kanshan/TensorRT-LLMs
1
0
Fork 0
mirror of https://github.com/NVIDIA/TensorRT-LLM.git synced 2026-01-14 06:27:45 +08:00
Code Issues Actions 1 Packages Projects Releases Wiki Activity
2,884 Commits 25 Branches 59 Tags 2.1 GiB
2a30f11d63
Commit Graph

417 Commits

Author SHA1 Message Date
Yechan Kim
f77aca9f2c
[TRTLLM-7385][feat] Optimize Qwen2/2.5-VL performance (#7250) 
Signed-off-by: yechank <161688079+yechank-nvidia@users.noreply.github.com>
 2025-09-22 03:40:02 -07:00
HuiGao-NV
0dac1ddb74
[https://nvbugs/5525849][fix] Cherry-pick to fix mismatch of max seq len between kv cache manager and dummy requests (#7855) 
Signed-off-by: Hui Gao <huig@nvidia.com>
 2025-09-22 18:07:47 +08:00
HuiGao-NV
af34c9713a [https://nvbugs/5474169][fix] seq_len mismatch between kv cache manager and graph attn metadata (#7606) 
Signed-off-by: Hui Gao <huig@nvidia.com>
Signed-off-by: Wangshanshan <30051912+dominicshanshan@users.noreply.github.com>
 2025-09-22 14:28:38 +08:00
HuiGao-NV
123f5cbbf0 [https://nvbugs/5474169][fix]Adjust max seq len for kvcache for memory estimation (#7391) 
Signed-off-by: Hui Gao <huig@nvidia.com>
Signed-off-by: Wangshanshan <30051912+dominicshanshan@users.noreply.github.com>
 2025-09-22 14:28:38 +08:00
Stefan Niebler
8aead224fb
[https://nvbugs/5513423][fix] Correctly respect min_tokens in PyTorch Workflow (#7808) 
Signed-off-by: Stefan Niebler <82932102+stnie@users.noreply.github.com>
Co-authored-by: Daniel Cámpora <961215+dcampora@users.noreply.github.com>
 2025-09-21 22:15:18 -07:00
Enwei Zhu
639d4109a7
[None][fix] Disable torch.compile for CapturableGuidedDecoder (#7871) 
Signed-off-by: Enwei Zhu <21126786+syuoni@users.noreply.github.com>
 2025-09-22 10:04:30 +08:00
Ziyi Xiong
897c4dd23b
[https://nvbugs/5517404][fix] Use the correct cuda graph for dynamic spec dec (#7728) 
Signed-off-by: ziyixiong-nv <219238287+ziyixiong-nv@users.noreply.github.com>
 2025-09-21 08:20:48 +08:00
Yuxian Qiu
d6ebcf7c4a
[TRTLLM-6994][feat] FP8 Context MLA integration (Cherry-pick https://github.com/NVIDIA/TensorRT-LLM/pull/6059 from release/1.1.0rc2) (#7610) 
Signed-off-by: Yuxian Qiu <142763828+yuxianq@users.noreply.github.com>
 2025-09-19 09:40:49 +08:00
Ziyi Xiong
420f0fbcf5
[https://nvbugs/5522851][fix] Correct the logic to update kv_lens_cuda (#7790) 
Signed-off-by: ziyixiong-nv <219238287+ziyixiong-nv@users.noreply.github.com>
 2025-09-19 08:11:29 +08:00
sunnyqgg
80dd8fe197
[TRTLLM-6746][feat] Enable two-model spec dec for MTP Eagle (#7001) 
Signed-off-by: qgai <qgai@nvidia.com>
 2025-09-18 12:05:36 -04:00
Leslie Fang
870cfcf9a0
[None][chore] Remove executor config in create_py_executor (#7599) 
Signed-off-by: leslie-fang25 <leslief@nvidia.com>
 2025-09-18 14:24:58 +08:00
Yukun He
cd80e0a7f1
[None][fix] Make tile_tokens_dim calculation just in time before kernel launching. (#7529) 
tile_tokens_dim directly depends on the num_token, which is a dynamic shape during tuning and inference. When AutoTuner prepares dummy tensors with different num_tokens, it does not update the value of tile_tokens_dim automatically. Therefore, the value stored in the AutoTuner cache is misaligned, which will introduce a lot of cache misses during inference, which hurts perf a lot.

To avoid this issue, we move the calculation of tile_tokens_dim right before kernel launching, so that the value of tile_tokens_dim is always up to date with the num_tokens of the current input tensor used for the kernel runner.

