Commit Graph
88 Commits
Author SHA1 Message Date
shuwenn ff1fc1fbdf [mem_cache][5/N] refactor: extract host KV cache base layer into pool_host package (#27273) 2026-06-20 20:44:08 +08:00
shuwenn 941a11ada3 [HiCache] refactor: remove unused transfer buffer (#26003) 2026-06-19 09:46:04 +00:00
shuwenn b23477af44 bench: infer tokenizer from serving model info (#28195) 2026-06-15 20:03:12 -07:00
shuwenn f4029d0fc0 [HiCache] fix: clear storage reset state (#26009) 2026-06-12 23:51:26 -07:00
shuwenn 9d37e710b7 [Bench] Add consistent p90/p95/p99 percentiles for all latency metrics (#27662) 2026-06-12 13:44:38 -07:00
shuwennandClaude Opus 4.8 66076f2409 [mem_cache][3/N] refactor: move HiSparse allocators to allocator/hisparse.py (#26678)
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-06-11 19:50:24 +08:00
shuwennandQiaolin Yu 3600a9ac5f [SPEC] feat: init adaptive spec params from config (#27493)
Co-authored-by: Qiaolin Yu <liin1211@outlook.com>
2026-06-10 18:55:25 +00:00
shuwenn e57323cae9 [mem_cache][4/N] refactor: extract MambaTokenToKVPoolAllocator into allocator/ (#27256) 2026-06-07 10:46:29 +08:00
shuwenn 280280ace9 [SPEC] fix: import copy module for eagle sampling info clone (#27390) 2026-06-06 12:12:46 +00:00
shuwenn efe24704b7 [HiCache] feat: truncate mamba prefetch length to available host KV size (#26945) 2026-06-05 03:01:37 +08:00
shuwenn 1af53f67b2 [mem_cache][2/N] refactor: move SWATokenToKVPoolAllocator to allocator/swa.py (#26676) 2026-06-04 15:10:37 +08:00
shuwenn 1a57145975 [codex] Fix adaptive metrics test flake (#27135) 2026-06-03 19:07:17 -07:00
shuwenn f65aae8493 [HiCache] fix: truncate prefetch key on degraded allocation (#25991) 2026-06-03 22:06:18 +08:00
f4e7a98fe5 [HiCache] feat: support draft offload for mooncake (#24984)
Co-authored-by: huangtingwei9988 <141888744+huangtingwei9988@users.noreply.github.com>
Co-authored-by: stmatengss <11641725+stmatengss@users.noreply.github.com>
2026-06-02 21:42:04 -07:00
shuwenn f40b2ca9d3 chore(CODEOWNERS): add allocator/ owners and @alphabetc1 to mem_cache (#27003) 2026-06-01 22:07:15 -07:00
a0670b5ba3 [SPEC] feat: add adaptive speculative decoding metrics (#25940)
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Co-authored-by: Jarrod Barnes <jbarnes850@gmail.com>
2026-06-01 13:53:30 -07:00
shuwenn c06220159b [mem_cache][1/N] refactor: split allocator.py into allocator/ subpackage (#26675) 2026-05-31 20:31:29 +08:00
shuwennandmaodoudou168 68706e615a [SPEC] fix: use effective max draft tokens for adaptive spec initiali… (#26354)
Co-authored-by: maodoudou168 <maodoudou168@users.noreply.github.com>
2026-05-28 13:33:11 -07:00
shuwenn 36eb72bf12 [UnifiedTree] fix: backup SWA-split parent before child under write-through (#25065) 2026-05-25 00:21:57 +08:00
shuwennandClaude Opus 4.7 66ef97c00f [Lint] Fix Optional[X] = (None,) typo defaults in two dataclasses (#25252)
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-15 00:32:40 -07:00
shuwennandClaude Opus 4.7 d6b28b4a69 [Refactor] Remove dead key_convert_fn / convert_to_bigram_key (#25161)
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-13 13:54:36 -07:00
shuwenn 5ed9a494d0 [UnifiedTree] fix: allow partial match on evicted+backuped nodes (#24943) 2026-05-13 11:43:12 +08:00
shuwenn 5495026a3b [HiCache] feat: default storage prefetch timeout (#23309) 2026-05-11 18:49:35 -07:00
shuwenn b4d347e86e [SPEC V2] fix: skip stale state updates in spec-v2 overlap (#23456) 2026-05-09 23:56:24 -07:00
shuwenn d5564c2a96 fix(fa3): translate page table to SWA loc in EAGLE3 topk>1 spec metadata (#24617) 2026-05-09 18:22:45 +08:00
shuwennandClaude Opus 4.7 9d12f9e6fa [HiCache] ci: lower est_time for test_hicache_spec_file_storage (#24713)
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-09 00:33:18 -07:00
