kanshan/TensorRT-LLMs
1
0
Fork 0
mirror of https://github.com/NVIDIA/TensorRT-LLM.git synced 2026-02-16 15:55:08 +08:00
Code Issues Actions 1 Packages Projects Releases Wiki Activity
5,110 Commits 29 Branches 68 Tags 2 GiB
ca499d600d
Commit Graph

420 Commits

Author SHA1 Message Date
Wanli Jiang
421eb9e39c
[None][feat] Optimize NemotronH model with elementwise and nvfp4 fusion (#11273) 
Signed-off-by: Wanli Jiang <35160485+Wanli-Jiang@users.noreply.github.com>
 2026-02-12 09:25:31 -05:00
William Zhang
ca9537e17c
[TRTLLM-10858][feat] Multi-image support for EPD disagg (#11264) 
* Why?

Prior to this commit, we only supported a single multimodal input for
E/P/D disaggregated serving.

* What?

This commit does a minor refactor of the multimodal embedding handles
that cross process boundaries to enable this.
Existing unit tests are updated accordingly to test this.

The `RequestOutput` has its `mm_embedding_handle` replaced in favor of
`disaggregated_params`, addressing a previous TODO.

Signed-off-by: William Zhang <133824995+2ez4bz@users.noreply.github.com>
 2026-02-11 20:50:00 -08:00
Harris Nover
2d5ebb3fe8
[None][chore] Merge residual+hidden into layer norm at the end of each NemotronH MTP, and remove a % operation (#11406) 
Signed-off-by: Harris Nover <249353502+hnover-nv@users.noreply.github.com>
 2026-02-11 12:01:36 -05:00
Iman Tabrizian
7d992972b2
[TRTLLM-10273][feat] Move MambaCacheManager from Python to C++ (#10540) 
Signed-off-by: Iman Tabrizian <10105175+tabrizian@users.noreply.github.com>
 2026-02-10 07:20:56 -08:00
Ziyi Xiong
e76b634251
[TRTLLM-10321][feat] Support different KV cache layout for one-model spec dec (#10502) 
Signed-off-by: ziyixiong-nv <219238287+ziyixiong-nv@users.noreply.github.com>
 2026-02-10 05:16:02 +08:00
Guiju Zhang
c37531c3f7 [TRTLLM-10669][fix] Fix Eagle3 draft model weight loading for throughput checkpoint (#11010) 
Signed-off-by: Guiju Zhang <7135567+cascade812@users.noreply.github.com>
Signed-off-by: Wangshanshan <30051912+dominicshanshan@users.noreply.github.com>
 2026-02-09 23:53:40 +08:00
William Zhang
abb8106c01 [https://nvbugs/5835925][fix] Add EPD disagg support for Qwen3 VL MoE (#10962) 
* Why?

Trying to instantiate a `MultimodalEncoder` for a Qwen3 VL MoE model
would fail during weight loading.

* What?

This commit fixes the bug, alongside:
- explicit, intentional support for EPD for Qwen3 VL MoE.
- extends EPD unit tests for Qwen3 VL MoE, albeit with dummy weights.
- unit tests for the weight mapper fixes.

Signed-off-by: William Zhang <133824995+2ez4bz@users.noreply.github.com>
Signed-off-by: Wangshanshan <30051912+dominicshanshan@users.noreply.github.com>
 2026-02-09 23:53:40 +08:00
yifeizhang-c
5521c7b7e7
[TRTLLM-9457][feat] Add cute dsl fp8 gemm for Blackwell (#10130) 
Added FP8 cute dsl gemm and batch gemm.

