EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

Commit Graph

  • 5ada33ffa0 Bump rustls from 0.23.16 to 0.23.18 in /rust (#2182) dependabot[bot] 2024-11-26 03:22:33 +08:00
  • 254fd130e2 [CI] Split test cases in CI for better load balancing (#2180) Lianmin Zheng 2024-11-25 04:58:16 -08:00
  • 538fa0ae13 [Fix] Avoid calling fill_vocab_mask for terminated requests (#2175) Yixin Dong 2024-11-25 04:31:25 -05:00
  • 55842eb81a feat: fused_moe fp8 monkey patch (#2174) Yineng Zhang 2024-11-25 17:06:36 +08:00
  • a866b65e1d Bump rust router to 0.0.8 Byron Hsu 2024-11-24 23:17:38 -08:00
  • 4b0a1c9365 Replace prob based with threshold based load balancing (#2170) Byron Hsu 2024-11-24 23:17:11 -08:00
  • 8e1adb8441 Allow overwrite flashinfer use_tensorcore (#2169) Lianmin Zheng 2024-11-24 20:58:17 -08:00
  • dd44173dad [Fused moe] add tuning fused configs for qwen2 57b and mixtral 8x7b (#2167) Xiaoyu Zhang 2024-11-25 10:37:50 +08:00
  • 8912b7637f Fix docs (#2164) Lianmin Zheng 2024-11-24 08:25:56 -08:00
  • be0124bda0 Rename triton_fused_moe -> fused_moe_triton (#2163) Lianmin Zheng 2024-11-24 08:12:35 -08:00
  • fe5d3e818f Balance CI tests (#2162) Lianmin Zheng 2024-11-24 07:38:52 -08:00
  • 731146f6cb Fix mixed chunked prefill in overlap mode (#2158) Lianmin Zheng 2024-11-24 07:17:37 -08:00
  • fa27161380 fix: use torch.sum for compatible (#2161) Yineng Zhang 2024-11-24 22:37:04 +08:00
  • 5652c56535 Update CI threshold & Improve code style (#2159) Lianmin Zheng 2024-11-24 06:29:38 -08:00
  • e3938b2f9c feat: update other MoE models deps (#2156) Yineng Zhang 2024-11-24 21:36:34 +08:00
  • c211e7b669 Simplify batch update (#2154) Lianmin Zheng 2024-11-24 04:47:10 -08:00
  • d90c3d6b8b fix: resolve end-of-file-fixer (#2157) Yineng Zhang 2024-11-24 20:38:26 +08:00
  • 9e8f8fbf95 feat: update gitignore and add tuning config for FusedMoE (#2155) Yineng Zhang 2024-11-24 20:24:58 +08:00
  • b509db5832 feat: remove the dependency on FusedMoE (#2153) Yineng Zhang 2024-11-24 20:09:27 +08:00
  • dbe1729395 Merged three native APIs into one: get_server_info (#2152) Henry Hyeonmok Ko 2024-11-24 01:37:58 -08:00
  • 84a1698d67 Update release-pypi-router.yml Byron Hsu 2024-11-23 17:35:25 -08:00
  • 32293a299c Improve sglang router (#2148) Byron Hsu 2024-11-23 17:34:24 -08:00
  • 7921690893 add prefix match for certain tenant (#2147) Byron Hsu 2024-11-23 15:35:02 -08:00
  • bbb81c2457 Add more api routes (completion, health, etc) to the router (#2146) Byron Hsu 2024-11-23 15:10:26 -08:00
  • 52f58fc42a fix dp_rank env (#2144) Byron Hsu 2024-11-23 11:46:21 -08:00
  • 145c0ddc2d update router doc (#2143) Byron Hsu 2024-11-23 11:01:04 -08:00
  • 505d7f71a6 Bump sglang-router to 0.0.5 (#2142) Byron Hsu 2024-11-23 08:35:46 -08:00
  • cbedd1db1d [router] cache-aware load-balancing router v1 (#2114) Byron Hsu 2024-11-23 08:34:48 -08:00
  • ad47749b82 fix: resolve bench_serving args (#2139) Yineng Zhang 2024-11-23 17:45:42 +08:00
  • 751c3a037c Fix dp print message (#2138) Lianmin Zheng 2024-11-23 01:22:26 -08:00
  • 60769be14d Add concurrency option for benchmark (#2136) Yunmeng 2024-11-23 17:07:07 +08:00
  • a78d8f8db3 [CI] Fix test cases (#2137) Lianmin Zheng 2024-11-23 01:00:07 -08:00
  • c5f865013e Fix grid size in Triton decoding kernel (#2134) Ke Bao 2024-11-23 16:51:46 +08:00
  • d98fa1e93d Add simple CPU offloading support. (#2081) Jani Monoses 2024-11-23 08:23:53 +02:00
  • 865233e256 Add initial support for intel Gaudi accelerators (#2121) Ankur Neog 2024-11-23 09:52:23 +05:30
