Commit Graph
100 Commits
Author SHA1 Message Date
Liangsheng Yin 73dfd2dfb1 [Test] Enhance radix cache test for spec cases (#12394) 2025-10-30 18:46:28 +08:00
Liangsheng Yin 42f8ea4030 [Test] Fix session control test (#12336) 2025-10-29 18:28:04 +08:00
Liangsheng Yin 14cbe42fd3 Refactor abortion in event loop (#12312) 2025-10-29 18:25:20 +08:00
Liangsheng Yin ea385ae85a Fix ITL metrics when using openai endpoint with spec (#12156) 2025-10-26 18:06:25 +08:00
Liangsheng Yin 5994e6c373 Do not use MagicMock to mock server_args in tests (#12154) 2025-10-26 16:00:46 +08:00
Liangsheng Yin 8491c794ad [misc] depdencies & enviroment flag (#12113) 2025-10-26 14:52:35 +08:00
Liangsheng Yin bda3758fac [log] Make forward iter count optional (#12116) 2025-10-26 14:51:07 +08:00
Liangsheng Yin 6c18addb6f Revert "Support nvidia/NVIDIA-Nemotron-Nano-9B-v2-FP8/NVFP4" (#12015) 2025-10-23 21:27:58 +08:00
Liangsheng Yin 32852fe9e9 Move memory runtime checker to mixin class (#12014) 2025-10-23 20:53:26 +08:00
Liangsheng YinandXiaoyu Zhang 9d61205dac [lint] improve ruff check (#11922)
Co-authored-by: Xiaoyu Zhang <35585791+BBuf@users.noreply.github.com>
2025-10-22 11:32:50 +08:00
Liangsheng Yin 97710ccd1a Fix flush cache API for spec v2 (#11918) 2025-10-21 23:01:16 +08:00
Liangsheng Yin da5bde4d16 Tiny fix main lint (#11862) 2025-10-20 19:57:24 +08:00
Liangsheng Yin 7a020e0f3b [Test] Add basic matched stop for beta eagle (#11833) 2025-10-20 01:17:00 +08:00
Liangsheng Yin 48738af7f9 [CI] always print back trace in retry() (#11834) 2025-10-20 01:12:49 +08:00
Liangsheng Yin d658f0497e [overlap-spec] fix stop condition and trimming (#11819) 2025-10-19 22:00:20 +08:00
Liangsheng Yin 57e25de756 Revert "Fix: Dynamic RoPE Cache Expansion to Prevent Position-ID Out-of-Bounds in EAGLE + Long-Sequence Workloads" (#11827) 2025-10-19 19:44:06 +08:00
Liangsheng Yin b288f4f440 Improve send_sone script (#11817) 2025-10-19 11:28:16 +08:00
Liangsheng Yin a93f10a722 [overlap-spec] support page size > 1 (#11772) 2025-10-19 02:09:13 +08:00
Liangsheng Yin cde5a6e30f Abstraction for spec worker and code cleanup (#11643) 2025-10-17 23:31:36 +08:00
Liangsheng Yin d88ac9bc9a [overlap-spec] Make plan stream an option (#11724) 2025-10-17 15:48:57 +08:00
Liangsheng Yin ce11dd82dc [CI] Try fix broken event loop init (#11746) 2025-10-17 13:30:17 +08:00
Liangsheng Yin 8221f9ae8b Tiny cleanup some eagle unused codes (#11660) 2025-10-15 17:24:08 +08:00
Liangsheng Yin 5ea96ac7cc Reduce one step decode for draft model. (#11561) 2025-10-14 23:52:04 +08:00
Liangsheng Yin 5a33c3aae7 Optimize Triton Draft Backend (#11556) 2025-10-14 20:08:32 +08:00
Liangsheng Yin acc2327bbd Move deep gemm related arguments to sglang.srt.environ (#11547) 2025-10-14 00:34:35 +08:00
Liangsheng Yin bfadb5ea5f Adjust overlap event loop (#11507) 2025-10-14 00:33:19 +08:00
Liangsheng Yin 54a46a264d Remove tp_worker.worker (#11548) 2025-10-13 22:38:48 +08:00
