./glm_tiny TF=1 ./glm 64 16 16 2>/dev/null | grep -oE "[0-9]+/[0-9]+ positions" || true) + r=$(SNAP=./glm_tiny TF=1 ./colibri 64 16 16 2>/dev/null | grep -oE "[0-9]+/[0-9]+ positions" || true) echo " engine self-test: ${r:-?} (expected 32/32)" fi diff --git a/c/telemetry.h b/c/telemetry.h new file mode 100644 index 0000000..3c3fe85 --- /dev/null +++ b/c/telemetry.h @@ -0,0 +1,189 @@ +/* telemetry.h — dashboard protocol lines, stats/usage persistence, hardware probe. + * Include after Model/Cfg/QT/ESlot/shards and st.h are defined; requires + * qt_bytes(), now_s(), rss_gb(), edisk_s(), and the g_cuda_* globals (ifdef). */ +#ifndef TELEMETRY_H +#define TELEMETRY_H + +static int64_t tbytes(int O,int I,int bits){ + if(bits>=16) return (int64_t)O*I*4; + if(bits>=5) return (int64_t)O*I + (int64_t)O*4; + return (int64_t)O*((I+1)/2) + (int64_t)O*4; +} + +static int64_t expert_bytes_probe(Model *m, int ebits){ + Cfg *c=&m->c; int64_t eb=0; char nm[256]; + snprintf(nm,sizeof(nm),"model.layers.%d.mlp.experts.0.gate_proj.weight",c->first_dense); + if(st_nbytes(&m->S,nm)>0){ + const char *suf[3]={"gate_proj","up_proj","down_proj"}; + for(int k=0;k<3;k++){ + snprintf(nm,sizeof(nm),"model.layers.%d.mlp.experts.0.%s.weight",c->first_dense,suf[k]); + eb+=st_nbytes(&m->S,nm); + snprintf(nm,sizeof(nm),"model.layers.%d.mlp.experts.0.%s.weight.qs",c->first_dense,suf[k]); + int64_t q=st_nbytes(&m->S,nm); if(q>0) eb+=q; + } + } + if(eb<=0) eb = tbytes(c->moe_inter,c->hidden,ebits)*2 + tbytes(c->hidden,c->moe_inter,ebits); + return eb; +} + +/* BRAIN MAP: per-turn expert hit bitmap for the dashboard. */ +static uint8_t **g_ehit; +static void ehit_mark(Model *m, int layer, int eid){ + if(!g_ehit){ Cfg *c=&m->c; + g_ehit=calloc(c->n_layers+1,sizeof(uint8_t*)); + for(int i=0;i<=c->n_layers;i++) g_ehit[i]=calloc(c->n_experts,1); + } + g_ehit[layer][eid]=1; +} + +/* CPU model + cores + RAM (GB); empty/zero where unavailable. */ +static void hw_probe(char *cpu, size_t cn, int *cores, double *ram_total, double *ram_avail){ + cpu[0]=0; +#ifdef _WIN32 +#if defined(__x86_64__) || defined(__i386__) + { unsigned int r[12]={0}; unsigned int *w=r; + for(unsigned int f=0x80000002u; f<=0x80000004u; f++,w+=4) + __get_cpuid(f,&w[0],&w[1],&w[2],&w[3]); + char *b=(char*)r; b[47]=0; while(*b==' ')b++; + snprintf(cpu,cn,"%s",b); } +#endif +#else + FILE *ci=fopen("/proc/cpuinfo","r"); + if(ci){ char ln[256]; + while(fgets(ln,sizeof(ln),ci)) if(!strncmp(ln,"model name",10)){ + char *p=strchr(ln,':'); if(p){ p++; while(*p==' ')p++; + int n=(int)strlen(p); if(n>0&&p[n-1]=='\n')p[--n]=0; + snprintf(cpu,cn,"%s",p); } break; } + fclose(ci); } +#endif + *cores=0; +#ifdef _WIN32 + { SYSTEM_INFO si; GetSystemInfo(&si); *cores=(int)si.dwNumberOfProcessors; } +#elif defined(_SC_NPROCESSORS_ONLN) + *cores=(int)sysconf(_SC_NPROCESSORS_ONLN); +#endif + *ram_total=*ram_avail=0; +#ifdef _WIN32 + compat_meminfo(ram_total,ram_avail); +#else + FILE *mi=fopen("/proc/meminfo","r"); + if(mi){ char ln[256]; double mt=0,ma=0; + while(fgets(ln,sizeof(ln),mi)){ + if(sscanf(ln,"MemTotal: %lf",&mt)==1) *ram_total=mt/1e6; + if(sscanf(ln,"MemAvailable: %lf",&ma)==1) *ram_avail=ma/1e6; + } fclose(mi); } +#endif +} + +static void hwinfo_emit(Model *m){ + Cfg *c=&m->c; (void)c; + char cpu[256]; int cores; double ram_total,ram_avail; + hw_probe(cpu,sizeof(cpu),&cores,&ram_total,&ram_avail); + int ngpu=0; double vram_total=0; + char gpu_name[128]=""; +#ifdef COLI_CUDA + ngpu=g_cuda_ndev; vram_total=m->gpu_expert_bytes/1e9; + for(int i=0;i0) + snprintf(gpu_name,sizeof(gpu_name),"CUDA device x%d",g_cuda_ndev); +#endif + printf("HWINFO %d %.1f %.1f %d %.1f %s|%s\n", + cores,ram_total,ram_avail,ngpu,vram_total,cpu,gpu_name); + fflush(stdout); +} + +static void tiers_emit(Model *m){ + Cfg *c=&m->c; int nsp=0; + for(int i=0;in_layers;i++) if(m->L[i].sparse) nsp++; + int total=(nsp+(m->has_mtp?1:0))*c->n_experts; + int pinned=0,lru=0; + for(int i=0;i<=c->n_layers;i++){ pinned+=m->npin?m->npin[i]:0; lru+=m->ecn?m->ecn[i]:0; } + int vram=0; double vram_gb=0; +#ifdef COLI_CUDA + vram=m->gpu_expert_count; vram_gb=m->gpu_expert_bytes/1e9; +#endif + int ram=pinned-vram+lru; if(ram<0) ram=0; + int disk=total-vram-ram; if(disk<0) disk=0; + double eb=(double)expert_bytes_probe(m,m->ebits); + printf("TIERS %d %d %d %.2f %.2f\n",vram,ram,disk,vram_gb,ram*eb/1e9); + fflush(stdout); +} + +static void emap_emit(Model *m){ + Cfg *c=&m->c; + int rows=0; + for(int i=0;in_layers;i++) if(m->L[i].sparse) rows++; + int has_mtp = m->has_mtp && m->eusage[c->n_layers]; + if(has_mtp) rows++; + int cols=c->n_experts; + char *hex=malloc((size_t)rows*cols*2+1); int w=0; + for(int i=0;i<=c->n_layers;i++){ + int is_row = (in_layers && m->L[i].sparse) || (i==c->n_layers && has_mtp); + if(!is_row) continue; + for(int e=0;epin[i]; + for(int z=0;znpin[i];z++) if(P[z].eid==e){ +#ifdef COLI_CUDA + tier = P[z].g.cuda?2:1; +#else + tier = 1; +#endif + break; } + if(!tier && m->ecache && m->ecache[i]) + for(int z=0;zecn[i];z++) if(m->ecache[i][z].eid==e){ tier=1; break; } + uint32_t u = m->eusage[i]?m->eusage[i][e]:0; + int heat=0; while(u){ heat++; u>>=1; } if(heat>63) heat=63; + int b=(tier<<6)|heat; + hex[w++]="0123456789abcdef"[b>>4]; hex[w++]="0123456789abcdef"[b&15]; + } + } + hex[w]=0; + printf("EMAP %d %d %s\n",rows,cols,hex); fflush(stdout); free(hex); +} + +static void hits_emit(Model *m){ + Cfg *c=&m->c; if(!g_ehit) return; + int rows=0; + for(int i=0;in_layers;i++) if(m->L[i].sparse) rows++; + int has_mtp = m->has_mtp && m->eusage[c->n_layers]; + if(has_mtp) rows++; + int cols=c->n_experts, nb=(rows*cols+7)/8; + uint8_t *bm=calloc(nb,1); int bit=0; + for(int i=0;i<=c->n_layers;i++){ + int is_row = (in_layers && m->L[i].sparse) || (i==c->n_layers && has_mtp); + if(!is_row) continue; + for(int e=0;e>3]|=1<<(bit&7); g_ehit[i][e]=0; } + } + char *hex=malloc((size_t)nb*2+1); int w=0; + for(int b=0;b>4]; hex[w++]="0123456789abcdef"[bm[b]&15]; } + hex[w]=0; + printf("HITS %d %d %s\n",rows,cols,hex); fflush(stdout); free(hex); free(bm); +} + +static void stats_dump_q(Model *m, const char *path, int quiet){ + char tmp[2100]; snprintf(tmp,sizeof(tmp),"%s.tmp",path); + FILE *f=fopen(tmp,"w"); if(!f){ if(!quiet) perror(tmp); return; } + Cfg *c=&m->c; int64_t tot=0, nz=0; + for(int i=0;i<=c->n_layers;i++){ if(!m->eusage[i]) continue; + for(int e=0;en_experts;e++) if(m->eusage[i][e]){ fprintf(f,"%d %d %u\n",i,e,m->eusage[i][e]); tot+=m->eusage[i][e]; nz++; } } + fclose(f); rename(tmp,path); + if(!quiet) fprintf(stderr,"[STATS] %lld selections across %lld distinct experts -> %s\n",(long long)tot,(long long)nz,path); +} +static void stats_dump(Model *m, const char *path){ stats_dump_q(m,path,0); } + +static char g_usage_path[2100]=""; +static int64_t usage_load(Model *m, const char *path){ + FILE *f=fopen(path,"r"); if(!f) return 0; + Cfg *c=&m->c; int l,e; uint32_t cnt; int64_t tot=0; + while(fscanf(f,"%d %d %u",&l,&e,&cnt)==3) + if(l>=0&&l<=c->n_layers&&e>=0&&en_experts&&m->eusage[l]){ m->eusage[l][e]+=cnt; tot+=cnt; } + fclose(f); return tot; +} +static void usage_save(Model *m){ if(g_usage_path[0]) stats_dump_q(m,g_usage_path,1); } + +#endif /* TELEMETRY_H */ diff --git a/c/tests/bench_dsa_select.c b/c/tests/bench_dsa_select.c index a1f2014..676e4c2 100644 --- a/c/tests/bench_dsa_select.c +++ b/c/tests/bench_dsa_select.c @@ -24,7 +24,7 @@ * Run: make tests/bench_dsa_select && ./tests/bench_dsa_select (not in TEST_BINS) */ #define main coli_glm_main_unused -#include "../glm.c" +#include "../colibri.c" #undef main #include diff --git a/c/tests/bench_topp.c b/c/tests/bench_topp.c index 9c0682f..b284308 100644 --- a/c/tests/bench_topp.c +++ b/c/tests/bench_topp.c @@ -22,7 +22,7 @@ * Run: make tests/bench_topp && ./tests/bench_topp (not in TEST_BINS -- not a gate) */ #define main coli_glm_main_unused -#include "../glm.c" +#include "../colibri.c" #undef main #include diff --git a/c/tests/test_dsa_select.c b/c/tests/test_dsa_select.c index 99fecb0..ef3fd7b 100644 --- a/c/tests/test_dsa_select.c +++ b/c/tests/test_dsa_select.c @@ -31,7 +31,7 @@ * In-memory only (no scratch files), so it builds clean on the Windows MinGW CI * job without the unmerged compat shim. */ #define main coli_glm_main_unused -#include "../glm.c" +#include "../colibri.c" #undef main #include diff --git a/c/tests/test_i4_grouped.c b/c/tests/test_i4_grouped.c index 2a9521e..04e8d88 100644 --- a/c/tests/test_i4_grouped.c +++ b/c/tests/test_i4_grouped.c @@ -18,7 +18,7 @@ * index, a wrong group boundary or a swapped nibble cannot hide under it — * those are O(1) relative errors, not O(1e-6). */ #define main coli_glm_main_unused -#include "../glm.c" +#include "../colibri.c" #undef main static uint32_t rng_state=0xC0FFEEu; diff --git a/c/tests/test_idot.c b/c/tests/test_idot.c index b7109d4..343535c 100644 --- a/c/tests/test_idot.c +++ b/c/tests/test_idot.c @@ -7,7 +7,7 @@ * (sign-trick kernels must treat |−128| as 128 unsigned, not saturate to 127), * and random data at qrow_i8's contract (|x| <= 127, w full int8 range). */ #define main coli_glm_main_unused -#include "../glm.c" +#include "../colibri.c" #undef main static uint32_t rng_state=0x12345678u; diff --git a/c/tests/test_kv_alloc.c b/c/tests/test_kv_alloc.c index 40ec00d..d080c08 100644 --- a/c/tests/test_kv_alloc.c +++ b/c/tests/test_kv_alloc.c @@ -4,7 +4,7 @@ * arrays twice on the second call -> allocator abort. No model file needed: * the CPU path of kv_alloc only reads c->n_layers/kv_lora/qk_rope. */ #define main coli_glm_main_unused -#include "../glm.c" +#include "../colibri.c" #undef main int main(void){ diff --git a/c/tests/test_makefile_platform.py b/c/tests/test_makefile_platform.py index 3c72880..50b4c7a 100644 --- a/c/tests/test_makefile_platform.py +++ b/c/tests/test_makefile_platform.py @@ -35,7 +35,7 @@ class MakefilePlatformTests(unittest.TestCase): env["PATH"] = "" result = subprocess.run( - [MAKE, "--no-print-directory", "-B", "-n", "glm"], + [MAKE, "--no-print-directory", "-B", "-n", "colibri"], cwd=C_DIR, env=env, text=True, @@ -43,7 +43,7 @@ class MakefilePlatformTests(unittest.TestCase): check=True, ) - self.assertIn("-o glm.exe", result.stdout) + self.assertIn("-o colibri.exe", result.stdout) self.assertIn("-fopenmp", result.stdout) self.assertIn("-static", result.stdout) diff --git a/c/tests/test_sample_nan.c b/c/tests/test_sample_nan.c index fc33268..4f31def 100644 --- a/c/tests/test_sample_nan.c +++ b/c/tests/test_sample_nan.c @@ -6,7 +6,7 @@ * iniettato il token scelto e' l'argmax dei FINITI (mai 0 per default), su ogni posizione * del NaN inclusa lo[0]; (c) nessun NaN sopravvive in g_pbuf. */ #define main coli_glm_main_unused -#include "../glm.c" +#include "../colibri.c" #undef main #include diff --git a/c/tests/test_stops.c b/c/tests/test_stops.c index 00f922b..5e7e9d2 100644 --- a/c/tests/test_stops.c +++ b/c/tests/test_stops.c @@ -20,7 +20,7 @@ * Defense 2 is what makes this robust against checkpoints we don't control: * even with BOTH configs mutilated, a control token cannot leak into a reply. */ #define main coli_glm_main_unused -#include "../glm.c" +#include "../colibri.c" #undef main static const char *TOKJSON = diff --git a/c/tests/test_topp.c b/c/tests/test_topp.c index 9276507..49f45d6 100644 --- a/c/tests/test_topp.c +++ b/c/tests/test_topp.c @@ -24,7 +24,7 @@ * No scratch files: the test runs entirely in memory (no mkdtemp), so it builds clean on * the Windows MinGW CI job without the unmerged compat shim (#352). */ #define main coli_glm_main_unused -#include "../glm.c" +#include "../colibri.c" #undef main #include diff --git a/c/tests/test_uring.c b/c/tests/test_uring.c index e64b9f2..625c79b 100644 --- a/c/tests/test_uring.c +++ b/c/tests/test_uring.c @@ -6,7 +6,7 @@ #include #include #define main coli_glm_main_unused -#include "../glm.c" +#include "../colibri.c" #undef main static int fail(const char *s){ fprintf(stderr,"FAIL: %s\n",s); return 1; } From 420a0720c31808753d03784de56a5fcd1a51a61d Mon Sep 17 00:00:00 2001 From: ZacharyZcR Date: Sun, 19 Jul 2026 04:02:00 +0800 Subject: [PATCH 18/22] docs: add simplified Chinese and Italian README, update language nav MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit New files: README.zh-CN.md — simplified Chinese (大陆用词) README.it.md — Italian (the project's "mother tongue") All four READMEs now link to each other in a consistent nav bar. Updated zh-TW to reflect glm.c → colibri.c rename and new headers. --- README.it.md | 260 ++++++++++++++++++++++++++++++++++++++++++++++++ README.md | 2 +- README.zh-CN.md | 237 +++++++++++++++++++++++++++++++++++++++++++ README.zh-TW.md | 12 ++- 4 files changed, 506 insertions(+), 5 deletions(-) create mode 100644 README.it.md create mode 100644 README.zh-CN.md diff --git a/README.it.md b/README.it.md new file mode 100644 index 0000000..76e3c21 --- /dev/null +++ b/README.it.md @@ -0,0 +1,260 @@ +