  • 66d4859acf Revert "Only stream output on tp rank 0" (#2130) Lianmin Zheng 2024-11-22 15:46:16 -08:00
  • e1b63624d7 Only stream output on tp rank 0 (#2124) Lianmin Zheng 2024-11-22 15:13:44 -08:00
  • c35cd1f8c7 Expose max total num tokens from Runtime & Engine API (#2092) Henry Hyeonmok Ko 2024-11-22 15:10:10 -08:00
  • 72f87b723b feat(pre-commit): trim unnecessary notebook metadata from git history (#2127) Xuehai Pan 2024-11-23 05:04:51 +08:00
  • 62a4a339eb docs: fix module docstrings and copyright headers (#2077) Xuehai Pan 2024-11-22 22:16:53 +08:00
  • 2797bc3422 fix: add xgrammar dependency (#2126) Yineng Zhang 2024-11-22 20:53:11 +08:00
  • 9a00e6f453 chore: bump v0.3.6 (#2120) Yineng Zhang 2024-11-22 19:27:30 +08:00
  • 4f8c3aeafc minor: update gsm8k threshold (#2125) Yineng Zhang 2024-11-22 19:23:58 +08:00
  • 2369e88209 [minor] Clean up unused imports (#2122) Lianmin Zheng 2024-11-22 01:50:42 -08:00
  • ad30d5cf9a Benchmark with Pytorch Profiler easily (#2110) bjmsong 2024-11-22 15:29:50 +08:00
  • dfec7fca06 Rename sglang.bench_latency to sglang.bench_one_batch (#2118) Lianmin Zheng 2024-11-21 20:07:48 -08:00
  • 8048c28c11 Fix #2037 - Context length check does not take into out pad tokens for visual models (#2106) Jake Poznanski 2024-11-21 19:05:41 -08:00
  • 30af7dfb34 [router] add base_gpu_id server args & merged radix tree python reference (#2115) Byron Hsu 2024-11-21 17:13:33 -08:00
  • f6f713797b Add support for Qwen2-VL-based embedding models (#2055) James Xu 2024-11-21 17:24:25 -05:00
  • f35cb46cc3 ROCm: Fix MoE padding for none FP8 cases (#2111) HAI 2024-11-21 12:23:21 -08:00
  • 7f8fcd39cd Turn off autotune for scaled mm for fp8 dynamic quant in torchao (#2116) Jerry Zhang 2024-11-21 12:19:49 -08:00
  • 5c6a41facf Error out when torchao-config option is not recognized (#2107) Jerry Zhang 2024-11-20 17:37:28 -08:00
  • 722530fa01 Enable overlap scheduler by default for the triton attention backend (#2105) Lianmin Zheng 2024-11-20 02:58:35 -08:00
  • 56a347f7d3 Move test_session_id.py to playground (#2104) Lianmin Zheng 2024-11-20 01:28:27 -08:00
  • 3295cd8af2 Allow skipping warmup in bench_offline_throughput.py (#2103) Lianmin Zheng 2024-11-20 01:25:21 -08:00
  • 5942dfc00a [feat] Add session control (#2073) Ying Sheng 2024-11-20 00:36:53 -08:00
  • 63a395b985 Update nightly-eval.yml (#2100) Lianmin Zheng 2024-11-19 22:15:02 -08:00
  • 7d671e4ad2 Enable overlap by default (#2067) Lianmin Zheng 2024-11-19 22:07:58 -08:00
  • 699384cb01 Set schedule policy more conservative for DP attention (#2096) Ke Bao 2024-11-20 12:57:18 +08:00
  • ffd20fcd03 Make constrained decoding work for overlap scheduler (#2095) Lianmin Zheng 2024-11-19 15:04:43 -08:00
  • 55bd97f3e5 minor: add dataset dump and questions shuffle (#2093) Yineng Zhang 2024-11-20 06:07:27 +08:00
  • e57c3e12b8 Use native fp8 format on MI300X (#2094) HAI 2024-11-19 14:06:29 -08:00
  • f239268fad minor: update gsm8k eval (#2091) Yineng Zhang 2024-11-19 20:36:55 +08:00
  • 929c7621af Fix: incorrect top_logprobs in chat completion (#2088) Alexander Waitz 2024-11-19 04:21:36 -08:00
  • b7a065eae3 Use cuda event wait and synchronization instead of busy waiting (#2089) Lianmin Zheng 2024-11-19 00:21:46 -08:00
  • b110453802 Simplify logits penalizer (#2086) Lianmin Zheng 2024-11-18 17:48:28 -08:00
  • 3b44bbeecf Allow passing extra request body to bench_offline_throughput.py (#2085) Lianmin Zheng 2024-11-18 14:59:15 -08:00