Also, the tile_tokens_dim is calculated based on the number of tokens of a tuned bucket, instead of the original token number. Because we only tune the value for the buckets, not for the raw input token number, to avoid unexpected misalignment between tile_tokens_dim and the token number.

This PR also removes the warmup requests with the extra input shapes, which are triggered in the CUDA graph warmup phase.

Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
 2025-09-18 10:58:52 +08:00
Netanel Haber
a5cfc8368f
[https://nvbugs/5508536][fix] Revert #7041: Move stop_criteria to sample_async (#7041) (#7796) 
Signed-off-by: Netanel Haber <nhaber@nvidia.com>
Signed-off-by: Netanel Haber <58652339+netanel-haber@users.noreply.github.com>
Co-authored-by: Mike Iovine <miovine@nvidia.com>
 2025-09-17 21:27:01 -04:00
HuiGao-NV
a49cfb3e68
[https://nvbugs/5516666][fix] cherrypick fix to the CUDA graph warmup issue when using speculative decoding (#7737) 
Signed-off-by: Fanrong Li <23290157+lfr-0531@users.noreply.github.com>
Co-authored-by: Fanrong Li <23290157+lfr-0531@users.noreply.github.com>
Co-authored-by: Tao Li @ NVIDIA <tali@nvidia.com>
Co-authored-by: Signed-off-by: Hui Gao <huig@nvidia.com>
 2025-09-17 06:24:20 +08:00
xiweny
c076a02b38
[TRTLLM-4629] [feat] Add support of CUDA13 and sm103 devices (#7568) 
Signed-off-by: Xiwen Yu <13230610+VALLIS-NERIA@users.noreply.github.com>
Signed-off-by: Tian Zheng <29906817+Tom-Zheng@users.noreply.github.com>
Signed-off-by: Daniel Stokes <dastokes@nvidia.com>
Signed-off-by: Zhanrui Sun <zhanruis@nvidia.com>
Signed-off-by: Xiwen Yu <xiweny@nvidia.com>
Signed-off-by: Jiagan Cheng <jiaganc@nvidia.com>
Signed-off-by: Yiqing Yan <yiqingy@nvidia.com>
Signed-off-by: Bo Deng <deemod@nvidia.com>
Signed-off-by: ZhanruiSunCh <184402041+ZhanruiSunCh@users.noreply.github.com>
Signed-off-by: xiweny <13230610+VALLIS-NERIA@users.noreply.github.com>
Co-authored-by: Tian Zheng <29906817+Tom-Zheng@users.noreply.github.com>
Co-authored-by: Daniel Stokes <dastokes@nvidia.com>
Co-authored-by: Zhanrui Sun <zhanruis@nvidia.com>
Co-authored-by: Jiagan Cheng <jiaganc@nvidia.com>
Co-authored-by: Yiqing Yan <yiqingy@nvidia.com>
Co-authored-by: Bo Deng <deemod@nvidia.com>
Co-authored-by: Zhanrui Sun <184402041+ZhanruiSunCh@users.noreply.github.com>
 2025-09-16 09:56:18 +08:00
Ziyi Xiong
536e8776cd
[TRTLLM-6668][feat] Enable overlap scheduler for two-model spec decoding (#7651) 
Signed-off-by: ziyixiong-nv <219238287+ziyixiong-nv@users.noreply.github.com>
 2025-09-16 07:33:44 +08:00
jmydurant
7deefb3d2b
[TRTLLM-7192][feat] optimize MLA chunked prefill && support fp8 mla chunked prefill (#7477) 
Signed-off-by: Mingyang Jiang <13463932+jmydurant@users.noreply.github.com>
 2025-09-15 21:43:49 +08:00
Zheng Duan
24fc1f9acf
[None][fix] using arrival time in llmapi when creating LlmRequest in pytorch workflow (#7553) 
Signed-off-by: zhengd-nv <200704041+zhengd-nv@users.noreply.github.com>
 2025-09-15 07:26:01 -04:00
DylanChen-NV
d5df0af017
[https://nvbugs/5467981][fix] Fix Qwen2.5-VL fails with cuda graph padding (#7122) 
Signed-off-by: Dylan Chen <191843203+DylanChen-NV@users.noreply.github.com>
 2025-09-15 15:02:34 +08:00
Chang Liu
47e37755a3