+ colibrì — motore piccolo, modello immenso +

+ +

+ English · 简体中文 · 繁體中文 · Italiano +

+ +**Motore piccolo, modello immenso.** Esegui **GLM-5.2 (744 miliardi di parametri, MoE)** su un computer consumer con ~25 GB di RAM — in C puro, zero dipendenze, caricando gli expert dal disco in streaming. + +Colibrì è un runtime MoE leggero e che preserva la qualità: tratta VRAM, RAM e +disco come un'unica gerarchia di memoria gestita. Se la memoria veloce non basta +il modello rallenta, ma la policy predefinita **non cambia mai silenziosamente la +precisione del modello né la semantica del router**. + +``` +$ ./coli chat + 🐦 colibrì v1.0 — GLM-5.2 · 744B MoE · int4 · streaming CPU + ✓ ready in 32s · resident 9.9 GB + › ciao! + ◆ Ciao! 😊 Come posso aiutarti oggi? +``` + +## Guardalo in azione + +

+ dashboard web di colibrì — metriche live, pannello hardware, livelli degli expert +

+

La dashboard web (./coli web): un modello da 744B a 4 tok/s, TTFT 1.6 s, disco 0 — +residenza completa degli expert su 6× RTX 5090, con metriche token in tempo reale, breakdown dei tempi per turno, +la barra dei livelli VRAM/RAM/disco e il mini-cervello live nell'angolo.

+ +

+ la pagina Brain — 19.456 expert come una corteccia vivente +

+

La pagina Brain: tutti i 19.456 expert come una corteccia vivente — il colore indica +il livello di archiviazione, la luminosità il calore di routing, e ogni expert instradato in un turno +lampeggia bianco. Passando il cursore si vede l'affinità +tematica misurata dell'expert.

+ +

+ la pagina Atlas — l'atlante misurato degli expert come una galassia 3D +

+

La pagina Atlas: l'atlante +misurato degli expert come una galassia 3D — 13.260 expert caratterizzati, 1.041 specialisti +replicabili che si raggruppano per argomento (poesia, legge, cinese, SQL…). La posizione deriva +dall'affinità di routing misurata, non da un embedding appreso. Trascinare per ruotare.