Signed-off-by: Yifei Zhang <219273404+yifeizhang-c@users.noreply.github.com>
 2026-02-06 09:49:30 +08:00
jthomson04
d778b26062
[None][fix] Reduce host memory usage during model loading (#11119) 
Signed-off-by: jthomson04 <jwillthomson19@gmail.com>
 2026-02-05 08:57:40 -08:00
Izzy Putterman
3ef8a4639b
[None][feat] Nemotron H: Eagle3 support (#11131) 
Signed-off-by: Izzy Putterman <iputterman@nvidia.com>
 2026-02-02 10:26:25 -05:00
Michal Guzek
fafc22e3d4 [https://nvbugs/5691730][fix] Have LoRa bf16 ckpts work with Llama 3.3-70B-fp8 (#9808) 
Signed-off-by: Michal Guzek <mguzek@nvidia.com>
Signed-off-by: Michal Guzek <moraxu@users.noreply.github.com>
Signed-off-by: Jin Li <59594262+liji-nv@users.noreply.github.com>
Co-authored-by: Jin Li <59594262+liji-nv@users.noreply.github.com>
Signed-off-by: Wangshanshan <30051912+dominicshanshan@users.noreply.github.com>
 2026-02-02 16:26:46 +08:00
Balaram Buddharaju
531f85dc9b
[None][feat] Perfect routing for Deepseek models (#11127) 
Signed-off-by: Balaram Buddharaju <169953907+brb-nv@users.noreply.github.com>
 2026-01-30 23:46:35 -05:00
Necofish
144b61715f
[None][fix] Add missing absolute pe in Qwen3-VL Vision Encoder (#11065) 
Signed-off-by: Necofish <liuxiangyang@mail.ustc.edu.cn>
 2026-01-30 09:59:36 +09:00
Balaram Buddharaju
c7a86f89de
[TRTLLM-10264][feat] Support attention DP + Helix CP (#10477) 
Signed-off-by: Balaram Buddharaju <169953907+brb-nv@users.noreply.github.com>
 2026-01-29 02:57:13 -05:00
sunnyqgg
ff0dd6076e
[TRTLLM-10062][feat] Enable MTP for Nemotron Super (#10754) 
Signed-off-by: qgai <qgai@nvidia.com>
 2026-01-26 11:23:26 -05:00
Mike Iovine
f02948d956 [https://nvbugs/5803813][fix] Fix llama 4 min latency (#10724) 
Signed-off-by: Mike Iovine <6158008+mikeiovine@users.noreply.github.com>
Signed-off-by: Wangshanshan <30051912+dominicshanshan@users.noreply.github.com>
 2026-01-25 18:12:21 +08:00
Venky
b3146d095d
[TRTC-122][feat] Eagle3 Specdec UX improvements (#10124) 
Signed-off-by: Venky Ganesh <23023424+venkywonka@users.noreply.github.com>
 2026-01-22 07:24:11 -08:00
Taylor Yeonbok Lee
895bb94b3d
[#8241][feat] Support model_kwargs for pytorch backend (#10351) 
Signed-off-by: Taylor Yeonbok Lee <249374542+taylor-yb-lee@users.noreply.github.com>
 2026-01-21 20:51:38 -08:00
Yechan Kim
70caa779a4
[None][feat] K-EXAONE MTP support (#10796) 
Signed-off-by: yechank <161688079+yechank-nvidia@users.noreply.github.com>
