url/colibri
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Commit Graph

  • 309f20c939 E5: moe_submit relies on the residency set instead of per-buffer useResource: monotophic 2026-07-18 13:56:54 -04:00
  • 0049ea15ae E5: MTLResidencySet lifecycle scaffolding (init/register/unregister/shutdown) monotophic 2026-07-18 13:56:06 -04:00
  • 22560ad15b README: vision, the JIT-for-weights framing, roadmap, acknowledgements ZacharyZcR 2026-07-20 00:08:28 +08:00
  • bca4e95d92 site: redesign — measured atlas everywhere, vision, algorithm, models ZacharyZcR 2026-07-20 00:08:28 +08:00
  • 3e45074bf2 site: official website — animated demo, expert brain, 3-D atlas, Pages deploy ZacharyZcR 2026-07-19 23:31:37 +08:00
  • 7d01d21023 fix Windows async expert loader ZacharyZcR 2026-07-18 18:44:27 +08:00
  • 1d9f554715 cuda: cache layernorm weights on-device in pipe_layer_sparse (kill 152 sync H2D/token) + overlap-window profiling counters ZacharyZcR 2026-07-17 15:41:38 +08:00
  • ab55f4900c cuda: COLI_GROUP_ASYNC=1 — async expert-group issue/take with CPU/GPU overlap at decode (opt-in, +6-8% measured) ZacharyZcR 2026-07-17 15:32:09 +08:00
  • 4586d33c60 fix: MTP-head probe uses the last expert by index, not a hardcoded 255 Nicholas Beerbower 2026-07-19 10:54:01 -04:00
  • 27b4c4263a Added Dockerfile and instructions for using Colibrì through a docker container Davide Quack 2026-07-19 16:49:29 +02:00
  • e9225d2ffc docs: dual-SSD streaming env vars in ENVIRONMENT.md Steve Markgraf 2026-07-19 15:02:14 +02:00
  • 63966ba3f8 Dual-SSD streaming: COLI_MODEL_MIRROR reads experts from two model copies Steve Markgraf 2026-07-14 01:23:18 +02:00
  • 26d6b2d662 sync: dev noobdev-ph 2026-07-19 21:47:34 +08:00
  • 468b190db9 Merge remote-tracking branch 'upstream/main' into feat/gpu-backend-hardening noobdev-ph 2026-07-19 21:39:57 +08:00
  • 61004dcb84 Merge pull request #391 from ZacharyZcR/refactor/split-colibri Vincenzo Fornaro 2026-07-19 15:29:40 +02:00
  • 8a9a0fca4d plan: auto-tune heuristics — bottleneck classification + parameter decisions ZacharyZcR 2026-07-19 04:50:45 +08:00
  • 083fda5b0a fix: update test_logit_nan to include colibri.c instead of glm.c ZacharyZcR 2026-07-19 21:18:33 +08:00
  • bc69a9a6d0 fix: remove duplicate argmax_v — use NaN-safe version from sample.h ZacharyZcR 2026-07-19 21:15:54 +08:00
  • 4a7f630cad serve: #401 diagnostics + optional single-system-block tool rendering (experiment) p401-experiment JustVugg 2026-07-19 15:15:47 +02:00
  • 93b4a8e78e ci: fix Windows CUDA DLL check — glm.exe → colibri.exe ZacharyZcR 2026-07-19 04:27:10 +08:00
  • e486574442 web: i18n support — English, 简体中文, 繁體中文, Italiano ZacharyZcR 2026-07-19 04:22:46 +08:00
  • 420a0720c3 docs: add simplified Chinese and Italian README, update language nav ZacharyZcR 2026-07-19 04:02:00 +08:00
  • f853ea8a0b refactor: split glm.c into colibri.c + 4 header modules ZacharyZcR 2026-07-19 03:56:56 +08:00
