Commit Graph

3288 Commits

Author SHA1 Message Date
Yineng Zhang
051068673c chore: update config (#9591) 2025-08-25 03:41:09 -07:00
fzyzcjy
9dcdf5da03 Tiny fix wrong comments (#9589) 2025-08-25 03:08:10 -07:00
Yineng Zhang
ebd9dbe71b fix: revert #8593 (#9581) 2025-08-25 01:29:06 -07:00
Yineng Zhang
938e986e15 chore: upgrade flashinfer 0.2.14.post1 (#9578) 2025-08-25 00:12:17 -07:00
Yuhao Zhou
17d5eda887 bugfix for undefined logging functions in HarmonyBrowserTool & HarmonyPythonTool (#9229) 2025-08-25 00:10:35 -07:00
fzyzcjy
71a7f1d86f Offload tensors by sharding on GPU (#9536) 2025-08-25 00:02:49 -07:00
fzyzcjy
433266c125 Reintroduce memory usage fix (#9535) 2025-08-25 00:02:31 -07:00
Qi Yuhang
fda4792620 Update CUTLASS 4.2 & Enable K-Major Scale Factor for SM90 FP8 Blockwise Group GEMM (#9559) 2025-08-24 23:24:43 -07:00
miter
a0b22f2f17 remove redundant rank0_log function. (#9560)
Co-authored-by: linhuang <linhuang@ruijie.com.cn>
2025-08-24 23:17:55 -07:00
SCDESPERTATE
b5c6529e17 [PD] Improve disaggregation metrics output: update the metrics to keep reflecting real stats (#7317) 2025-08-24 23:16:43 -07:00
Beichen Ma
dd6ec02965 Add target module validation for init adapters (#9429) 2025-08-24 20:24:50 -07:00
Yineng Zhang
e0ab167db0 chore: bump v0.5.1.post1 (#9558) 2025-08-24 01:14:17 -07:00
Yineng Zhang
c807cd7c75 chore: update configurer (#9557) 2025-08-24 01:05:00 -07:00
Vincent Zhong
327f7b7c87 fix(grok): remove duplicate replicate_lm_head configuration (#9549) 2025-08-23 19:49:24 -07:00
Lianmin Zheng
97a38ee85b Release 0.5.1 (#9533) 2025-08-23 07:09:26 -07:00
Lianmin Zheng
86d10d220f Update grok.py and tiktoken tokenizer (#9532) 2025-08-23 05:40:18 -07:00
hzh0425
83871aa12d feat(hicache): Supports 3fs-hicache compatibility with dp-attention (#9372) 2025-08-23 02:08:32 -07:00
fzyzcjy
b1b3f0b38f Partially unify triton per token group quant kernels (#9485) 2025-08-23 02:07:31 -07:00
fzyzcjy
34e5e11f0f Tiny make device_loading_context more static (#9478) 2025-08-23 02:07:15 -07:00
fzyzcjy
2600fc0d47 Overlapped weight offload (#8034) 2025-08-23 02:06:46 -07:00
hlu1
ccd3fb946e [fix] Fix mxfp4 triton MoE tp bug (#9473)
Signed-off-by: Hao Lu <14827759+hlu1@users.noreply.github.com>
2025-08-23 01:48:40 -07:00
Chang Su
c9dd70fbde tool-call(dsv3): Improve deepseek-v3 chat template and tool_choice = required (#9525) 2025-08-23 01:46:56 -07:00
Yineng Zhang
6b2b8bf0e1 fix: blackwell dsv3 fp8 issue temporary solution (#9530) 2025-08-23 01:33:21 -07:00
fzyzcjy
0374304a2c Add enable_flashinfer_mxfp4_bf16_moe for higher precision and slower moe backend (#9004) 2025-08-23 15:38:40 +08:00
Chanh Nguyen
127d4b0d5e Support GC Freezing to improve latency & throughput (#9241)
Co-authored-by: Chanh Nguyen <cnguyen@linkedin.com>
Co-authored-by: Liangsheng Yin <hnyls2002@gmail.com>
2025-08-23 13:43:09 +08:00
Moein Khazraee
7e880286b5 Add support for extensions of interface and pre-registrations to NIXL HiCache (#9211)
Co-authored-by: Zhiqiang Xie <xiezhq@stanford.edu>
2025-08-22 20:06:13 -07:00
sogalin
c4500233ff Add Qwen3-30B-A3B-Thinking-2507 support on AMD GPUs. (#9456) 2025-08-22 13:14:42 -07:00
