Files

T

FABIOTESS 932f3678b0 convert: --up-bits/--gate-bits/--down-bits per-projection expert quant

Routed experts (gate/up/down) can now take different bit widths via PROJ_BITS.
Motivating config: --xbits 4 --up-bits 3 puts up_proj at int3-g64 (fmt=5, this
PR's format) while gate/down stay int4 — ~8% fewer expert bytes on disk and per
token, at ~zero quality cost. Backed by the OLMoE per-projection ablation posted
to #168: up@int3 matches int4-g64 (56.2 vs 55.8), up@int2 craters (-16pp).

This also supplies the definition the #404 resume manifests already depend on:
current dev records dict(PROJ_BITS) in check_or_record_params and the --indir
progress file in four places, but the global was never defined — every one of
those paths NameErrors at runtime today. The manifests were written for this
interface; this commit is the other half.

Validated: synthetic GLM fixture with --up-bits 3 yields int3-g64 up_proj
(O*(I/64)*24B weight + group scales) and int4-per-row gate/down.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

2026-07-20 17:31:05 +01:00

expert_atlas

tools: one expert atlas, not two — retire expert_atlas.py, analyze.py gains --web for the dashboard

2026-07-15 14:16:36 +08:00

__init__.py

Organize project tools and local workflows: c/tools, c/scripts, c/tests, root Makefile (flat C core untouched) (#22 )

2026-07-10 10:08:39 +02:00

bench_ux.sh

GPU resident pipeline: batch CUDA attention, head-sharded kv_b, prefill expert groups, W4A16 mixed dispatch (#111 )

2026-07-14 18:18:05 +02:00

benchmark_cuda_fixture.py

profile effective CPU expert bandwidth

2026-07-18 05:34:41 +08:00

clean.py

Makefile: portable clean/test-c via python helpers — works without sh.exe on native Windows (#179 , fixes #172 )

2026-07-14 13:49:26 +02:00

convert_fp8_to_int4.py

convert: --up-bits/--gate-bits/--down-bits per-projection expert quant

2026-07-20 17:31:05 +01:00

convert_olmoe_merged.py

fix(prefetch): address copilot code quality reviews on safety and concurrency

2026-07-16 14:36:53 +02:00

convert_olmoe.py

convert_olmoe.py: wire --ebits through to quantization (#323 )

2026-07-16 21:59:16 +03:00

download_glm52.py

Translate user-facing runtime output to English, machine prefixes preserved, + CLI output test (#67 , #85 )

2026-07-12 13:38:40 +02:00

efficiency.py

tests: efficiency suite — tiny-model regression gates + full-model optimization dossier

2026-07-20 10:49:07 -04:00

eval_glm.py

security: harden safetensors/JSON parsers against malicious model files

2026-07-15 08:13:20 +02:00

fetch_benchmarks.py

bench: survive transient hub errors — retry with backoff, per-task isolation, atomic writes (#304 )

2026-07-16 13:20:35 +03:00

gen_unicode.py

Translate user-facing runtime output to English, machine prefixes preserved, + CLI output test (#67 , #85 )

2026-07-12 13:38:40 +02:00

glm_fp8_emit.py

diskio: KV write batching + persistent handle; generators: unfuse experts

2026-07-15 02:32:12 -04:00

iq3xxs_grid.json

tools: IQ3_XXS-codebook scheme in the quant ablation (#452 step 1)

2026-07-20 12:55:44 +08:00

make_glm_bench_model.py

diskio: KV write batching + persistent handle; generators: unfuse experts

2026-07-15 02:32:12 -04:00

make_glm_oracle.py

Merge pull request #322 from woolcoxm/fix/oracle-transformers-pin

2026-07-16 19:46:20 +02:00

make_olmoe_real_oracle.py

refactor(prefetch): clean up unused variables, dead functions, and hardcoded paths

2026-07-16 14:22:04 +02:00

quant_ablation.py

tools: IQ3_XXS-codebook scheme in the quant ablation (#452 step 1)

2026-07-20 12:55:44 +08:00

README.md

Organize project tools and local workflows: c/tools, c/scripts, c/tests, root Makefile (flat C core untouched) (#22 )

2026-07-10 10:08:39 +02:00

repair_mtp_int8.py

tools: repair_mtp_int8.py — fix int4-converted MTP heads in place; warn on --mtp --ebits <8

2026-07-18 16:27:29 -04:00

route_coupling_report.py

routing: cross-layer coupling trace dump + pair-table builder + opt-in coupled prefetch source (#176 )

2026-07-14 16:13:19 +02:00

route_pairs.py

routing: cross-layer coupling trace dump + pair-table builder + opt-in coupled prefetch source (#176 )

2026-07-14 16:13:19 +02:00

run_tests.py

run_tests: normpath test binaries — forward-slash relative paths fail CreateProcess on Windows (#196 )

2026-07-14 18:01:12 +02:00

test_olmoe_real.py

refactor(prefetch): clean up unused variables, dead functions, and hardcoded paths

2026-07-16 14:22:04 +02:00

README.md

Tools

These scripts support model preparation and offline engineering work. They are not runtime dependencies of the C engine.

convert_fp8_to_int4.py, download_glm52.py: model preparation
make_glm_oracle.py, make_glm_bench_model.py: deterministic fixtures
benchmark_cuda_fixture.py, eval_glm.py, fetch_benchmarks.py: benchmarks
gen_unicode.py: tokenizer table generation

Run them from c/, for example:

python3 tools/convert_fp8_to_int4.py --selftest
python3 tools/make_glm_bench_model.py --output /tmp/colibri-bench