+ +## L'idea + +Un modello Mixture-of-Experts da 744B attiva solo ~40B parametri per token — e +solo ~11 GB di quelli cambiano da un token all'altro (gli expert instradati): + +

+ solo ~5.4% dei parametri è attivo per token +

+ +Il modello non ha bisogno di *stare* in memoria veloce — ha bisogno di essere +**piazzato**: + +- la **parte densa** (attenzione, expert condivisi, embedding — ~17B parametri) + resta **residente in RAM a int4** (~9.9 GB); +- i **19.456 expert instradati** (75 layer MoE × 256 + la testa MTP, ~19 MB + ciascuno a int4) stanno **su disco** (~370 GB) e vengono **caricati on demand + in streaming**, con una cache LRU per layer, un hot-store pinnato che impara, + e un livello VRAM opzionale. + +Il motore è un singolo file C (`c/colibri.c`) più header piccoli. Niente BLAS, +niente Python a runtime, niente GPU obbligatoria. + +## Come funziona + +### Il percorso di ogni token + +

+ instrada → unione → piazza → sovrapponi → impara +

+ +Ogni layer di ogni token percorre gli stessi cinque passi. L'obiettivo +progettuale è che **il piazzamento decide solo la velocità** — le decisioni +del router e la precisione dei pesi sono identiche sia che un expert risponda +dalla VRAM sia dal disco. + +### Una gerarchia di memoria, non un requisito di memoria + +

+ residenza expert a tre livelli: VRAM / RAM / NVMe +

+ +Lo stesso motore copre l'intero spettro: su un portatile da 25 GB tutto viene +caricato dal disco in streaming (lento, ma corretto); su un host grande l'intero +set di expert diventa residente (`CUDA_EXPERT_GB=auto PIN_GB=all`) e il disco +esce completamente dal percorso di decode. Tra i livelli c'è una **cache che +impara**: il motore registra quali expert il *tuo* carico di lavoro instrada +(`.coli_usage`, aggiornato a ogni turno) e fissa automaticamente i più caldi — +colibrì diventa letteralmente più veloce man mano che lo usi. Sugli host +multi-socket, `COLI_NUMA=1` interlaccia i pesi residenti tra i controller di +memoria ([#82](https://github.com/JustVugg/colibri/issues/82)). + +### Mai aspettare il disco due volte + +I miss nella cache costano caro, quindi il motore investe la maggior parte +della sua astuzia per evitarli e sovrapporli: le tre matrici di ogni expert sono +memorizzate contigue e lette con un unico `pread`; un pool I/O asincrono +limitato (`PIPE=1`, attivo per default) carica gli expert mancanti mentre quelli +residenti calcolano; le posizioni in batch leggono ogni expert unico una sola +volta (**batch-union**); un thread di lookahead del router (`PILOT=1`) fa il +prefetch degli expert del layer successivo — il routing è misurabilmente +**prevedibile al 71.6% un layer in anticipo**. Sulle GPU, la pipeline residente +(`COLI_CUDA_PIPE=2`) mantiene il flusso residuo on-device tra i layer, così il +loop CPU degli expert procede senza interruzioni; su Apple Silicon un backend +[Metal](docs/metal.md) sperimentale esegue la matmul batch degli expert sulla +GPU a memoria unificata. + +### Modello fedele, stato compresso + +Il forward pass è validato **token-esatto contro un oracle `transformers`** +(teacher-forcing 32/32). L'attenzione MLA memorizza uno stato KV compresso — 576 +float/token invece di 32.768 (**57× più piccolo**) — e lo persiste tra i +riavvii (`.coli_kv`): le conversazioni riaprono "calde", senza alcun re-prefill, +byte-identiche a una sessione ininterrotta. L'attenzione sparsa DSA (il +lightning indexer di GLM-5.2) è implementata fedelmente e validata forzando la +selezione di tutte le chiavi per riprodurre esattamente l'attenzione densa. + +### Decodifica speculativa, onestamente + +La testa MTP nativa di GLM-5.2 propone token che il modello principale verifica +in un unico forward batch — 2.2–2.8 token/forward quando conviene. Due regole +conquistate a caro prezzo sono i default: la testa MTP deve essere **int8** (le +teste int4 crollano al 0–4% di accettazione, +[#8](https://github.com/JustVugg/colibri/issues/8)), e draft e verifica devono +calcolare **la stessa funzione** — `SPEC_PIN=1` fissa entrambi sulla stessa +famiglia di kernel ([#163](https://github.com/JustVugg/colibri/issues/163) +contiene l'intera indagine forense). I draft forzati da grammatica +([`GRAMMAR=file.gbnf`](docs/grammar-draft.md)) aggiungono accettazione quasi +gratuita sull'output JSON vincolato. Se la speculazione conviene dipende dalla +temperatura della cache — misura, e usa `DRAFT=0` quando non paga. + +## Cosa ottiene + +