Liangsheng Yin 516738b096 Depreate global_server_args_dict (#11528) 2025-10-13 19:34:43 +08:00
Liangsheng Yin 1083e7e3df Deprecate global_server_args_dict (#11331) 2025-10-13 01:20:47 +08:00
Liangsheng Yin 2157d12ae8 [CI] fix lint (#11509) 2025-10-13 01:07:21 +08:00
Liangsheng Yin f49419061d Move args from global_config to environ (#11332) 2025-10-12 21:29:31 +08:00
Liangsheng Yin 01e59e8247 Fix CI break by express-laned PRs. (#11499) 2025-10-12 21:06:06 +08:00
20a6c0a63d Beta spec-overlap for EAGLE (#11398)
Co-authored-by: Lianmin Zheng <15100009+merrymercy@users.noreply.github.com>
Co-authored-by: Hanming Lu <69857889+hanming-lu@users.noreply.github.com>
2025-10-12 11:02:22 +08:00
Liangsheng Yin 739daa63e4 Adjust logits metada init for target verify (#11467) 2025-10-11 21:17:04 +08:00
Liangsheng YinandShangming Cai c882b5ae75 [CI] improve disaggregation CI. (#11264)
Signed-off-by: Shangming Cai <csmthu@gmail.com>
Co-authored-by: Shangming Cai <csmthu@gmail.com>
2025-10-08 21:40:56 +08:00
Liangsheng Yin 4b4dc132fa Rename ngram_utils -> ngram_info (#11316) 2025-10-08 11:49:46 +08:00
Liangsheng Yin 3ddd7dc9f8 Introduce future indices (#11301) 2025-10-07 22:24:02 +08:00
Liangsheng Yin 501dfa6b42 Remove sampling info events and overlap thread file (#11300) 2025-10-07 21:34:25 +08:00
1519a89cfd Remove overlap thread (#11210)
Co-authored-by: Lianmin Zheng <15100009+merrymercy@users.noreply.github.com>
Co-authored-by: Hanming Lu <69857889+hanming-lu@users.noreply.github.com>
2025-10-07 20:12:12 +08:00
Liangsheng Yinandgemini-code-assist[bot] 8a8a608af9 [ci] fix pp test (#11294)
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
2025-10-07 14:20:04 +08:00
Liangsheng Yin 4cb5a5235e Tiny skip_sample adjust (#11225) 2025-10-05 23:41:04 +08:00
Liangsheng Yin 4726c9197f [minor] fix the lint (#11198) 2025-10-04 01:04:58 +08:00
Liangsheng Yin 04b86b3c5c [hot-fix] Fix CI break which caused by adding thinking_mode in eval (#11192) 2025-10-03 18:29:27 +08:00
Liangsheng Yin de89ef49da [CI]] Tee server logs to both file and stdout/stderr using PIPE (#11185) 2025-10-03 12:31:13 +08:00
Liangsheng Yin 3c699772c9 Introduce naming convention in io_struct and base sglang io classes. (#10133) 2025-10-03 10:55:13 +08:00
Liangsheng Yin 7ff740a6ce Remove dp balance metadata and minimul token balance. (#11170) 2025-10-03 01:48:15 +08:00
Liangsheng Yin bfcd9b2433 [grpc] style fix for grpc compilation. (#11175) 2025-10-03 01:44:29 +08:00
Liangsheng Yin 458611de77 Unify forward output datastructure (#11124) 2025-10-03 00:28:57 +08:00
Liangsheng Yin 7d00479950 Clean up ascend allocator (#11152) 2025-10-02 20:34:26 +08:00
Liangsheng Yin 6a29003410 Remove unused pack .item() in paged allocator. (#11156) 2025-10-02 18:01:21 +08:00
Liangsheng Yin 25e7dbe8af Fix ngram spec with page size > 1 (#11135) 2025-10-02 12:34:23 +08:00