[TRTLLM-6903][feat] Support chunked prefill for multimodal models (#6843) 
Signed-off-by: Chang Liu (Enterprise Products) <9713593+chang-l@users.noreply.github.com>
 2025-09-14 20:10:10 -07:00
Leslie Fang
d219a4f225
[None][chore] remove executor config in kv cache creator (#7526) 
Signed-off-by: leslie-fang25 <leslief@nvidia.com>
 2025-09-10 21:14:44 +08:00
Zheyu Fu
c353ff342e
[None][feat] Make the should_use_spec_decode logic a bit smarter (#7112) 
Signed-off-by: Zheyu Fu <zheyuf@NVIDIA.com>
 2025-09-10 12:53:59 +08:00
Richard Huo
dcd110cfac
[None][chore] add TorchLlmArgs to the connector api (#7493) 
Signed-off-by: richardhuo-nv <rihuo@nvidia.com>
 2025-09-09 09:05:59 -04:00
Chuang Zhu
77657a1c12
[TRTLLM-7361][feat] KV cache transfer for uneven pp (#7117) 
Signed-off-by: Chuang Zhu <111838961+chuangz0@users.noreply.github.com>
 2025-09-08 13:37:46 -04:00
Leslie Fang
3e0073e86b
[None][chore] remove executor config in instantiate sampler (#7516) 
Signed-off-by: leslie-fang25 <leslief@nvidia.com>
 2025-09-08 09:02:40 -07:00
Yan Chunwei
205c3a144c
[None][chore] expose tokens_per_block into KvCacheConfig (#5911) 
Signed-off-by: Superjomn <328693+Superjomn@users.noreply.github.com>
Signed-off-by: Yan Chunwei <328693+Superjomn@users.noreply.github.com>
 2025-09-07 21:14:10 -04:00
Netanel Haber
0fee8cd028
[TRTLLM-7153] [feat] Move stop_criteria to sample_async (#7041) 
Signed-off-by: Netanel Haber <nhaber@nvidia.com>
 2025-09-07 17:36:49 +03:00
Raayan Dhar
bae9560e62
[https://nvbugs/5448767][fix] sync termination of requests across PP ranks (#7455) 
Signed-off-by: raayandhar <rdhar@nvidia.com>
Signed-off-by: Lizhi Zhou <1432185+reasonsolo@users.noreply.github.com>
Co-authored-by: Lizhi Zhou <1432185+reasonsolo@users.noreply.github.com>
 2025-09-07 08:45:49 -04:00
Chang Liu
99b98f1374
[TRTLLM-7440][fix] Split fused_input_embed to separate out host sync (#7280) 
Signed-off-by: Chang Liu (Enterprise Products) <9713593+chang-l@users.noreply.github.com>
 2025-09-06 23:11:39 -04:00
Chang Liu
23500b55c3
[TRTLLM-7398][feat] Support KV cache salting for secure KV cache reuse (#7106) 
Signed-off-by: Chang Liu (Enterprise Products) <9713593+chang-l@users.noreply.github.com>
Signed-off-by: Chang Liu <9713593+chang-l@users.noreply.github.com>
 2025-09-06 17:58:32 -04:00
QI JUN
12ecb864c2
[None][chore] share input_ids buffers among different cuda graphs (#7236) 
Signed-off-by: junq <22017000+QiJune@users.noreply.github.com>
 2025-09-06 17:49:42 -04:00
Leslie Fang
9eb3911470
[None][chore] Remove executor_config in create_py_executor_instance (#7463) 
Signed-off-by: leslie-fang25 <leslief@nvidia.com>
 2025-09-05 20:56:03 +08:00
Shunkangz
bddf183e15
[None][feat] Add Request specific exception (#6931) 
Signed-off-by: Shunkang <182541032+Shunkangz@users.noreply.github.co>
 2025-09-04 18:43:42 -04:00
Enwei Zhu
1745102e72
[TRTLLM-7027][feat] Fuse d2t to logitsBitmaskKernel and fix a race condition in one-model spec (#7481) 
Signed-off-by: Enwei Zhu <21126786+syuoni@users.noreply.github.com>
Co-authored-by: Jin Li <59594262+liji-nv@users.noreply.github.com>
 2025-09-04 23:30:14 +08:00
Izzy Putterman
26b133f3a7
[None][feat] MultiLayer Eagle (#7234) 
Signed-off-by: Izzy Putterman <iputterman@nvidia.com>