+ velocità di decode misurata per classe hardware +

+ +Stesso motore, stesso container int4 — cambia solo dove risiedono gli expert. +Punti salienti dalle [tabelle benchmark complete](docs/benchmarks.md): + +- **6× RTX 5090, residenza completa:** 5.8–6.8 tok/s in decode, TTFT ~13 s + ([log dell'esperimento](docs/experiments/glm52-6x5090-2026-07-12.md)); +- **desktop solo-CPU da 128 GB:** ~1.8 tok/s a cache calda + ([#200](https://github.com/JustVugg/colibri/issues/200)); +- **singola RTX 5070 Ti, classe laptop:** 1.07 tok/s tramite la pipeline + GPU-residente ([#273](https://github.com/JustVugg/colibri/issues/273)); +- **macchina di sviluppo da 25 GB:** 0.05–0.1 tok/s a freddo — il punto di + partenza dimostrato da cui è nato il progetto, e ancora oggi la baseline onesta. + +La qualità è misurata, non presunta: il costo di quantizzazione del container +int4 e le ablazioni su granularità delle scale e rotazione sono in +[docs/benchmarks.md](docs/benchmarks.md#quality-benchmark) e +[#108](https://github.com/JustVugg/colibri/issues/108)/[#81](https://github.com/JustVugg/colibri/issues/81). + +## Per iniziare + +### 1. Scarica il modello + +Un container **GLM-5.2 int4** pre-convertito è su Hugging Face — **usa la +versione con le teste MTP int8**: + +**https://huggingface.co/mateogrgic/GLM-5.2-colibri-int4-with-int8-mtp** + +> ⚠️ Il mirror originale contiene teste MTP int4 → accettazione dei draft allo 0% +> ([#8](https://github.com/JustVugg/colibri/issues/8)). Verifica la tua versione: +> `ls -l /out-mtp-*` — int8 (corretto) è `3527131672 / 5366238584 / 1065950496`. + +Oppure converti tu stesso dalla sorgente FP8 — un unico comando riprendibile che +non richiede mai i 756 GB completi su disco contemporaneamente: + +```bash +cd c && ./setup.sh # verifica gcc/OpenMP, compila, autotest +./coli convert --model /nvme/glm52_i4 # scarica e converti shard per shard (python, una tantum) +``` + +### 2. Esegui + +```bash +COLI_MODEL=/nvme/glm52_i4 ./coli chat # budget RAM, cache e MTP rilevati automaticamente +COLI_MODEL=/nvme/glm52_i4 ./coli plan # mostra il piazzamento pianificato VRAM/RAM/disco +COLI_MODEL=/nvme/glm52_i4 ./coli doctor # controllo di idoneità (sola lettura) +./coli web --model /nvme/glm52_i4 # API + dashboard web sulla stessa porta +./coli serve --model /nvme/glm52_i4 # solo API compatibile OpenAI +``` + +Il motore a runtime è puro C — python si usa solo per il convertitore (una tantum) +e per il gateway API opzionale. + +### 3. Approfondisci + +| argomento | documento | +|---|---| +| Benchmark, dati dalla comunità, misurazioni di qualità | [docs/benchmarks.md](docs/benchmarks.md) | +| Parametri di tuning, policy, cache che impara, prefetch | [docs/tuning.md](docs/tuning.md) | +| Build nativa su Windows 11 (con CUDA DLL) | [docs/windows.md](docs/windows.md) | +| Backend CUDA, livello expert in VRAM, residenza completa | [docs/cuda.md](docs/cuda.md) | +| Backend Metal per Apple Silicon | [docs/metal.md](docs/metal.md) | +| API compatibile OpenAI, KV slot, dashboard web | [docs/api.md](docs/api.md) | +| Draft forzati da grammatica (output strutturato) | [docs/grammar-draft.md](docs/grammar-draft.md) | +| Inventario delle variabili d'ambiente | [docs/ENVIRONMENT.md](docs/ENVIRONMENT.md) | + +## Sostenere il progetto + +colibrì è nato come progetto di una sola persona su un portatile con 12 core +e 25 GB di RAM; oggi i suoi numeri arrivano da una comunità di macchine reali. +Se ti è utile: + +- ⭐ metti una stella al repository e condividilo; +- 🐛 apri issue con i numeri di benchmark del tuo hardware — i datapoint + fanno avanzare questo progetto più di qualsiasi altra cosa; +- 💬 contattaci via GitHub issues per sponsorizzare lo sviluppo o donare hardware. + +## Struttura del repository + +``` +Makefile punto d'ingresso root per build/check +c/ +├── colibri.c motore principale +├── quant.h kernel matmul quantizzati (SIMD multi-architettura) +├── sample.h campionamento, RNG, set di stop +├── kv_persist.h persistenza KV su disco (.coli_kv) +├── telemetry.h protocollo dashboard, statistiche, usage +├── st.h, tok.h, json.h header di runtime +├── backend_cuda.* livello CUDA opzionale +├── Makefile build e check locali +├── coli CLI utente +├── openai_server.py gateway HTTP compatibile OpenAI +├── setup.sh setup locale in un solo comando +├── tools/ conversione offline, fixture e benchmark +├── scripts/ helper per conversioni lunghe +└── tests/ test C e Python senza dipendenze +web/ UI browser (puro client API OpenAI) +desktop/ shell desktop Tauri v2 che racchiude la web UI +docs/ documentazione di riferimento, esperimenti, media +``` + +Il percorso a runtime resta intenzionalmente piatto e leggibile: `colibri.c` +più i suoi header. Dalla radice del repository, `make`, `make check` e +`make clean` delegano al Makefile del motore. + +## Perché "colibrì" + +Il colibrì pesa pochi grammi, sta sospeso nel vuoto e visita un migliaio di +fiori al giorno. Questo motore tiene in vita un gigante da 744 miliardi di +parametri con le razioni di un colibrì: 25 GB di RAM, dodici core CPU e +tanta pazienza col disco. + +Il nome è rimasto in italiano perché questa è la lingua in cui è stato scritto +il primo prototipo — i commenti nel codice lo testimoniano ancora. + +## Licenza + +Apache 2.0. I pesi di GLM-5.2 sono rilasciati da Z.ai sotto licenza MIT. diff --git a/README.md b/README.md index f9cf3f0..d6f9bcb 100644 --- a/README.md +++ b/README.md @@ -3,7 +3,7 @@

- English · 繁體中文 + English · 简体中文 · 繁體中文 · Italiano

**Tiny engine, immense model.** Run **GLM-5.2 (744B-parameter MoE)** on a consumer machine with ~25 GB of RAM — in pure C, with zero dependencies, by streaming experts from disk. diff --git a/README.zh-CN.md b/README.zh-CN.md new file mode 100644 index 0000000..352de33 --- /dev/null +++ b/README.zh-CN.md @@ -0,0 +1,237 @@ +

+ colibrì——小巧引擎，庞大模型 +

+ +

+ English · 简体中文 · 繁體中文 · Italiano +

+ +**小巧引擎，庞大模型。**只需约 25 GB 内存，就能在消费级电脑上运行 **GLM-5.2（744B 参数的 MoE）**——以零依赖的纯 C 实现，从磁盘流式加载专家。 + +Colibrì 是一套轻量、保持模型质量的 MoE 运行时，将 VRAM、RAM +与存储设备视为统一管理的内存层级。高速内存不足可能降低速度， +但默认策略**绝不会在未告知的情况下改变模型精度或路由语义**。 + +``` +$ ./coli chat + 🐦 colibrì v1.0 — GLM-5.2 · 744B MoE · int4 · streaming CPU + ✓ ready in 32s · resident 9.9 GB + › ciao! + ◆ Ciao! 😊 Come posso aiutarti oggi? +``` + +## 实际运行效果 + +

+ colibrì 网页仪表盘——实时指标、硬件面板与专家存储层级 +

+

网页仪表盘（./coli web）：744B 模型达到 4 tok/s、TTFT 1.6 秒、磁盘读取 0—— +在 6× RTX 5090 上让所有专家常驻，并实时显示 token 指标、每轮耗时明细、 +VRAM／RAM／磁盘层级条，以及角落的实时迷你大脑。

+ +

+ 大脑页面——以实时皮层呈现 19,456 个专家 +

+

大脑（Brain）页面：将全部 19,456 个专家呈现为活的皮层——颜色代表存储层级， +亮度代表路由热度，每轮被路由到的专家都会闪白。将光标停在专家上，即可查看其 +实测主题亲和度。

+ +

+ 图谱页面——以 3D 星系呈现实测专家图谱 +

+

图谱（Atlas）页面：将实测专家图谱 +呈现为 3D 星系——共 13,260 个已分析专家，其中 1,041 个可复现的专门专家会按主题聚集 +（诗歌、法律、中文、SQL……）。位置取自实测路由亲和度，而非学习出的嵌入向量。拖拽即可旋转。