Liangsheng Yin 73d4a5f879 Organize spec-related data structures (#10735) 2025-10-01 09:45:30 +08:00
Liangsheng Yin 9710f718fb [Eval] Add --repeat in run_eval (#11101) 2025-09-30 23:35:54 +08:00
Liangsheng Yin 4a762041d7 move environ into sglang.srt to avoid break SRT auto sync. (#10791) 2025-09-23 02:04:20 -07:00
Liangsheng Yin f5a2faf2b8 Introduce FutureMap (#10715) 2025-09-22 23:27:30 -07:00
Liangsheng Yin 632b7d8cc9 Use simulate acc len from sglang.environ (#10771) 2025-09-23 12:59:50 +08:00
Liangsheng Yin 5ccf0b03bd [bench] Fix random seed in bench_one_batch_server (#10548) 2025-09-17 19:30:32 +08:00
Liangsheng Yin fa5d0bf6a5 Remove wrong imports from sglang.python (#10493) 2025-09-15 22:12:21 -07:00
Liangsheng Yin c3c26f76b3 [Env] minimal version for organizing envs (#10479) 2025-09-16 03:51:25 +08:00
Liangsheng Yin 2cf811a9da Fix --dataset-path in bench_one_batch_server (#10475) 2025-09-16 02:55:02 +08:00
Liangsheng Yin 305c9e8c2d [4/N]DP refactor: support watching mode get_load and shortest queue strategy (#10201) 2025-09-15 10:06:08 +08:00
Liangsheng Yin 55a6e644b0 [Hack] Add pd-disaggregation decode polling interval (#10411) 2025-09-14 10:18:23 +08:00
Liangsheng Yin 6897e06b69 Remove repeatedly lists adding in init_incremental_detokenization (#10412) 2025-09-14 10:05:52 +08:00
Liangsheng Yin 83d55ac51f [1/N]DP refactor: Improve dp rank scheduling in PD disaggregation mode. (#10169) 2025-09-09 12:27:55 +08:00
Liangsheng Yin 2c2b19b18b [CI] fix ambiguous argument in testing hybrid attentions. (#10161) 2025-09-08 18:16:52 +08:00
Liangsheng YinandHuang Long 72f9fc5f11 Monkey patch uvicorn multi worker is_alive timeout (#10159)
Co-authored-by: Huang Long <121648372+llll114@users.noreply.github.com>
2025-09-08 17:43:23 +08:00
Liangsheng Yin 78f139812a [1/N] DP-Refactor: move communicators into tokenizer_communicator_mixin (#10028) 2025-09-08 16:27:37 +08:00
Liangsheng Yin e719bb0e84 [1/2] Refactor multi-tokenizer manager (#10074) 2025-09-07 19:13:34 +08:00
Liangsheng Yin 6e95f5e5bd Simplify Router arguments passing and build it in docker image (#9964) 2025-09-05 12:13:55 +08:00
Liangsheng Yin 27e8ffed37 [1/N] DP-refactor: move dp balance code into scheduler's mixin class (#10004) 2025-09-04 16:53:58 +08:00
Liangsheng Yinandgemini-code-assist[bot] 5dfcd6c207 add proctitle for tokenizers (#9952)
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
2025-09-03 13:31:38 +08:00
Liangsheng Yin 6d3c20cf5b fix set_interal_state API (#9850) 2025-09-01 01:31:35 +08:00
Liangsheng YinandLianmin Zheng 836873b99f Fix memory leak when aborting decode request in PD-Disagg (#9817)
Co-authored-by: Lianmin Zheng <15100009+merrymercy@users.noreply.github.com>
2025-08-30 14:36:03 +08:00
Liangsheng Yinandgemini-code-assist[bot] a23c30205d Raise error when topk>1 and page>1 for paged attention backends. (#9784)
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
2025-08-29 12:47:34 +08:00