 2026-01-22 13:43:00 +09:00
Daniil
0434db5bf7
[None][feat] GLM-4.5-Air support (#10653) 
Signed-off-by: Daniil Kulko <kulkodaniil@gmail.com>
 2026-01-22 11:42:09 +08:00
benzh-2025
4c8468c5d3
[None][fix] default disable gemm+allreduce fusion (#10656) 2026-01-20 12:31:17 +08:00
Anish Shanbhag
faa80e73fd
[None][feat] Auto download speculative models from HF for pytorch backend, add speculative_model field alias (#10099) 
Signed-off-by: Anish Shanbhag <ashanbhag@nvidia.com>
 2026-01-14 21:06:07 -08:00
彭晋韬(jtao peng)
211c44b951
[None][feat] Adding torch ext API for FusedAddRMSNormQuant kernel (#9905) 
Signed-off-by: jintaop <jintaop@nvidia.com>
 2026-01-15 07:29:15 +08:00
jmydurant
e7882d5c74
[None][feat] MiniMax M2 support (#10532) 
Signed-off-by: Mingyang Jiang <13463932+jmydurant@users.noreply.github.com>
 2026-01-14 17:38:58 +08:00
Yuxian Qiu
2acd03030a
[https://nvbugs/5781589][fix] Implement pp skip forward for all spec workers. (#10578) 
Signed-off-by: Yuxian Qiu <142763828+yuxianq@users.noreply.github.com>
 2026-01-14 09:36:35 +08:00
benzh-2025
6df2c8a074
[None][feat] add fp4 gemm + allreduce (#9729) 
Signed-off-by: benzh 
Signed-off-by: benzh-2025
 2026-01-13 21:11:13 +08:00
Tailing Yuan
38296a472b
[None][feat] Layer-wise benchmarks: make model init more general and support weights loading (#10562) 
Signed-off-by: Tailing Yuan <yuantailing@gmail.com>
 2026-01-13 19:17:03 +08:00
Guoming Zhang
bdaee87895
[TRTLLM-10060][feat] Enable attention dp for Nemotron Super v3. (#10347) 
Signed-off-by: nv-guomingz <137257613+nv-guomingz@users.noreply.github.com>
 2026-01-13 17:13:55 +08:00
Yechan Kim
8e0d20d901
[TRTLLM-10195][feat] K-EXAONE support (#10355) 
Signed-off-by: Jaedeok Kim <jaedeokk@nvidia.com>
Signed-off-by: yechank <161688079+yechank-nvidia@users.noreply.github.com>
Co-authored-by: Jaedeok Kim <jaedeokk@nvidia.com>
 2026-01-12 00:29:51 +09:00
Yechan Kim
7295af68ba
[None][fix] Enable AttentionDP on Qwen3-VL and fix test (#10435) 
Signed-off-by: yechank <161688079+yechank-nvidia@users.noreply.github.com>
 2026-01-10 00:13:26 +09:00
Yuxian Qiu
afa55c12b6
[None][fix] revert https://github.com/NVIDIA/TensorRT-LLM/pull/10445. (#10547) 
Signed-off-by: Yuxian Qiu <142763828+yuxianq@users.noreply.github.com>
 2026-01-08 21:50:04 -05:00
William Zhang
c0ae6bbdbe
[None][feat] EPD for Qwen3 VL (#10470) 
* Why?