 2025-09-04 10:49:13 -04:00
jianweiwu
7090b286b2
[None][fix] fix hunyuan_moe init bug (#7502) 
Signed-off-by: sorenwu <sorenwu@tencent.com>
 2025-09-04 03:06:00 -04:00
Leslie Fang
bd9ba97d89
[None][chore] Remove two unused parameters in create_py_executor (#7458) 
Signed-off-by: leslie-fang25 <leslief@nvidia.com>
 2025-09-04 07:31:31 +08:00
Enwei Zhu
5ff3a65b23
[TRTLLM-7028][feat] Enable guided decoding with speculative decoding (part 2: one-model engine) (#6948) 
Signed-off-by: Enwei Zhu <21126786+syuoni@users.noreply.github.com>
 2025-09-03 15:16:11 -07:00
Anurag Mukkara
ae5136831f
[https://nvbugs/5472947][fix] wait on isend handles before reusing buffers (#7462) 
Signed-off-by: Anurag Mukkara <134339030+amukkara@users.noreply.github.com>
 2025-09-03 13:20:02 +05:30
YueWeng
9a4f60687f
[https://nvbugs/5480289][fix] release slot manager in mtp MTPHiddenStatesManager (#7340) 
Signed-off-by: Yue Weng <25103990+yweng0828@users.noreply.github.com>
 2025-09-02 19:37:51 -07:00
Jinyang Yuan
572551b586
[None][perf] Autotune TRT-LLM Gen MoE when using CUDA graphs (#7285) 
Signed-off-by: Jinyang Yuan <154768711+jinyangyuan-nvidia@users.noreply.github.com>
 2025-09-03 10:08:59 +08:00
Leslie Fang
42697ea32a
[None][chore] rm executor config in kv cache connector (#7372) 
Signed-off-by: leslie-fang25 <leslief@nvidia.com>
 2025-09-03 08:13:13 +08:00
Leslie Fang
e81c50dbd2
[None][chore] Use llm args in create_py_executor (#7239) 
Signed-off-by: leslie-fang25 <leslief@nvidia.com>
 2025-09-01 16:27:55 -07:00
Mike Iovine
b3c57a7042
[TRTLLM-7353][feat] Implement capturable drafting loops for speculation (#7100) 
Signed-off-by: Mike Iovine <6158008+mikeiovine@users.noreply.github.com>
 2025-09-01 14:37:44 -04:00
Tian Zheng
e257cb3533
[None][feat] Support NVFP4 KV Cache (#6244) 
Signed-off-by: Tian Zheng <29906817+Tom-Zheng@users.noreply.github.com>
 2025-09-01 09:24:52 +08:00
Fanrong Li
37a1bd810f
[https://nvbugs/5481385][fix] Fix max_seq_len in cuda graph warmup and intermediate_size in fused_moe_deepgemm (#7345) 
Signed-off-by: Fanrong Li <23290157+lfr-0531@users.noreply.github.com>
Co-authored-by: Tao Li @ NVIDIA <tali@nvidia.com>
 2025-08-29 17:00:43 +08:00
Richard Huo
ce580ce4f5
[None][feat] KV Cache Connector API (#7228) 
Signed-off-by: jthomson04 <jwillthomson19@gmail.com>
Signed-off-by: richardhuo-nv <rihuo@nvidia.com>
Co-authored-by: jthomson04 <jwillthomson19@gmail.com>
Co-authored-by: Iman Tabrizian <10105175+Tabrizian@users.noreply.github.com>
Co-authored-by: Sharan Chetlur <116769508+schetlur-nv@users.noreply.github.com>
 2025-08-28 23:09:27 -04:00
Mike Iovine
8b216135f0
[None][refactor] Move draft token padding out of Drafter (#7134) 
Signed-off-by: Mike Iovine <6158008+mikeiovine@users.noreply.github.com>
 2025-08-27 11:07:50 +02:00
Shunkangz
ff4047414b
[None][opt] Balance the request based on number of tokens in AttentionDP (#7183) 
Signed-off-by: Shunkang <182541032+Shunkangz@users.noreply.github.co>
Co-authored-by: Shunkang <182541032+Shunkangz@users.noreply.github.co>
 2025-08-27 11:16:12 +08:00
Jin Li
028235404b
[TRTLLM-6633][feat] Padding for piecewise cudagraph (#6750) 
Signed-off-by: Jin Li <59594262+liji-nv@users.noreply.github.com>
 2025-08-26 18:31:33 -04:00