Liangsheng Yin d0934a5192 gpt-oss blog reproduction document (#9728) 2025-08-28 10:15:08 +08:00
Liangsheng Yinandgemini-code-assist[bot] 0ff7241995 Improve bench_one_batch_server script (#9608)
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
2025-08-26 10:38:37 +08:00
Liangsheng Yin 9b5f0f64f5 Fix tiny misalign with previous truncation setting in tokenizer_manager (#9430) 2025-08-21 14:05:35 +08:00
Liangsheng Yin eb19ccadae [bug] fix errors related to context length in SD (#9388) 2025-08-21 10:32:34 +08:00
Liangsheng Yinandgemini-code-assist[bot] 08ebdf79d0 Fix the --allow-auto-truncate argument in tokenizer manager. (#9391)
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
2025-08-20 16:56:47 +08:00
Liangsheng Yin 4d98e48649 Revert "[Misc] feat: Deepgemm update for sgl-kernel (#8790)" to fix kernel CI (#9260) 2025-08-17 22:59:50 +08:00
Liangsheng Yin 0c8594e67d Optional extension for green context (#9231) 2025-08-15 21:33:52 +08:00
Liangsheng Yin 445f9dca6e Runtime check CUDA driver version to avoid unresolved green context symbols (#9021) 2025-08-12 09:26:10 -07:00
Liangsheng Yin f9afa7dceb Fix docs for clip max new tokens (#9082) 2025-08-11 13:15:21 -07:00
Liangsheng Yin 7a27e798ca [CI] Do not trigger pd-disaggregation CI in draft PR (#8737) 2025-08-04 05:12:20 +08:00
Liangsheng Yin 9f47d686e5 Fix fused MoE when routed_scaling_factor is None (#8709) 2025-08-03 12:42:01 +08:00
Liangsheng Yin 603f5ce020 [Bug] fix green context's incompatibility with cuda < 12.4 (#8701) 2025-08-02 15:23:11 -07:00
Liangsheng Yin f9f0138f80 Revert "[1/2] sgl-kernel: Fuse routed scaling factor into select_experts" (#8706) 2025-08-02 20:14:30 +08:00
Liangsheng Yin e7dc163f57 add SVG logo (#8603) 2025-07-31 15:56:26 +08:00
Liangsheng Yin 76139bfba0 update mooncake in dockerfile (#7480) 2025-06-24 02:29:30 +08:00
Liangsheng Yin 25549433e8 Fix prefill OOM due to wrong token calculation when page > 1 (#7397) 2025-06-24 02:12:29 +08:00
Liangsheng Yin 05c9bc8956 [minor] simplify the TokenToKVPoolAllocator (#7414) 2025-06-22 12:37:18 +08:00
Liangsheng Yin 5ea5d22170 Fix CPU offloading for MLA memory pool (#7409) 2025-06-22 02:39:05 +08:00
Liangsheng Yin c494386728 minor fix (#7245) 2025-06-16 23:30:26 +08:00
Liangsheng Yin 930746d93c Improve log status (#7115) 2025-06-12 14:38:24 +08:00
Liangsheng Yinandfzyzcjy 78689d3393 PD Rust LB (PO2) (#6437)
Co-authored-by: fzyzcjy <5236035+fzyzcjy@users.noreply.github.com>
2025-05-29 20:50:10 +08:00
Liangsheng Yin 17d080b7ae Remove Cargo.lock, add it into .gitignore (#6438) 2025-05-20 12:01:32 +08:00
Liangsheng Yin a3e4e9bf9e Better PD initialization (#5751) 2025-05-07 01:12:57 +08:00
Liangsheng Yin 6d4d3bc81d Fix not "import os" (#6057) 2025-05-06 22:06:41 +08:00
Liangsheng Yin 40d9b8acce Improve overlap scheduling (#5788) 2025-04-28 11:19:16 +08:00
Liangsheng Yin beb65c7433 [PD]Reduce kv transfer threads (#5791) 2025-04-27 23:03:30 +08:00