shuwennandQiaolin-Yu d9dddd4d7d [SPEC V2][2/N] feat: adaptive spec support spec v2 (#23336)
Co-authored-by: Qiaolin-Yu <liin1211@outlook.com>
2026-05-07 18:33:47 -07:00
shuwennandClaude Opus 4.7 03147f66b8 ci: add /rerun-group to rerun all registered tests in a group (#24023)
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-04-29 10:24:16 -07:00
shuwenn 2c41ef4c93 [HiCache] feat: add draft KV cache backing for L2/L3 (#21125) 2026-04-28 23:47:31 -07:00
shuwenn 233048212a [HiCache][SPEC] fix: normalize storage prefetch key (#23631) 2026-04-28 15:53:20 -07:00
shuwennandClaude Opus 4.6 3e1c5e1b74 [HiCache][SPEC] fix: empty key after page alignment in match_prefix (#23387)
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-04-28 14:06:55 -07:00
shuwenn f30a6f4d7e [DOC] Add DFLASH speculative decoding documentation (#23553) 2026-04-24 17:18:46 -07:00
shuwennandClaude Opus 4.6 4befc31408 fix: pass v_head_dim to MHA KV pools and validate MiMo HiCache geometry (#23173)
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-04-21 19:48:45 -07:00
e3782d04d2 fix: fallback to triton for attention-sink models (flashinfer unsupported) (#23139)
Co-authored-by: Liangsheng Yin <hnyls2002@gmail.com>
Co-authored-by: hnyls2002 <lsyincs@gmail.com>
2026-04-21 13:48:50 -07:00
shuwennandQiaolin-Yu b65799cf83 [SPEC][1/N] feat: add adaptive speculative_num_steps for EAGLE topk=1 (#21599)
Co-authored-by: Qiaolin-Yu <liin1211@outlook.com>
2026-04-20 14:25:04 -07:00
shuwenn dbcf7459b5 fix: reset empty prefill batch fullness (#23138) 2026-04-20 14:14:00 -07:00
shuwennandZhangheng 90c76d665e [HiCache] fix: HiCacheFile component key suffixing (#22891)
Co-authored-by: Zhangheng <hzh0425@apache.org>
2026-04-17 13:06:28 -07:00
shuwenn ec5742f4ab fix: Auto-correct page_size for Mamba no_buffer radix cache mode (#20538) 2026-04-08 00:19:31 +08:00
shuwenn 4f84ce5807 [CI] ci: add test_http_server_auth.py to CI (#21866) 2026-04-03 16:32:18 +08:00
shuwennandZhiqiang Xie 9cb362f70e [HiCache] fix: Clone host indices to avoid memory leak (#21624)
Co-authored-by: Zhiqiang Xie <xiezhq@stanford.edu>
2026-04-01 17:42:07 -07:00
shuwenn a1c725bdc5 fix: pre-init tokenizer_manager to avoid AttributeError in shutdown (#21824) 2026-04-01 10:54:53 -07:00
shuwenn 6098c51bc2 fix(MiMo-V2-Flash): add mimo reasoning parser (#21414) 2026-04-02 00:47:27 +08:00
shuwenn 8e84f846cc [Diffusion] Add --uvicorn-access-log-exclude-prefixes to suppress noisy access logs (#20379) 2026-04-01 09:31:44 +08:00
shuwenn c34593f951 [HiCache] fix: graceful shutdown of pending async tasks in bench_mix.py (#20276) 2026-03-29 00:46:32 -07:00
shuwenn 18074e25dc fix: scheduler launch hang when non-current rank dies (#20287) 2026-03-29 00:28:45 -07:00
shuwenn 646573e4e8 fix: use get_rope_config() to support models without rope_parameters (#21135) 2026-03-26 11:22:12 -07:00
shuwenn 2fba2bdad1 refactor: Remove dead code from utils/common.py (#20668) 2026-03-21 21:54:17 -07:00
shuwenn 6c91590e1b [HiCache] refactor: hicache normalization flow and compatibility checks (#19669) 2026-03-20 18:38:44 -07:00
shuwenn 515b3a323d feat: support human-readable suffixes (25.6k, 1M, 1Mi) for token CLI (#20577) 2026-03-16 20:05:33 -07:00
shuwenn 42f18fe560 [HiCache] fix: release write-through lock_ref during decode (#20049) 2026-03-16 14:49:31 +08:00
shuwenn 1ac6a26464 fix: Nemotron chunk size alias (#20458) 2026-03-14 23:23:39 -07:00
shuwenn 538acb4c46 fix: Add .text property to HttpResponse to prevent AttributeError (#20518) 2026-03-14 22:59:32 -07:00
shuwenn acab24a76a fix: gracefully abort last request in retract_decode on OOM (#19881) 2026-03-11 15:13:03 -07:00
shuwenn 5a11ae19c1 [CI] fix: notebook ci often OOM (#20199) 2026-03-09 22:32:41 -07:00