+ +## 核心概念 + +744B 的专家混合（Mixture-of-Experts）模型，每个 token 只会激活约 40B 参数—— +其中每个 token 之间会变动的只有约 11 GB（被路由到的专家）： + +

+ 每个 token 只会激活约 5.4% 的参数 +

+ +所以模型不必完整**装进**高速内存，而是需要正确**放置**： + +- **稠密部分**（注意力、共享专家、嵌入——约 17B 参数）以 int4 + **常驻 RAM**（约 9.9 GB）； +- **19,456 个路由专家**（75 个 MoE 层 × 256，加上 MTP head；每个在 int4 下约 19 MB） + **存放在磁盘**（约 370 GB），并**按需流式加载**，配合逐层 LRU 缓存、 + 会学习的热门专家固定存储区，以及可选的 VRAM 层级。 + +引擎是一个 C 主文件（`c/colibri.c`）加上若干头文件。不需要 BLAS， +运行时不需要 Python，也不需要 GPU。 + +## 工作原理 + +### 每个 token 的处理路径 + +

+ 路由 → 并集 → 放置 → 重叠执行 → 学习 +

+ +每个 token 的每一层都会经过相同的五个步骤。设计目标是让 +**放置只决定速度**——无论专家是从 VRAM 还是磁盘响应，路由器的决策与权重精度都完全相同。 + +### 统一内存层级，取代单一内存门槛 + +

+ VRAM／RAM／NVMe 三层专家常驻架构 +

+ +同一套引擎覆盖完整硬件范围：在 25 GB 笔记本上，一切都从磁盘流式加载 +（慢，但结果正确）；在大内存主机上，则可让整组专家常驻 +（`CUDA_EXPERT_GB=auto PIN_GB=all`），让磁盘完全退出解码路径。 +两端之间有一层**学习型缓存**：引擎会记录*你的*工作负载路由到哪些专家 +（`.coli_usage`，每轮更新），并自动固定最热门的专家——colibrì 确实会越用越快。 +在多路主机上，`COLI_NUMA=1` 会将常驻权重交错分配到各内存控制器 +（[#82](https://github.com/JustVugg/colibri/issues/82)）。 + +### 绝不为同一次磁盘读取等待两遍 + +缓存未命中的代价很高，因此引擎大部分的巧思都用来避免或重叠这些读取： +每个专家的三个矩阵相邻存储，并以一次 `pread` 读取；有界异步 I/O 池 +（`PIPE=1`，默认启用）会在常驻专家计算时加载缺失的专家；批量位置只读取每个 +不重复专家一次（**批量并集**）；路由前瞻线程（`PILOT=1`）则预取下一层专家—— +实测显示，路由结果提前一层时有 **71.6% 的可预测性**。 +在 GPU 上，常驻管线（`COLI_CUDA_PIPE=2`）让残差流跨层保留在设备端， +使 CPU 专家循环不中断；在 Apple Silicon 上，实验性的 +[Metal 后端](docs/metal.md)会用统一内存 GPU 执行批量专家运算。 + +### 忠实模型，压缩状态 + +前向传播已通过 `transformers` oracle 验证为**逐 token 完全一致** +（teacher-forcing 32/32）。MLA 注意力存储压缩后的 KV 状态——每个 token 为 576 个 +浮点数，而非 32,768 个（**缩小 57×**）——并跨重启持久保存 +（`.coli_kv`）：对话可暖启恢复，不需重新 prefill，结果与不中断的会话 +逐字节相同。DSA 稀疏注意力（GLM-5.2 的 lightning indexer）已忠实实现， +并通过强制选取所有 key，验证可精确复现稠密注意力。 + +### 诚实的推测解码 + +GLM-5.2 原生 MTP head 会起草 token，再由主模型以一次批量前向传播验证—— +条件合适时每次 forward 可产生 2.2–2.8 个 token。两条来之不易的规则已成为默认值： +MTP head 必须是 **int8**（int4 head 的接受率会崩塌到 0–4%，见 +[#8](https://github.com/JustVugg/colibri/issues/8)），且草稿与验证必须计算 +**相同函数**——`SPEC_PIN=1` 会把两者固定在同一 kernel family +（完整取证过程见 [#163](https://github.com/JustVugg/colibri/issues/163)）。 +语法强制草稿（[`GRAMMAR=file.gbnf`](docs/grammar-draft.md)）可在受限 JSON 输出中， +以近乎免费的代价提高接受率。推测解码是否带来净收益取决于缓存热度——请实测， +若不划算就使用 `DRAFT=0`。 + +## 实际成果 + +

