kanshan/TensorRT-LLMs
1
0
Fork 0
mirror of https://github.com/NVIDIA/TensorRT-LLM.git synced 2026-01-30 07:33:48 +08:00
Code Issues Actions 1 Packages Projects Releases Wiki Activity
910 Commits 28 Branches 64 Tags 2 GiB
27bdd0c82d
Commit Graph

420 Commits

Author SHA1 Message Date
NVJiangShao
a6f2a1e918
Fix test_fused_moe_w4afp8 (#4393) 
Signed-off-by: Jiang Shao <91270701+StudyingShao@users.noreply.github.com>
 2025-05-16 17:21:33 +08:00
Daniel Cámpora
df19430629
chore: Mass Integration 0.19 (#4255) 
* fix: Fix/fused moe 0.19 (#3799)

* fix bug of stream init

Signed-off-by: bhsueh <11360707+byshiue@users.noreply.github.com>

* fix bug

Signed-off-by: bhsueh <11360707+byshiue@users.noreply.github.com>

---------

Signed-off-by: bhsueh <11360707+byshiue@users.noreply.github.com>

* fix: Add pre-download of checkpoint before benchmark. (#3772)

* Add pre-download of checkpoint before benchmark.

Signed-off-by: Frank Di Natale <3429989+FrankD412@users.noreply.github.com>

* Add missing remote code flag.

Signed-off-by: Frank Di Natale <3429989+FrankD412@users.noreply.github.com>

* Move from_pretrained to throughput benchmark.

Signed-off-by: Frank Di Natale <3429989+FrankD412@users.noreply.github.com>

* Move download and use snapshot_download.

Signed-off-by: Frank Di Natale <3429989+FrankD412@users.noreply.github.com>

* Removed trusted flag.

Signed-off-by: Frank Di Natale <3429989+FrankD412@users.noreply.github.com>

* Fix benchmark command in iteration log test.

Signed-off-by: Frank Di Natale <3429989+FrankD412@users.noreply.github.com>

---------

Signed-off-by: Frank Di Natale <3429989+FrankD412@users.noreply.github.com>

* [https://nvbugspro.nvidia.com/bug/5241495][fix] CUDA Graph padding with overlap scheduler (#3839)

* fix

Signed-off-by: Enwei Zhu <21126786+syuoni@users.noreply.github.com>

* fuse

Signed-off-by: Enwei Zhu <21126786+syuoni@users.noreply.github.com>

* fix

Signed-off-by: Enwei Zhu <21126786+syuoni@users.noreply.github.com>

* fix

Signed-off-by: Enwei Zhu <21126786+syuoni@users.noreply.github.com>

---------

Signed-off-by: Enwei Zhu <21126786+syuoni@users.noreply.github.com>

* TRTLLM-4875 feat: Add version switcher to doc (#3871)

Signed-off-by: Kaiyu Xie <26294424+kaiyux@users.noreply.github.com>

* waive a test (#3897)

Signed-off-by: Superjomn <328693+Superjomn@users.noreply.github.com>

* docs:fix https://nvbugs/5244616 by removing new invalid links. (#3939)

Signed-off-by: nv-guomingz <37257613+nv-guomingz@users.noreply.github.com>
Co-authored-by: nv-guomingz <37257613+nv-guomingz@users.noreply.github.com>

* fix: remote mpi session abort (#3884)

* fix remote mpi session

Signed-off-by: Superjomn <328693+Superjomn@users.noreply.github.com>

* fix

Signed-off-by: Superjomn <328693+Superjomn@users.noreply.github.com>

---------

Signed-off-by: Superjomn <328693+Superjomn@users.noreply.github.com>

* skip fp8 gemm for pre-hopper (#3931)

Signed-off-by: Ivy Zhang <25222398+crazydemo@users.noreply.github.com>

* [https://nvbugspro.nvidia.com/bug/5247148][fix] Attention DP with overlap scheduler (#3975)

* fix

Signed-off-by: Enwei Zhu <21126786+syuoni@users.noreply.github.com>

* update multigpu list

Signed-off-by: Enwei Zhu <21126786+syuoni@users.noreply.github.com>

* fix namings

Signed-off-by: Enwei Zhu <21126786+syuoni@users.noreply.github.com>

---------

Signed-off-by: Enwei Zhu <21126786+syuoni@users.noreply.github.com>

* Doc: Fix H200 DeepSeek R1 perf doc (#4006)

* fix doc

Signed-off-by: jiahanc <173873397+jiahanc@users.noreply.github.com>

* update perf number

Signed-off-by: jiahanc <173873397+jiahanc@users.noreply.github.com>

---------

Signed-off-by: jiahanc <173873397+jiahanc@users.noreply.github.com>

* Fix the perf regression caused by insufficient cache warmup. (#4042)

Force tuning up to 8192 sequence length for NVFP4 linear op. Also, make this runtime-selectable with UB enabled.

Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>

* doc: Update 0.19.0 release notes (#3976)

Signed-off-by: Kaiyu Xie <26294424+kaiyux@users.noreply.github.com>

* Optimize the AutoTuner cache access code to reduce host code overhead. (#4060)

The NVFP4 Linear op is very sensitive to the host overhead.
This PR introduces customizable `find_nearest_profile` and `get_cache_key_specifc`, which allow users to override the default method for generating the cache key.

Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>

* Update switcher (#4098)

Signed-off-by: Kaiyu Xie <26294424+kaiyux@users.noreply.github.com>

* doc: update release notes (#4108)

Signed-off-by: Kaiyu Xie <26294424+kaiyux@users.noreply.github.com>

* docs:update 0.19 doc. (#4120)

Signed-off-by: nv-guomingz <137257613+nv-guomingz@users.noreply.github.com>

* docs:add torch flow supported model list. (#4129)

Signed-off-by: nv-guomingz <137257613+nv-guomingz@users.noreply.github.com>

* doc: Release V0.19 Perf Overview Update (#4166)

Signed-off-by: zpatel <22306219+zbpatel@users.noreply.github.com>

* Fix readme of autodeploy.

Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>

* Update tensorrt_llm/_torch/pyexecutor/llm_request.py

Co-authored-by: Enwei Zhu <21126786+syuoni@users.noreply.github.com>
Signed-off-by: Daniel Cámpora <961215+dcampora@users.noreply.github.com>

* Revert mgmn worker node.

Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>

* Change to disable_overlap_scheduler.

Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>

---------

Signed-off-by: bhsueh <11360707+byshiue@users.noreply.github.com>
Signed-off-by: Frank Di Natale <3429989+FrankD412@users.noreply.github.com>
Signed-off-by: Enwei Zhu <21126786+syuoni@users.noreply.github.com>
Signed-off-by: Kaiyu Xie <26294424+kaiyux@users.noreply.github.com>
Signed-off-by: Superjomn <328693+Superjomn@users.noreply.github.com>
Signed-off-by: nv-guomingz <37257613+nv-guomingz@users.noreply.github.com>
Signed-off-by: Ivy Zhang <25222398+crazydemo@users.noreply.github.com>
Signed-off-by: jiahanc <173873397+jiahanc@users.noreply.github.com>
Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
Signed-off-by: nv-guomingz <137257613+nv-guomingz@users.noreply.github.com>
Signed-off-by: zpatel <22306219+zbpatel@users.noreply.github.com>
Signed-off-by: Daniel Campora <961215+dcampora@users.noreply.github.com>
Signed-off-by: Daniel Cámpora <961215+dcampora@users.noreply.github.com>
Co-authored-by: bhsueh_NV <11360707+byshiue@users.noreply.github.com>
Co-authored-by: Frank <3429989+FrankD412@users.noreply.github.com>
Co-authored-by: Enwei Zhu <21126786+syuoni@users.noreply.github.com>
Co-authored-by: Kaiyu Xie <26294424+kaiyux@users.noreply.github.com>
Co-authored-by: Yan Chunwei <328693+Superjomn@users.noreply.github.com>
Co-authored-by: nv-guomingz <137257613+nv-guomingz@users.noreply.github.com>
Co-authored-by: nv-guomingz <37257613+nv-guomingz@users.noreply.github.com>
Co-authored-by: Ivy Zhang <25222398+crazydemo@users.noreply.github.com>
Co-authored-by: jiahanc <173873397+jiahanc@users.noreply.github.com>
Co-authored-by: Yukun He <23156053+hyukn@users.noreply.github.com>
Co-authored-by: Zac Patel <22306219+zbpatel@users.noreply.github.com>
 2025-05-16 10:53:25 +02:00
Tracin
46c5a56444
Support dynamic per-tensor FP8 (#4250) 
* Support dynamic per-tensor FP8

Signed-off-by: Tracin <10434017+Tracin@users.noreply.github.com>

* Update test cases.

Signed-off-by: Tracin <10434017+Tracin@users.noreply.github.com>

---------

Signed-off-by: Tracin <10434017+Tracin@users.noreply.github.com>
 2025-05-16 13:33:58 +08:00
WeiHaocheng
54d28718c7
feat: support benchmark on scaffolding (#3328) (#4286) 
Signed-off-by: Fred Wei <20514172+WeiHaocheng@users.noreply.github.com>
 2025-05-16 12:28:49 +08:00
yuxianq
a1daa22970
doc: Add docstring for Attention and MLA module. (#4354) 
Signed-off-by: Yuxian Qiu <142763828+yuxianq@users.noreply.github.com>
Co-authored-by: hlu1 <14827759+hlu1@users.noreply.github.com>
 2025-05-16 09:37:04 +08:00
Suyog Gupta
b0f7522c82
[AutoDeploy]feat: Add an AutoDeploy compile backend that only calls torch.compile (#4240) 
* add a torch-compile backend