shuwenn 72f6dfcc31 fix: add ModelScope cache lookup and speculative path support (#20098) 2026-03-07 22:23:16 -08:00
shuwenn 7bd3dd9270 fix: image URL in notebook to use raw.githubusercontent.com (#20100) 2026-03-07 13:28:20 -08:00
shuwenn bdffb027a8 [CI] fix: handle missing repo in lora notebook (#19700) 2026-03-02 10:27:32 -08:00
shuwenn e3e71f275a docs: refactor speculative decoding doc (#19186) 2026-03-01 22:03:20 -05:00
shuwenn 0c224c3c62 docs: Embed release lookup tool into Sphinx documentation site (#19264) 2026-02-24 11:11:27 -08:00
shuwenn 4cf4f0859f [Doc] Convert the speculative decoding notebook to markdow (#18395) 2026-02-14 18:18:56 -08:00
shuwenn bd39de7d5e [Env] centralize hicache vars in environ.py (#17204) 2026-02-13 18:02:31 -08:00
shuwenn 3299c4f9c1 [CI] feat: add early exit to wait_for_server when process dies (#18602) 2026-02-13 16:46:09 -08:00
shuwenn bc2405e6c1 feat: support release lookup (#18450) 2026-02-13 10:47:02 +08:00
shuwenn 8da14aea88 [HiCache] fix: StorageMetricsCollector was initialized twice (#18354) 2026-02-10 00:53:00 -08:00
shuwenn f600965b0e [CI] fix: notebook ci may not working (#18417) 2026-02-07 22:26:35 -08:00
shuwenn ef1d0ea885 [Doc] add a summary section for spec decode document (#18323) 2026-02-05 16:34:31 -05:00
shuwenn 8b21dd4b77 [Doc] refine spec decode docs for SpecV2/STANDALONE/NGRAM (#18321) 2026-02-05 15:12:33 -05:00
shuwenn a723d1c5ef [model-gateway] ignore error for embeddings/classify in PD router (#15931) 2026-01-26 22:49:14 -05:00
shuwenn 57e432d951 fix: preserve disconnect events in api key middleware (#17253) 2026-01-26 22:48:24 -05:00
shuwenn fd3b179ffd [HiCache][HA 1/N] Support HiCache storage runtime attach/detach (#15892) 2026-01-26 19:33:19 -08:00
shuwenn 8fb45523f3 feat: support bitsandbytes quantization algorithm (#15325) 2026-01-19 18:36:56 -08:00
shuwenn 71279e31f7 [CI] fix test_vlm_models.py (#17049) 2026-01-19 10:00:00 -08:00
shuwenn 8ec160ed46 feature: support uvicorn access log filter(disable logging /metrics) (#15513) 2026-01-15 20:00:06 -08:00
shuwenn 9227d9f60c [Docs] sort and update server_arguments.md (#17163) 2026-01-15 12:07:18 -05:00
shuwennandXinyuan Tong de94d793ad feat: support qwen3(-VL) rerank scoring&chat template (#16403)
Signed-off-by: Xinyuan Tong <xinyuantong.cs@gmail.com>
Co-authored-by: Xinyuan Tong <xinyuantong.cs@gmail.com>
2026-01-15 00:45:46 +08:00
shuwenn 48c2aca9ba [Env] centralize pd vars in environ.py (#16264) 2026-01-14 23:06:01 +08:00
shuwennandSimo Lin cd33694585 feat: add --admin-api-key for finer-grained endpoint auth (#15908)
Co-authored-by: Simo Lin <linsimo.mark@gmail.com>
2026-01-13 20:21:55 -08:00
shuwenn ac11e6a7c5 Add alphabetc1 into CI_PERMISSION (#16198) 2025-12-31 11:31:47 +08:00
shuwenn c0fc7a89e7 [sgl-kernel] fix: make sgl-kernel build respect MAX_JOBS (#15575) 2025-12-31 10:44:45 +08:00
shuwenn 7380ec9d55 [CI] fix test_mla_deepseek_v3.py (#16096) 2025-12-29 10:18:02 -08:00
shuwenn fe3d47fc9d fix: warn once per env var key (#15846) 2025-12-26 18:41:15 +08:00
shuwenn 53f974b973 fix: potential crash for missing stream attribute (#15644) 2025-12-23 23:24:31 +08:00
shuwenn 758b9067a0 [CI] fix UT assert error in test_tokenizer_manager.py (#15646) 2025-12-23 22:24:33 +08:00
shuwenn 1d9ba2ce4d [model-gateway] Use UUIDs for router-managed worker resources (#15540) 2025-12-21 07:40:03 -10:00
shuwenn 1d90b194b2 [model-gateway] bugfix: backward compatibility for GET endpoints (#15413) 2025-12-20 10:37:53 -10:00
shuwenn 5a0ad7310e fix: update model name after weights update (#15416) 2025-12-19 21:53:14 +08:00
shuwennandLiangsheng Yin fb17845723 fix: unreachable error check in retraction (#15433)
Co-authored-by: Liangsheng Yin <lsyincs@gmail.com>
2025-12-19 16:33:34 +08:00
shuwenn 621061f017 [Bugfix] input prompt was not logged (#13936) 2025-11-28 16:00:51 +08:00