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
2,563 Commits 25 Branches 59 Tags 2.1 GiB
b4d41d6604
Commit Graph

1113 Commits

Author SHA1 Message Date
brb-nv
0253036a4e [None][chore] Add docs for Gemma3 VLMs (#6880) 
Signed-off-by: Balaram Buddharaju <169953907+brb-nv@users.noreply.github.com>
Signed-off-by: Wangshanshan <30051912+dominicshanshan@users.noreply.github.com>
 2025-09-01 11:02:31 +08:00
Yukun He
e106045fda [None][fix] Complete the last missing allreduce op in Llama3/4. (#6850) 
The allreduce op of the last decoder layer is missing in some circumstances for the models Llama3 and Llama4.

Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
Signed-off-by: Wangshanshan <30051912+dominicshanshan@users.noreply.github.com>
 2025-09-01 11:02:31 +08:00
Anurag Mukkara
b821883b25 [None][fix] Revert phi4-mm aggregate mode (#6907) 
Signed-off-by: Anurag Mukkara <134339030+amukkara@users.noreply.github.com>
Signed-off-by: Wangshanshan <30051912+dominicshanshan@users.noreply.github.com>
 2025-09-01 11:02:31 +08:00
2ez4bz
cf0c47ca2d [None][fix] Fix batching bug in Mistral3 model (#6841) 
Prior to this commit, if multiple requests with images were in the same
batch, the batching logic for the images would fail.

This commit fixes it, and adds unit tests for it that were verified to
fail prior to the fix.

Signed-off-by: William Zhang <133824995+2ez4bz@users.noreply.github.com>
Signed-off-by: Wangshanshan <30051912+dominicshanshan@users.noreply.github.com>
 2025-09-01 11:02:31 +08:00
2ez4bz
2480aedb73 [TRTLLM-5252][feat] Add fp8 support for Mistral Small 3.1 (#6731) 
This commit adds some level of FP8 support to Mistral Small 3.1 by:

* disabling quantization for the vision sub-model since `modelopt` does
  support quantizing it (yet).
* extending existing accuracy tests to use a modelopt produced FP8
  checkpoint.

Signed-off-by: William Zhang <133824995+2ez4bz@users.noreply.github.com>
Signed-off-by: Wangshanshan <30051912+dominicshanshan@users.noreply.github.com>
 2025-09-01 11:02:31 +08:00
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
Zongfei Jing
a7ed26dd8b
[TRTLLM-6747][feat] Merge add sparse exp and shared exp into local reduction (#7369) 
Signed-off-by: Zongfei Jing <20381269+zongfeijing@users.noreply.github.com>
 2025-08-31 21:20:00 -04:00
Yiqing Yan
ec595a8e29
[None][chore] Bump version to 1.1.0rc2 (#7394) 
Signed-off-by: Yiqing Yan <yiqingy@nvidia.com>
 2025-08-31 10:20:38 +08:00
Zhongdongming Dai
9bb0c9500e
[None][docs] Update Dynasor paper info (#7137) 
Signed-off-by: Zhongdongming Dai <zhongdongmin@nvidia.com>
 2025-08-29 18:47:47 -07:00
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
Chang Liu
31b0f0fb0c
[https://nvbugs/5445466][fix] Eliminate race when loading HF dynamic modules (#7268) 
Signed-off-by: Chang Liu (Enterprise Products) <9713593+chang-l@users.noreply.github.com>
 2025-08-29 12:36:30 +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
Shiyu Li
b093d94d34
[https://nvbugs/5445466][fix] Bypass MLP TP split for MNNVL in DeepSeek V3 to avoid hanging. (#6886) 
Signed-off-by: Shiyu Li <shili@nvidia.com>
 2025-08-28 15:17:48 -07:00
dongfengy
367ff88a5e
[None][feat] Refactor llama4 for multimodal encoder IFB (#6844) 
Signed-off-by: Dongfeng Yu <dongfengy@nvidia.com>
 2025-08-28 13:22:19 -07:00
Nikita Korobov
a419b77fb5
[None][fix] mxfp4 padding bug for TRT-LLM and CUTLASS MoE backends (#7214) 
Signed-off-by: Nikita Korobov <14355239+nekorobov@users.noreply.github.com>
 2025-08-28 10:08:05 -07:00
Zongfei Jing
53163bf1df
[TRTLLM-6876][feat] Add low precision all2all for mnnvl (#7155) 
Signed-off-by: Zongfei Jing <20381269+zongfeijing@users.noreply.github.com>
 2025-08-28 18:26:16 +08:00
Pengyun Lin
c1e7fb9042
[TRTLLM-7207][feat] Chat completions API for gpt-oss (#7261) 
Signed-off-by: Pengyun Lin <81065165+LinPoly@users.noreply.github.com>
 2025-08-28 10:22:06 +08:00
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
dongxuy04
abdb2735be
[None][fix] Fix possible hang issue in WideEP and move some tests to pre-merge (#7262) 
Signed-off-by: Dongxu Yang <78518666+dongxuy04@users.noreply.github.com>
 2025-08-27 01:39:24 -04:00
Yukun He
bed5bc9f2e
[None][chore] Wrap the swiglu into custom op to avoid redundant device copy. (#7021) 
A redundant D2D copy is observed when enabling torch.compile for the Llama model due to the swiglu triton kernel, which brings perf overhead. Use a custom op to wrap the swiglu op to avoid this overhead.

Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
 2025-08-27 13:02:10 +08:00
Iman Tabrizian
bc84758626
[None][feat] Add logging for OAI disagg server (#7232) 2025-08-26 21:02:03 -07:00
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
Fanrong Li
e12868bc00
[None][fix] Remove and fuse some element-wise ops in the ds-r1-fp8 model (#7238) 
Signed-off-by: Fanrong Li <23290157+lfr-0531@users.noreply.github.com>
 2025-08-27 10:35:38 +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
Fridah-nv
0f947c64cb
[None][doc] Update autodeploy README.md, deprecate lm_eval in examples folder (#7233) 
Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
 2025-08-26 10:47:57 -07:00
Frank
78ecfbb4a4
[None][fix] Fix data type of KV Cache percentage in bench. (#7230) 
Signed-off-by: Frank Di Natale <3429989+FrankD412@users.noreply.github.com>
 2025-08-26 12:28:09 -04:00
Void
040f4c70d3
[None][perf] Accelerate global scale calculations for deepEP fp4 combine (#7126) 
Signed-off-by: Yilin Zhang <18275976+yilin-void@users.noreply.github.com>
 2025-08-27 00:13:13 +08:00
Zheng Duan
cf50ba2980
[TRTLLM-6549][feat] add perf metrics endpoint to openai server and openai disagg server (#6985) 
Signed-off-by: zhengd-nv <200704041+zhengd-nv@users.noreply.github.com>
 2025-08-26 15:34:44 +08:00
qixiang-99
b165f8bc97
fix/improve kvcache allocation in PyTorch runtime (#5933) 
Signed-off-by: qixiang-99 <203170375+qixiang-99@users.noreply.github.com>
 2025-08-26 12:40:22 +08:00
William Zhang
92576488d3
[None][feat] Skip prefetching consolidated safetensors when appropriate (#7013) 
* Why?

Some models (e.g. anything produced by Mistral) can have both sharded
safetensors and a consolidated safetensor in the same checkpoint
directory. In such cases, prefetching both to memory is a waste of time,
and memory.

* What?

This commit skips over consolidated safetensors when they are not the
only safetensor file present in the checkpoint directory

Signed-off-by: William Zhang <133824995+2ez4bz@users.noreply.github.com>
 2025-08-25 23:56:21 -04:00
Leslie Fang
20922b7d1f
[None][chore] Create PyExecutor from TorchLlmArgs Part 1 (#7105) 
Signed-off-by: leslie-fang25 <leslief@nvidia.com>
 2025-08-26 10:42:01 +08:00
Grzegorz Kwasniewski
2101d46d68
[TRTLLM-6342][feat] TP Sharding read from the model config (#6972) 
Signed-off-by: greg-kwasniewski1 <213329731+greg-kwasniewski1@users.noreply.github.com>
Co-authored-by: Suyog Gupta <41447211+suyoggupta@users.noreply.github.com>
 2025-08-25 15:41:27 -07:00
Lucas Liebenwein
97d550b4ba
[None] [AutoDeploy] canonicalize_graph before shape prop for consistent state_dict (#7223) 
Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>
 2025-08-25 16:59:57 -04:00
Bo Li
bf1b958f1a
[TRTLLM-7319][perf] Fuse slicing into MoE. (#6728) 
Signed-off-by: Bo Li <22713281+bobboli@users.noreply.github.com>
Signed-off-by: Sergey Klevtsov <sklevtsov@nvidia.com>
Co-authored-by: Sergey Klevtsov <sklevtsov@nvidia.com>
 2025-08-25 16:52:30 -04:00
Daniel Cámpora
e8e7e52892
[None][chore] Refactored the handle logits pp communication (#7154) 
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
 2025-08-25 16:14:08 -04:00
Frank
788fc62d23
[None][fix] Update to pull LLM from a central location. (#6458) 
Signed-off-by: Frank Di Natale <3429989+FrankD412@users.noreply.github.com>
 2025-08-25 13:07:29 -07:00
QI JUN
bea5e07fb7
[None][refactor] refactor the CUDA graph runner to manage all CUDA graphs (#6846) 
Signed-off-by: junq <22017000+QiJune@users.noreply.github.com>
 2025-08-25 20:52:05 +08:00
shaharmor98
b32e00e9fd
[None][chore] remove CLI support for mamba cache dtype setting (#7119) 
Signed-off-by: Shahar Mor <17088876+shaharmor98@users.noreply.github.com>
 2025-08-25 08:08:51 -04:00
amitz-nv
a1e03af0f4
[TRTLLM-7346][fix] Improve performance of PyTorchModelEngine._get_lora_params_from_requests (#7033) 
Signed-off-by: Amit Zuker <203509407+amitz-nv@users.noreply.github.com>
 2025-08-25 10:37:40 +03:00
Enwei Zhu
be6d92f09f
[None][fix] Fix MoE load balancer config loading (#7150) 
Signed-off-by: Enwei Zhu <21126786+syuoni@users.noreply.github.com>
 2025-08-25 01:42:54 -04:00
Yukun He
9c5b464fe0
[None][feat] Apply AutoTuner to fp8_block_scale_deep_gemm to trigger JIT ahead of time. (#7113) 
Because deep_gemm.gp8_gemm_nt will trigger many JIT processes during the inference phase, we need to sweep these shapes ahead of time. Apply the AutoTuner framework to achieve this and retain the potential capability to tune the swap_ab flag.

Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
 2025-08-25 10:48:31 +08:00
ajrasane
068056677f
[None][chore] Enable auto deploy accuracy test in CI (#7179) 
Signed-off-by: ajrasane <131806219+ajrasane@users.noreply.github.com>
Signed-off-by: Suyog Gupta <41447211+suyoggupta@users.noreply.github.com>
Co-authored-by: Suyog Gupta <41447211+suyoggupta@users.noreply.github.com>
 2025-08-24 08:42:30 -07:00
dongxuy04
19a0ea363b
[TRTLLM-6743][feat] Optimize and refactor alltoall in WideEP (#6973) 
Signed-off-by: Dongxu Yang <78518666+dongxuy04@users.noreply.github.com>
Signed-off-by: Fred Wei <20514172+WeiHaocheng@users.noreply.github.com>
Signed-off-by: Dongxu Yang <dongxuy@nvidia.com>
Co-authored-by: Fred Wei <20514172+WeiHaocheng@users.noreply.github.com>
 2025-08-24 08:15:29 -04:00
amitz-nv
35e0ae484a
[https://nvbugs/5467232][fix] Fix load_torch_hf_lora to override lora_config.trtllm_modules_to_hf_modules with default only when it has no value (#7132) 
Signed-off-by: Amit Zuker <203509407+amitz-nv@users.noreply.github.com>
 2025-08-24 15:00:24 +03:00
Grace Ho
3d54a1a521
[None] [feat] nsys profile output kernel classifier (#7020) 
Signed-off-by: Grace Ho <grho@nvidia.com>
 2025-08-23 00:57:37 -04:00
Frank
81fd468fec
[None][fix] Correct KV cache percentage report out. (#7102) 
Signed-off-by: Frank Di Natale <3429989+FrankD412@users.noreply.github.com>
 2025-08-22 10:28:57 -07:00
Izzy Putterman
b36460d7b5
[None][feat] Deepseek: Start Eagle work (#6210) 
Signed-off-by: Izzy Putterman <iputterman@nvidia.com>
Co-authored-by: Mike Iovine <miovine@nvidia.com>
 2025-08-22 12:57:17 -04:00
tomeras91
c232ba8157
[TRTLLM-4921][feat] Enable chunked prefill for Nemotron-H (#6334) 
Signed-off-by: Tomer Asida <57313761+tomeras91@users.noreply.github.com>
Signed-off-by: tomeras91 <57313761+tomeras91@users.noreply.github.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Co-authored-by: coderabbitai[bot] <136622811+coderabbitai[bot]@users.noreply.github.com>
 2025-08-22 12:15:20 -04:00
Suyog Gupta
e3de5758a3
[#7136][feat] trtllm-serve + autodeploy integration (#7141) 
Signed-off-by: Suyog Gupta <41447211+suyoggupta@users.noreply.github.com>
 2025-08-22 08:30:53 -07:00
Yiqing Yan
907bc22fcb
[None][chore] Bump version to 1.1.0rc2 (#7167) 
Signed-off-by: Yiqing Yan <yiqingy@nvidia.com>
 2025-08-22 22:02:28 +08:00