Nicolas Perrin f4d8ec6e03 perf(gemm): route the unrouted K3 decode projections (#1200) ...

Signed-off-by: nperrin-fr <nperrin.ai@gmail.com>

2026-08-22 11:03:35 -04:00

..

agentic_benchmark

feat(spec): frontier-anchored multi-layered MTP (#1080)

2026-08-14 22:18:56 +08:00

ci

ci: move Kimi K3 benchmarks to MI350 (#1203)

2026-08-22 00:47:54 -07:00

ci_system

ci: cache verified remote wheels on Slurm (#1202)

2026-08-21 23:06:38 -07:00

cli

fix(cli): allocate the gateway metrics port right before spawn (#1090)

2026-08-13 17:48:27 -07:00

gemm_tuning

perf(gemm): route the unrouted K3 decode projections (#1200)

2026-08-22 11:03:35 -04:00

long_context_benchmark/tokenspeed

ci(perf): add 1M-context perf bench for qwen3.5 (#264)

2026-05-28 18:14:17 +08:00

random_benchmark/tokenspeed

fix(ci): stabilize Kimi K3 perf metrics (#935)

2026-08-04 16:21:26 -07:00

runtime

perf(metadata): fuse the paged write-location and page-table expansions (#1198)

2026-08-22 11:02:54 -04:00

__init__.py

Initial Release

2026-05-06 13:27:02 +00:00

long_prompt.txt

Initial Release

2026-05-06 13:27:02 +00:00

runners.py

refactor(pd): unify paged-cache contracts across models and transfer modes (#1057)

2026-08-14 01:43:02 -07:00

test_bench_timeout.py

feat(models): add Qwen2 / Qwen2.5 dense LM support (#39)

2026-05-14 11:35:54 -07:00

test_cache_transfer_layout.py

refactor(pd): clarify cache transfer contract ownership (#1108)

2026-08-15 13:37:03 -07:00

test_host_cache_storage.py

feat(cache): add unified L2 cache and decode retraction (#1002)

2026-08-09 14:51:48 -07:00

test_l2_layerwise_load.py

feat(cache): add unified L2 cache and decode retraction (#1002)

2026-08-09 14:51:48 -07:00

test_utils.py

fix(ci): load EvalScope datasets from Hugging Face (#1010)

2026-08-08 21:33:07 -07:00