  • 8f33bd153b Merge pull request #420 from JustVugg/p417-metal-lfru Vincenzo Fornaro 2026-07-19 15:02:41 +02:00
  • cfcc742591 metal: advance the LFRU recency clock on the GPU-prerouted decode path (#417) p417-metal-lfru JustVugg 2026-07-19 14:58:32 +02:00
  • ebc851edb3 Merge pull request #415 from JustVugg/docs-quickstart Vincenzo Fornaro 2026-07-19 14:25:24 +02:00
  • 40a3596354 nix: python3 is only needed for checks, not for the actual build Attila Oláh 2026-07-19 14:09:45 +02:00
  • 845af6378d docs: beginner-friendly Quick Start guide for Linux/Windows/macOS (#414) docs-quickstart JustVugg 2026-07-19 14:08:19 +02:00
  • 4775f28ceb nix: copy missing version.py to the environment Attila Oláh 2026-07-19 13:50:45 +02:00
  • 51fe03f615 nix: set main program to coli Attila Oláh 2026-07-19 11:36:13 +02:00
  • da97f0dbf1 nix: build on darwin with -march=native Attila Oláh 2026-07-19 10:59:16 +02:00
  • 58fd0e557f nix: commit lockfile, ignore result Attila Oláh 2026-07-19 10:48:00 +02:00
  • bcb984a61f nix: use getExe instead of hardcoding the path Attila Oláh 2026-07-19 10:40:19 +02:00
  • e93d574e13 nix: add the python env to check inputs Attila Oláh 2026-07-19 10:29:46 +02:00
  • a6d0a6c4a7 nix: use with pkgs in some package lists Attila Oláh 2026-07-19 10:21:50 +02:00
  • 73aef4d010 nix: configure and apply a formatter Attila Oláh 2026-07-19 10:14:24 +02:00
  • ac39be6b62 flake: remove unnecessary rec Attila Oláh 2026-07-19 10:06:27 +02:00
  • 3ffe4bb75e Merge pull request #413 from JustVugg/p-sec-trustboundary Vincenzo Fornaro 2026-07-19 13:11:06 +02:00
  • 72e36772f5 security: reject malformed model tensors at the untrusted-mirror boundary p-sec-trustboundary JustVugg 2026-07-19 13:05:56 +02:00
  • fae4b2cc3e Merge pull request #412 from JustVugg/p396-packaging Vincenzo Fornaro 2026-07-19 12:42:29 +02:00
  • 22509fccde Merge pull request #362 from EgonRuiter/prefetcher-v3 JustVugg 2026-07-19 12:40:43 +02:00
  • ca39e5333f packaging: single version source + honest editable-install semantics (on top of #396) p396-packaging JustVugg 2026-07-19 12:38:11 +02:00
  • 741d46ba25 Merge branch 'pr396' into p396-packaging JustVugg 2026-07-19 12:36:18 +02:00
  • 72874f38a2 Merge pull request #411 from JustVugg/dev v1.0.0 Vincenzo Fornaro 2026-07-19 12:30:32 +02:00
  • 3cd2674f68 release: fix the Windows job shell — 'msys2 {0}', not 'msys2' (+ inherit PATH for 7z) JustVugg 2026-07-19 12:29:33 +02:00
  • 819941ee2c Merge pull request #400 from JustVugg/dev Vincenzo Fornaro 2026-07-19 12:25:43 +02:00
  • e25aeecc4c Merge pull request #410 from JustVugg/p403-rss-guard Vincenzo Fornaro 2026-07-19 11:29:11 +02:00
  • 2f8fefd701 Merge pull request #408 from JustVugg/p401-tools-e2e Vincenzo Fornaro 2026-07-19 11:06:14 +02:00
  • c90e2cc438 glm: measured-RSS guard — the RAM budget enforces itself at the safe point (#403) p403-rss-guard JustVugg 2026-07-19 11:02:31 +02:00