  • 80e2c4a8de Fix chunked prefill with output logprob (#2083) Lianmin Zheng 2024-11-18 13:16:28 -08:00
  • 66318ffe96 Rename layer_idx to layer_id for consistency (#2078) Jani Monoses 2024-11-18 23:00:02 +02:00
  • 766192610e feat: update torch 2.5.1 (#2069) Yineng Zhang 2024-11-18 21:29:13 +08:00
  • 2a3992b6f1 support set role as 'tool' (#2075) yukavio 2024-11-18 17:06:59 +08:00
  • 4af3f889fc Simplify flashinfer indices update for prefill (#2074) Lianmin Zheng 2024-11-18 00:02:36 -08:00
  • df7fe4521a Crash the CI jobs on model import errors (#2072) Lianmin Zheng 2024-11-17 22:18:11 -08:00
  • a7164b620f Tune the threshold for accuracy tests in CI (#2071) Lianmin Zheng 2024-11-17 21:51:00 -08:00
  • 116685337e Fix cuda illegal memory access in overlap mode (#2070) Lianmin Zheng 2024-11-17 21:29:30 -08:00
  • a9e90b4bce [Minor] Fix styles for overlap mode (#2068) Lianmin Zheng 2024-11-17 19:49:20 -08:00
  • 8c280cee55 add phi-3 small support (#2062) Tanjiro 2024-11-17 18:47:43 -08:00
  • 9c745d078e [Performance] Update xgrammar-related constrained decoding (#2056) DarkSharpness 2024-11-18 09:58:49 +09:00
  • ebaa2f3199 Rename arguments --disable-nan-detection to --enable-nan-detection (#2066) Lianmin Zheng 2024-11-17 16:53:44 -08:00
  • 62832bb272 Support cuda graph for DP attention (#2061) Ke Bao 2024-11-18 08:29:20 +08:00
  • 11f881d173 Deprecate --disable-flashinfer and --disable-flashinfer-sampling (#2065) Lianmin Zheng 2024-11-17 16:20:58 -08:00
  • 38625e2139 Remove monkey_patch_vllm_dummy_weight_loader (#2064) Lianmin Zheng 2024-11-17 15:48:12 -08:00
  • c1f401fc58 Revert "chore: update torch v2.5.1" (#2063) Lianmin Zheng 2024-11-17 15:29:38 -08:00
  • 3b878863f7 chore: update torch v2.5.1 (#1849) Yineng Zhang 2024-11-18 00:06:00 +08:00
  • f719d9aebc Launch dp ranks in parallel (#2053) Lianmin Zheng 2024-11-16 17:13:36 -08:00
  • edad373135 Fix illegal memory access in overlap mode & Use more fused triton kernels for building meta data (#2051) Lianmin Zheng 2024-11-16 16:14:23 -08:00
  • 976bc302e5 Support DP MLA (#1970) Ke Bao 2024-11-16 17:01:43 +08:00
  • 2f2e07439c Fix weight update for data parallelism (#2050) Lianmin Zheng 2024-11-16 00:30:39 -08:00
  • 2ffe0a7363 Add get_amdgpu_memory_capacity() (#2049) HAI 2024-11-15 22:51:48 -08:00
  • cf2489762b Add Tensor Parallel to torch_native_llama (#1876) Ke Wen 2024-11-15 21:26:00 -08:00
  • e5c6715003 Fix core (MI300X) with --enable-overlap (#2048) HAI 2024-11-15 21:24:42 -08:00
  • 023d0a73df fix small typos in docs (#2047) Xiaoyu Zhang 2024-11-16 03:09:10 +08:00
  • 32c9a7ec11 Release v0.3.5.post2 (#2046) Lianmin Zheng 2024-11-15 06:54:00 -08:00
  • b01df48cf2 [Fix] Adjust default chunked prefill size and cuda graph max bs according to GPU memory capacity (#2044) Lianmin Zheng 2024-11-15 06:21:57 -08:00
  • c29b98e043 Fix json benchmark (#2043) Lianmin Zheng 2024-11-15 05:33:43 -08:00
  • 954f4e6bd6 benchmark json schema (#2030) DarkSharpness 2024-11-15 22:06:19 +09:00
  • 2558d6a675 Fix the default arguments of bench_offline_throughput.py & simplify detokenizer manager (#2042) Lianmin Zheng 2024-11-15 05:02:44 -08:00
  • 29ebe3dff4 fix: align enable_overlap_scheduler naming between code and docs (#2038) ws 2024-11-15 19:39:10 +08:00
  • f6dd648620 Offline LLM Engine Benchmark Throughput (#1968) zolinthecow 2024-11-14 21:59:33 -08:00
  • ea53c63bad Expose no_stop_trim and skip_special_tokens in openai api (#2039) Lianmin Zheng 2024-11-14 19:09:21 -08:00