+ 各硬件级别的实测解码速度 +

+ +同一套引擎、同一个 int4 容器——硬件只会改变专家的存放位置。 +[完整 benchmark 表格](docs/benchmarks.md)中的重点如下： + +- **6× RTX 5090，全部常驻：**解码 5.8–6.8 tok/s，TTFT 约 13 秒 + （[实验记录](docs/experiments/glm52-6x5090-2026-07-12.md)）； +- **128 GB、仅使用 CPU 的台式机：**热缓存后约 1.8 tok/s + （[#200](https://github.com/JustVugg/colibri/issues/200)）； +- **单张 RTX 5070 Ti 的笔记本级主机：**通过 GPU 常驻管线达到 1.07 tok/s + （[#273](https://github.com/JustVugg/colibri/issues/273)）； +- **25 GB 开发机：**冷启动 0.05–0.1 tok/s——这是项目起步时已证实的下限， + 也仍是诚实的基准。 + +质量来自测量，而非假设：int4 容器的量化损失，以及 scale granularity／rotation +消融实验，收录于 [docs/benchmarks.md](docs/benchmarks.md#quality-benchmark)、 +[#108](https://github.com/JustVugg/colibri/issues/108) 与 +[#81](https://github.com/JustVugg/colibri/issues/81)。 + +## 开始使用 + +### 1. 获取模型 + +Hugging Face 上已有预转换的 **GLM-5.2 int4** 容器——请务必使用 +**含 int8 MTP head 的版本**： + +**https://huggingface.co/mateogrgic/GLM-5.2-colibri-int4-with-int8-mtp** + +> ⚠️ 原始镜像使用 int4 MTP head → 草稿接受率为 0% +>（[#8](https://github.com/JustVugg/colibri/issues/8)）。请检查你的版本： +> `ls -l /out-mtp-*`——正确的 int8 大小为 `3527131672 / 5366238584 / 1065950496`。 + +你也可以自行从 FP8 源转换——只需一条可断点续传的命令，且任何时候都不需要 +在磁盘上同时存放完整的 756 GB： + +```bash +cd c && ./setup.sh # 检查 gcc/OpenMP、构建并运行自测 +./coli convert --model /nvme/glm52_i4 # 逐 shard 下载并转换（仅此一次需要 python） +``` + +### 2. 运行 + +```bash +COLI_MODEL=/nvme/glm52_i4 ./coli chat # 自动检测 RAM 预算、缓存与 MTP +COLI_MODEL=/nvme/glm52_i4 ./coli plan # 查看规划的 VRAM／RAM／磁盘配置 +COLI_MODEL=/nvme/glm52_i4 ./coli doctor # 只读就绪检查 +./coli web --model /nvme/glm52_i4 # 在同一端口提供 API 与网页仪表盘 +./coli serve --model /nvme/glm52_i4 # 仅提供 OpenAI 兼容 API +``` + +引擎运行时是纯 C——python 只供一次性转换工具与可选的 API gateway 使用。 + +### 3. 深入了解 + +| 主题 | 文档 | +|---|---| +| Benchmark、社区实测数据、质量测量 | [docs/benchmarks.md](docs/benchmarks.md) | +| 调优选项、策略、学习型缓存、预取 | [docs/tuning.md](docs/tuning.md) | +| Windows 11 原生构建（含 CUDA DLL） | [docs/windows.md](docs/windows.md) | +| CUDA 后端、VRAM 专家层级、全部常驻 | [docs/cuda.md](docs/cuda.md) | +| Apple Silicon Metal 后端 | [docs/metal.md](docs/metal.md) | +| OpenAI 兼容 API、KV slots、网页仪表盘 | [docs/api.md](docs/api.md) | +| 语法强制草稿（结构化输出） | [docs/grammar-draft.md](docs/grammar-draft.md) | +| 环境变量完整清单 | [docs/ENVIRONMENT.md](docs/ENVIRONMENT.md) | + +## 支持项目 + +colibrì 最初由一人使用 12 核心、25 GB RAM 的笔记本开发； +如今它的数据来自社区中各种真实机器。如果这个项目对你有用： + +- ⭐ 为仓库加星并分享； +- 🐛 以 issue 提交你的硬件 benchmark 数据——实测数据比任何其他事都更能推动项目； +- 💬 若想赞助开发或捐赠硬件，请通过 GitHub issues 联系。 + +## 仓库结构 + +``` +Makefile 根目录构建／检查入口 +c/ +├── colibri.c 引擎主文件 +├── quant.h 量化 matmul 内核（SIMD 多架构） +├── sample.h 采样与 stop-set 管理 +├── kv_persist.h .coli_kv 磁盘持久化 +├── telemetry.h 仪表盘协议、统计与用量持久化 +├── st.h, tok.h, json.h 运行时头文件 +├── backend_cuda.* 可选的 CUDA 层级 +├── Makefile 构建与本地检查 +├── coli 用户界面 CLI +├── openai_server.py OpenAI 兼容 HTTP gateway +├── setup.sh 一条命令完成本地设置 +├── tools/ 离线转换、fixtures 与 benchmarks +├── scripts/ 长时间转换辅助工具 +└── tests/ 零依赖的 C 与 Python 测试 +web/ 浏览器 UI（纯 OpenAI API client） +desktop/ 封装网页 UI 的 Tauri v2 桌面 shell +docs/ 参考文档、实验与媒体文件 +``` + +运行时路径刻意保持扁平、易读：`colibri.c` 加上若干头文件。 +在仓库根目录执行 `make`、`make check` 与 `make clean`， +都会转发给引擎的 Makefile。 + +## 为什么叫"colibrì" + +蜂鸟只有几克重，能在原地悬停，并在一天内造访上千朵花。 +这套引擎只用蜂鸟般的配给，就能让 744B 参数的巨人运转： +25 GB RAM、十二个 CPU 核心，以及对磁盘的大量耐心。 + +## 许可证 + +Apache 2.0。GLM-5.2 权重由 Z.ai 以 MIT 许可发布。 diff --git a/README.zh-TW.md b/README.zh-TW.md index 1662617..2576eba 100644 --- a/README.zh-TW.md +++ b/README.zh-TW.md @@ -3,7 +3,7 @@

- English · 繁體中文 + English · 简体中文 · 繁體中文 · Italiano

**小巧引擎，龐大模型。**只要約 25 GB 記憶體，就能在消費級電腦上執行 **GLM-5.2（744B 參數的 MoE）**——以零相依套件的純 C 實作，從硬碟串流載入專家。 @@ -60,7 +60,7 @@ VRAM／RAM／硬碟層級長條，以及角落的即時迷你大腦。