  • 0f606bc446 test: skip the tools e2e suite on Windows (shebang mock engine) p401-tools-e2e JustVugg 2026-07-19 11:01:26 +02:00
  • 2122c004d9 serve: end-to-end tool-calling regression test + unparsed-marker diagnosis (#401) JustVugg 2026-07-19 10:56:00 +02:00
  • d94c68149d Merge pull request #352 from woolcoxm/fix/test-stops-windows JustVugg 2026-07-19 10:42:41 +02:00
  • 66b5e57bc4 Merge pull request #397 from NeuralNotwerk/fix/mtp-int8-repair Vincenzo Fornaro 2026-07-19 10:40:07 +02:00
  • 3f99d2bbb7 Merge pull request #395 from Stonki13/fix/windows-cuda-detection Vincenzo Fornaro 2026-07-19 10:40:01 +02:00
  • 19bab420a0 Merge pull request #404 from anrasi/fix/indir-meta-resume Vincenzo Fornaro 2026-07-19 10:39:56 +02:00
  • 36d389bf50 Merge pull request #402 from anrasi/fix/coli-run-think Vincenzo Fornaro 2026-07-19 10:39:50 +02:00
  • 6690c4f58e Merge branch 'p394' into trialsafe JustVugg 2026-07-19 01:28:16 +02:00
  • e5b1e19022 Merge branch 'p366' into trialsafe JustVugg 2026-07-19 01:26:51 +02:00
  • 4d780e2b46 Merge branch 'p378' into trialsafe JustVugg 2026-07-19 01:25:40 +02:00
  • d86a6b93ad packaging: pyproject.toml + editorconfig + clang-format ZacharyZcR 2026-07-19 06:10:14 +08:00
  • d2d3a7b559 Fix CUDA detection on Windows: cuda_binary()/cuda_linkage() always returned False Stonki13 2026-07-18 23:18:16 +02:00
  • 05bba7994c release: version infrastructure + GitHub Release workflow ZacharyZcR 2026-07-19 05:01:41 +08:00
  • d6676c10e9 tools: repair_mtp_int8.py — fix int4-converted MTP heads in place; warn on --mtp --ebits <8 recviking 2026-07-18 16:27:29 -04:00
  • 308e269f46 convert(fp8->int4): --indir copies the full metadata set and resumes interrupted passes (#383) andres-gb10 2026-07-18 21:11:40 +01:00
  • 799fe41f88 coli: honor THINK=1 in run mode andres-gb10 2026-07-18 16:45:19 +01:00
  • 6a2ab47f3b convert: resolve --mtp/--indexer shards from the local index instead of scanning all of them (#383) bokiko 2026-07-18 15:15:26 +03:00
  • caa49f7fa0 Merge pull request #384 from ZacharyZcR/perf/auto-numa Vincenzo Fornaro 2026-07-18 12:52:08 +02:00
  • 3fed654473 Merge pull request #372 from ZacharyZcR/feat/paged-ragged-kv Vincenzo Fornaro 2026-07-18 12:48:07 +02:00
  • 4fcbea9056 Merge branch 'p370' into t370 JustVugg 2026-07-18 12:41:07 +02:00
  • 9c17dc6d29 plan NUMA interleave on multi-socket Linux ZacharyZcR 2026-07-18 18:38:40 +08:00
  • 6d3a6168e4 cuda: keep ragged KV resident across decode steps ZacharyZcR 2026-07-18 04:14:46 +08:00
  • f8e4dc95b5 serve: honor --ngen when the client omits max_tokens; convert: print the resolved plan (#382, #383) JustVugg 2026-07-18 12:28:27 +02:00
  • 8a1dccae3b docs(api): tell coding-CLI users about the prefill cost before they hit it (#373) JustVugg 2026-07-18 02:33:23 +02:00
  • 605b0e28f9 Merge pull request #365 from ZacharyZcR/feat/ragged-batch-attention-upstream Vincenzo Fornaro 2026-07-18 02:04:43 +02:00
  • b09273f73d serve: MTP status message tells the truth in multiplexed mode (#358) JustVugg 2026-07-18 01:57:17 +02:00