Hubert Lu
f445a1d9a3 [AMD] Fix Llama 4 FP8 accuracy issues on MI300X (#7699) 2025-08-22 13:13:45 -07:00
datdo-msft
110a65989b [MTP] Force greedy sampling on AMD (#9127) 2025-08-22 11:14:43 -07:00
Wenxuan Tan
0f587e80d3 Use Tensor Core Decode when gqa group size >= 4 (#8624)
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
2025-08-22 23:25:15 +08:00
huangtingwei
6078d5fcc0 [HiCacheStorage] backup optimization for MLA model (#8865)
Co-authored-by: Zhiqiang Xie <xiezhq@stanford.edu>
2025-08-22 18:03:51 +08:00
pansicheng
70cf4abccc 3fs zerocopy (#9109)
Co-authored-by: Zhiqiang Xie <xiezhq@stanford.edu>
2025-08-22 17:56:38 +08:00
Xuchun Shang
cebf45994b [bugfix] Make --enable-hierarchical-cache and --disable-radix-cache mutually exclusive (#9452)
Signed-off-by: Xuchun Shang <xuchun.shang@linux.alibaba.com>
2025-08-22 17:49:52 +08:00
Qiaolin Yu
9c0c1e30b2 Disable torch.compile for get_last_loc_large_page_size_large_top_k (#9507)
Co-authored-by: ispobock <ispobaoke@gmail.com>
2025-08-22 02:05:02 -07:00
Mick
a1f011d09a minor: determine mm attn backend based on platforms (#9303) 2025-08-22 01:08:41 -07:00
Qiaolin Yu
9ec314c6ac Support speculative decoding in the trtllm_mha attention backend (#9331)
Co-authored-by: ispobock <ispobaoke@gmail.com>
2025-08-21 23:53:35 -07:00
Elfie Guo
243e745d07 Add trtllm_mla and cutlass_mla for ragged fmha for chunked prefill (#9480) 2025-08-21 23:01:36 -07:00
timmy-feng
61a0e600df torch.compile() mrope (#9487) 2025-08-21 23:01:08 -07:00
Yongfei Xu
9708d353b7 Support MHA with chunked prefix cache for flashinfer/flashmla backend, support page size > 1 for MHA chunked prefix (#8616)
Co-authored-by: xuyongfei.xyf <xuyongfei.xyf@antgroup.com>
2025-08-21 18:19:44 -07:00
Pavani Majety
3cc3d9b950 Add Support for Page Size greater than 1 for Flashinfer MLA Backend (#8593)
Signed-off-by: Pavani Majety <pmajety@nvidia.com>
2025-08-21 18:15:06 -07:00
Xinyuan Tong
6c855db82c Revert "bugfix: Fix output_ids extraction in detokenizer_manager" (#9467) 2025-08-21 17:24:25 -07:00
Yineng Zhang
0f9318f7d0 feat: update auto_choose_speculative_params (#9470) 2025-08-21 17:12:12 -07:00
Yineng Zhang
849957bc76 fix: tmp revert gpt oss tp sharding on hopper (#9469) 2025-08-21 17:03:21 -07:00
Stefan He
cded039b57 [FA3] Init Spec Page Table only when Spec is enabled to save ~40MB (#9455) 2025-08-21 15:11:38 -07:00
zixuanzhang226
275f9df381 feat: add fused moe config for GLM-4.5-Air-FP8 on B200 (#9463) 2025-08-21 15:10:20 -07:00
Xinyuan Tong
e8449ab515 Add deepseek v3.1 thinking parser support and update docs (#9464)
Signed-off-by: Xinyuan Tong <xinyuantong.cs@gmail.com>
2025-08-21 15:09:40 -07:00
Yineng Zhang
4746aaea41 fix: support fb fp8 (#9462) 2025-08-21 14:31:43 -07:00
gongwei-130
10d34f74e2 fix: should return a invalid request response when schema missing (#9461) 2025-08-21 14:06:50 -07:00
gongwei-130
9ba7253094 accomendate reasoning_effort set in chat_template_kwargs (#9458) 2025-08-21 13:22:03 -07:00
Hongbo Xu
9c8e4f69c3 [5/n]decouple quantization implementation from vLLM dependency (#9454) 2025-08-21 12:52:07 -07:00