**存放在硬碟**（約 370 GB），並**隨需串流載入**，搭配逐層 LRU 快取、 會學習的熱門專家固定儲存區，以及選用的 VRAM 層級。 -引擎由單一 C 檔（`c/glm.c`）與少量標頭檔組成。不需要 BLAS， +引擎由主 C 檔（`c/colibri.c`）與多個標頭檔模組組成。不需要 BLAS， 執行階段不需要 Python，也不需要 GPU。 ## 運作方式 @@ -203,7 +203,11 @@ colibrì 最初是由一人使用 12 核心、25 GB RAM 的筆電開發； ``` Makefile 根目錄建置／檢查入口 c/ -├── glm.c 單檔 GLM 引擎 +├── colibri.c GLM 引擎主檔 +├── quant.h 量化 matmul kernel +├── sample.h 取樣與 stop-set +├── kv_persist.h .coli_kv 磁碟持久化 +├── telemetry.h 儀表板協定、統計 ├── st.h, tok.h, json.h 執行階段標頭檔 ├── backend_cuda.* 選用的 CUDA 層級 ├── Makefile 建置與本機檢查 @@ -218,7 +222,7 @@ desktop/ 包裝網頁 UI 的 Tauri v2 桌面 shell docs/ 參考文件、實驗與媒體檔 ``` -執行階段路徑刻意維持扁平、易讀：`glm.c` 加上少量標頭檔。 +執行階段路徑刻意維持扁平、易讀：`colibri.c` 加上模組化標頭檔。 在儲存庫根目錄執行 `make`、`make check` 與 `make clean`， 都會轉交給引擎的 Makefile。 From e486574442724ff4ba79808a1ede254368b0638a Mon Sep 17 00:00:00 2001 From: ZacharyZcR Date: Sun, 19 Jul 2026 04:22:46 +0800 Subject: [PATCH 19/22] =?UTF-8?q?web:=20i18n=20support=20=E2=80=94=20Engli?= =?UTF-8?q?sh,=20=E7=AE=80=E4=BD=93=E4=B8=AD=E6=96=87,=20=E7=B9=81?= =?UTF-8?q?=E9=AB=94=E4=B8=AD=E6=96=87,=20Italiano?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Lightweight i18n without react-i18next: a LocaleProvider context + useLocale() hook with {{var}} interpolation, auto-detect from navigator.language, persisted to localStorage. 98 translation keys across 4 locale files (en/zh-CN/zh-TW/it). All user-visible strings in App/Brain/Profiling/ErrorBoundary are now t() calls. Language switcher added to the sidebar footer. Build clean (tsc + vite), 18 tests pass. --- web/src/App.tsx | 121 +++++++++++++++++++----------------- web/src/Brain.tsx | 43 +++++++------ web/src/ErrorBoundary.tsx | 21 +++++-- web/src/Profiling.tsx | 58 ++++++++---------- web/src/i18n/en.ts | 125 ++++++++++++++++++++++++++++++++++++++ web/src/i18n/index.ts | 78 ++++++++++++++++++++++++ web/src/i18n/it.ts | 113 ++++++++++++++++++++++++++++++++++ web/src/i18n/zh-CN.ts | 113 ++++++++++++++++++++++++++++++++++ web/src/i18n/zh-TW.ts | 113 ++++++++++++++++++++++++++++++++++ web/src/index.css | 4 +- web/src/main.tsx | 5 +- 11 files changed, 675 insertions(+), 119 deletions(-) create mode 100644 web/src/i18n/en.ts create mode 100644 web/src/i18n/index.ts create mode 100644 web/src/i18n/it.ts create mode 100644 web/src/i18n/zh-CN.ts create mode 100644 web/src/i18n/zh-TW.ts diff --git a/web/src/App.tsx b/web/src/App.tsx index a33cc88..39c9f3b 100644 --- a/web/src/App.tsx +++ b/web/src/App.tsx @@ -9,6 +9,7 @@ import { Database, Feather, Gauge, + Globe, HardDrive, KeyRound, Layers, @@ -34,6 +35,7 @@ import { Brain } from "./Brain" import { Profiling } from "./Profiling" import { persistPublicSettings, stored } from "@/lib/storage" import { cn } from "@/lib/utils" +import { useLocale } from "./i18n" const message = (role: ChatMessage["role"], content: string): ChatMessage => { let id: string @@ -42,15 +44,12 @@ const message = (role: ChatMessage["role"], content: string): ChatMessage => { } export default function App() { - // When the page is served by the engine itself (coli web), same-origin is the - // right default: no CORS, no manual endpoint editing. The Vite dev server - // (port 5173) keeps the classic default. + const { t, locale, setLocale, locales } = useLocale() + const servedByEngine = typeof window !== "undefined" && window.location.port !== "5173" && window.location.protocol.startsWith("http") const defaultBase = servedByEngine ? `${window.location.origin}/v1` : "http://127.0.0.1:8000/v1" const [baseUrl, setBaseUrl] = useState(() => { const saved = stored(localStorage, "colibri.baseUrl", defaultBase) - // migrate: a stored FACTORY default pointing at another origin would trip CORS - // when the page is engine-served — upgrade it to same-origin once. if (servedByEngine && saved === "http://127.0.0.1:8000/v1" && defaultBase !== saved) return defaultBase return saved }) @@ -120,7 +119,7 @@ export default function App() { const result = await getHealth(baseUrl, apiKey) if (!disposed) { setHealth(result); setHealthError("") } } catch (cause) { - if (!disposed) setHealthError(cause instanceof Error ? cause.message : "Runtime metrics unavailable") + if (!disposed) setHealthError(cause instanceof Error ? cause.message : "status.runtimeUnavailable") } } const timer = window.setInterval(() => void poll(), 5000) @@ -155,13 +154,13 @@ export default function App() { } catch (cause) { if (!controller.signal.aborted) { setHealth(null) - setHealthError(cause instanceof Error ? cause.message : "Runtime metrics unavailable") + setHealthError(cause instanceof Error ? cause.message : "status.runtimeUnavailable") } } } catch (cause) { if (controller.signal.aborted) return setConnected(false) - setError(cause instanceof Error ? cause.message : "Could not reach the server.") + setError(cause instanceof Error ? cause.message : "status.serverError") } finally { if (probeRef.current === controller) { probeRef.current = null; setConnecting(false) } } @@ -227,7 +226,7 @@ export default function App() { if (controller.signal.aborted) { updateMessages((current) => current.filter((item) => item.id !== assistant.id || item.content)) } else { - setError(cause instanceof Error ? cause.message : "Generation failed.") + setError(cause instanceof Error ? cause.message : "status.generationFailed") updateMessages((current) => current.filter((item) => item.id !== assistant.id || item.content)) } } finally { @@ -241,22 +240,22 @@ export default function App() {
-
ACTIVE MODEL{model}
+
{t("topbar.activeModel")}{model}
- - - + + +
- {loading && tokenCount > 0 ? {tokenCount} tokens : null} - {!loading && tokPerSec != null ? {tokPerSec.toFixed(1)} tok/s : null} + {loading && tokenCount > 0 ? {t("topbar.tokens", { n: tokenCount })} : null} + {!loading && tokPerSec != null ? {t("topbar.tokPerSec", { n: tokPerSec.toFixed(1) })} : null} {!loading && ttft != null ? TTFT {(ttft/1000).toFixed(1)}s : null} {!loading && lastRun?.usage ? {lastRun.usage.prompt_tokens}→{lastRun.usage.completion_tokens} : null} {lastRun?.queueWaitMs != null ? queue {Math.round(lastRun.queueWaitMs)}ms : null} - slot {cacheSlot + 1} - + {t("topbar.slot", { n: cacheSlot + 1 })} +
@@ -335,11 +342,11 @@ export default function App() { {!messages.length ? (
- COLIBRÌ ENGINE -

Ask the giant.
Keep the machine yours.

-

Connect to a local colibrì server and stream responses directly from your hardware. Nothing leaves the endpoint you choose.

+ {t("hero.title")} +

{t("hero.subtitle")}
{t("hero.tagline")}

+

{t("hero.description")}

- {["Explain how expert routing works", "Write a small C benchmark", "Compare RAM and VRAM caching"].map((item) => )} + {[t("prompts.routing"), t("prompts.benchmark"), t("prompts.caching")].map((item) => )}
) : ( @@ -347,7 +354,7 @@ export default function App() { {messages.map((item) => (
{item.role === "user" ? "Y" : }
-
{item.role === "user" ? "You" : "colibrì"}
{item.content || }
+
{item.role === "user" ? t("chat.you") : t("chat.colibri")}
{item.content || }
))}
@@ -356,10 +363,10 @@ export default function App() {
- {error &&
{error}
} + {error &&
{t(error)}
}
-