  • 679c0742b2 telemetry: split expert hits into pin-tier vs LRU ecache (#336) JustVugg 2026-07-18 01:43:50 +02:00
  • db09466d3d convert(fp8->int4): --mtp/--indexer respected on the --indir path, no more container overwrite (#355) JustVugg 2026-07-18 01:24:43 +02:00
  • eb6d4325dc docs: collapse WINDOWS.md/windows.md case collision into one file (#371) JustVugg 2026-07-18 01:15:48 +02:00
  • 4fb5b4e975 sampling: survive non-finite logits instead of emitting token 0 forever (#369) JustVugg 2026-07-18 01:10:33 +02:00
  • 6ce6cc4380 Merge pull request #374 from ZacharyZcR/perf/p0-execution-profile Vincenzo 2026-07-18 00:57:35 +02:00
  • d5dac9bd6c Merge pull request #376 from KingIcyCreamProjects/docs/connect-coding-cli Vincenzo 2026-07-18 00:57:31 +02:00
  • 1deaa250f5 Merge pull request #375 from bokiko/fix/cpu-absorb-fmt4 Vincenzo 2026-07-18 00:54:57 +02:00
  • 153ee16f61 bench: multi-seed bench_dsa_select — kill single-input pivot-luck spike (#357) woolcoxm 2026-07-17 18:17:29 -04:00
  • 570d738ed5 profile effective CPU expert bandwidth ZacharyZcR 2026-07-18 05:34:41 +08:00
  • 73e0e5ec82 docs(api): add "Connect a coding CLI or editor" recipe (#373) KingIcyCreamProjects 2026-07-17 16:24:33 -05:00
  • 946fcd4f9f glm: fmt=4 support in qt_addrow/qt_matvec_rows — CPU absorb path decoded grouped int4 as int2 bokiko 2026-07-18 00:19:39 +03:00
  • c3a90eca36 profile CPU GPU tier execution costs ZacharyZcR 2026-07-18 04:54:59 +08:00
  • 786f98d471 cuda: load ragged attention entry point on Windows ZacharyZcR 2026-07-18 03:50:48 +08:00
  • e2d39abd2d sampling: NaN-skip the mx scan too (review follow-up on #369) KingIcyCreamProjects 2026-07-17 14:39:14 -05:00
  • 7f70a8db5d sampling: guard against non-finite logits (was silent token-0 spew) KingIcyCreamProjects 2026-07-17 14:08:50 -05:00
  • 92ddc2234f Merge remote-tracking branch 'upstream/dev' into feat/gpu-backend-hardening noobdev-ph 2026-07-18 02:55:12 +08:00
  • ef97f3cf2b Windows: fix non-ASCII chat prompt corruption (ANSI codepage vs UTF-8) volognamor 2026-07-17 21:49:28 +03:00
  • 8b36736e5d Merge branch 'p357' into trial357 JustVugg 2026-07-17 20:47:21 +02:00
  • 70a58799d6 Merge branch 'p354' into trialsamp JustVugg 2026-07-17 20:45:53 +02:00
  • 1a956b0119 Merge pull request #331 from nbeerbower/st-pread-full Vincenzo 2026-07-17 20:42:42 +02:00
  • 19f267794f Merge pull request #329 from mohamedmastouri2000-boop/docs/windows-native Vincenzo 2026-07-17 20:42:38 +02:00
  • d2226e222f Merge pull request #346 from bm1016bm-svg/docs/readme-zh-tw Vincenzo 2026-07-17 20:42:35 +02:00
  • d6d28f0c5c Merge pull request #353 from woolcoxm/docs/environment-refresh-dev Vincenzo 2026-07-17 20:42:31 +02:00
  • c1e709eb96 Merge pull request #351 from LordMZTE/nix-flake Vincenzo 2026-07-17 20:42:28 +02:00
  • 3d0a28b7bf Merge pull request #350 from woolcoxm/fix/build-warnings-dev Vincenzo 2026-07-17 20:42:25 +02:00