We would like to support EPD disaggregated serving for Qwen3 VL.

* What?

This commit adds such support, and extends existing unit tests for
correctness checks.

Some minor (protected) interface changes had to be made to the
weight mapper as a side-effect.

Signed-off-by: William Zhang <133824995+2ez4bz@users.noreply.github.com>
 2026-01-08 06:45:54 -05:00
bhsueh_NV
bea61bb17d
[None][fix] Mistral large 3 few code refine (#10405) 
Signed-off-by: bhsueh <11360707+byshiue@users.noreply.github.com>
 2026-01-08 06:38:49 -05:00
Ziyi Xiong
7187afe7b9
[https://nvbugs/5781589][fix] Skip spec dec for non-last rank (#10445) 
Signed-off-by: ziyixiong-nv <219238287+ziyixiong-nv@users.noreply.github.com>
 2026-01-07 13:55:45 -05:00
Kanghwan
dc32bac9fc
[#4745][fix] Pass lora_params through Qwen2/3 model forward (#10174) 
Signed-off-by: Kanghwan Jang <861393+karljang@users.noreply.github.com>
 2026-01-07 15:30:17 +08:00
Xiao Xuan
46f035befe
[#2511][fix] eagle: qwen2 capture hidden states (#10091) 
Signed-off-by: SpicyNoodle <522169030@qq.com>
 2026-01-05 21:46:41 -05:00
Yukun He
d272f1a9bc
[TRTLLM-8821][feat] Apply AutoTuner to AllReduce Op for strategy tuning. (#8531) 
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
 2026-01-05 15:44:37 +08:00
Wanli Jiang
da0830670a
[TRTLLM-10065][feat] Add accuracy tests for super-v3 with multiple-gpus (#10234) 
Signed-off-by: Wanli Jiang <35160485+Wanli-Jiang@users.noreply.github.com>
 2026-01-05 09:41:49 +08:00
bhsueh_NV
0517b62789
[https://nvbugs/5772363][fix] fix bug of Mistral-Small-3.1-24B-Instruct-2503 (#10394) 
Signed-off-by: bhsueh <11360707+byshiue@users.noreply.github.com>
 2026-01-05 09:04:13 +08:00
dongfengy
afc533193d
[None][feat] Support nvfp4 for gptoss (#8956) 
Signed-off-by: Dongfeng Yu <dongfengy@nvidia.com>
 2026-01-04 08:57:44 -05:00
Izzy Putterman
bdf6953ddc
[None][feat] Eagle: MLA Based Eagle (#9677) 
Signed-off-by: Izzy Putterman <iputterman@nvidia.com>
 2026-01-02 13:45:07 -05:00
Necofish
73870ae4ad
[None][feat] support Qwen3-VL dense model in pytorch backend (#9060) 
Signed-off-by: Nekofish-L <liuxiangyang@mail.ustc.edu.cn>
 2025-12-31 17:54:26 +09:00
binghanc
692d8f2023
[TRTLLM-9455][feat] support for new checkpoint (#10082) 
Signed-off-by: binghanc <176802681+binghanc@users.noreply.github.com>
Co-authored-by: coderabbitai[bot] <136622811+coderabbitai[bot]@users.noreply.github.com>
 2025-12-30 14:46:39 +08:00
Guoming Zhang
1865020b6f
[TRTLLM-8577][feat] Clean the Qwen3-next code by removing Qwen3NextCo… (#10228) 
Signed-off-by: nv-guomingz <137257613+nv-guomingz@users.noreply.github.com>
 2025-12-27 22:49:55 +08:00
Wanli Jiang
14554ab3f3
[None][feat] Support multi-gpu running for nemotron-v3-nano and super (#10118) 
Signed-off-by: Wanli Jiang <35160485+Wanli-Jiang@users.noreply.github.com>
 2025-12-26 11:23:14 +08:00
bhsueh_NV
db3430f589
[None][feat] Support VLM part for Mistral Large 3 (#10188) 
Signed-off-by: bhsueh <11360707+byshiue@users.noreply.github.com>
 2025-12-25 11:20:58 -05:00
Necofish
8614cd3439
[None][fix] fix: resolve GPU memory imbalance in concurrent weight loading (#6472) 
Signed-off-by: Necofish <liuxiangyang@mail.ustc.edu.cn>
Signed-off-by: Nekofish-L <liuxiangyang@mail.ustc.edu.cn>
Signed-off-by: Jie Li <lijie@nvidia.com>
Co-authored-by: Jie Li <lijie@nvidia.com>
 2025-12-24 09:43:09 -05:00
Tailing Yuan
648196f8ae
[TRTLLM-9432][feat] Reduce synchronization and recompilation for qwen3-next (#9691) 
Signed-off-by: Tailing Yuan <yuantailing@gmail.com>
 2025-12-23 10:14:29 +08:00
William Zhang
a6a88985cf
[TRTLLM-9409][feat] Pass MRoPE tensors for EPD disagg (#9758) 
* Why?

Certain VLMs like the Qwen family need more than just the multimodal
embeddings in the language model, and need MRoPE position IDs and
deltas. Prior to this commit, only the embeddings could be communicated
from the encoder worker to the prefill worker.

* What?

This commit extends the `DisaggregatedParams` to include the MRoPE
information. It also adjusts several pieces of code required to
communicate that between E, P and D workers.

Closes TRTLLM-9409.

Signed-off-by: William Zhang <133824995+2ez4bz@users.noreply.github.com>
 2025-12-22 06:32:49 -05:00
bhsueh_NV
cd4b4f43fa
[None][feat] Support Eagle3 on Mistral Large3 (#9971) 
Signed-off-by: bhsueh <11360707+byshiue@users.noreply.github.com>
 2025-12-21 10:25:45 -05:00