Signed-off-by: Suyog Gupta <41447211+suyoggupta@users.noreply.github.com>

* readme changes

Signed-off-by: Suyog Gupta <41447211+suyoggupta@users.noreply.github.com>

* plumb torch-compile through build_and_run_ad.py

Signed-off-by: Suyog Gupta <41447211+suyoggupta@users.noreply.github.com>

* plumb torch-compile through build_and_run_ad.py

Signed-off-by: Suyog Gupta <41447211+suyoggupta@users.noreply.github.com>

* plumb torch-compile through build_and_run_ad.py

Signed-off-by: Suyog Gupta <41447211+suyoggupta@users.noreply.github.com>

* add torch-cudagraph backend

Signed-off-by: Suyog Gupta <41447211+suyoggupta@users.noreply.github.com>

* update readme

Signed-off-by: Suyog Gupta <41447211+suyoggupta@users.noreply.github.com>

* update readme

Signed-off-by: Suyog Gupta <41447211+suyoggupta@users.noreply.github.com>

* further enhanced compiler backends

Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>

* further enhance readme

Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>

* better specified defaults in simple_config.py

Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>

* fix typo in simple_config.py

Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>

* updated deepseek-v3 support

Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>

* revert accidental deletion in AD Readme

Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>

---------

Signed-off-by: Suyog Gupta <41447211+suyoggupta@users.noreply.github.com>
Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>
Co-authored-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>
 2025-05-16 08:38:15 +08:00
rakib-hasan
25407249a5
[TRTLLM-5054][fix] Removing repeated loading of input processor (#4161) 
removing repeated loading of input processor

Signed-off-by: Rakib Hasan <rhasan@nvidia.com>
 2025-05-16 08:04:58 +08:00
Lucas Liebenwein
4883121477
[AutoDeploy] fix: disable overlap scheduler until supported (#4365) 
Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>
 2025-05-15 16:19:30 -07:00
Yechan Kim
c6e2111f4e
feat: enhance trtllm serve multimodal (#3757) 
* feat: enhance trtllm serve multimodal

1. made the load_image and load_video asynchronous
2. add image_encoded input support to be compatible with genai-perf
3. support text-only on multimodal mdoels(currently, Qwen2-VL & Qwen2.5-VL)

Signed-off-by: yechank <161688079+yechank-nvidia@users.noreply.github.com>

* add test

Signed-off-by: yechank <161688079+yechank-nvidia@users.noreply.github.com>

* fix bandit

Signed-off-by: yechank <161688079+yechank-nvidia@users.noreply.github.com>

* trimming uils

Signed-off-by: yechank <161688079+yechank-nvidia@users.noreply.github.com>

* trimming for test

Signed-off-by: yechank <161688079+yechank-nvidia@users.noreply.github.com>

* genai perf command fix

Signed-off-by: yechank <161688079+yechank-nvidia@users.noreply.github.com>

* command fix

Signed-off-by: yechank <161688079+yechank-nvidia@users.noreply.github.com>

* refactor chat_utils

Signed-off-by: yechank <161688079+yechank-nvidia@users.noreply.github.com>

* stress test genai-perf command

Signed-off-by: yechank <161688079+yechank-nvidia@users.noreply.github.com>

---------

Signed-off-by: yechank <161688079+yechank-nvidia@users.noreply.github.com>
 2025-05-15 16:16:31 -07:00
yuxianq
4f8afe4cc6
feat: [nvbugs/5261055][nvbugs/5170160] non-invasive pipeline parallelism (#4034) 
Signed-off-by: Yuxian Qiu <142763828+yuxianq@users.noreply.github.com>
 2025-05-16 04:16:53 +08:00
Venky
5ebe32f06f
enh: Enable option in trtllm-bench build subcommand to avoid loading weights (#4142) 
* expose load_format 

Signed-off-by: Venky <23023424+venkywonka@users.noreply.github.com>

* yapf

Signed-off-by: Venky <23023424+venkywonka@users.noreply.github.com>

---------

Signed-off-by: Venky <23023424+venkywonka@users.noreply.github.com>
Co-authored-by: Frank <3429989+FrankD412@users.noreply.github.com>
 2025-05-16 03:50:53 +08:00
yuxianq
0e87fcc228
refactor: use x is None instead of x == None. (#4244) 
Signed-off-by: Yuxian Qiu <142763828+yuxianq@users.noreply.github.com>
 2025-05-15 20:00:04 +08:00
ixlmar
4ee82fc0fd
chore: reduce code duplication (#4297) 
Signed-off-by: ixlmar <206748156+ixlmar@users.noreply.github.com>
 2025-05-15 09:25:37 +01:00
Zongfei Jing
f0ca60a95d
Add allreduce and rmsnorm fusion for qwen3 (#4304) 
Signed-off-by: Zongfei Jing <20381269+zongfeijing@users.noreply.github.com>
 2025-05-15 16:22:11 +08:00
zhhuang-nv
97bc680cd8
feat: support kv cache reuse for MLA (#3571) 
* support kv cache reuse for MLA

load compressed_kv and k_pe and do up-projection
use 192/128 head size MLA context kernel
support Blackwell and Hopper now

Signed-off-by: Zhen Huang <145532724+zhhuang-nv@users.noreply.github.com>

* add CI test

Signed-off-by: Zhen Huang <145532724+zhhuang-nv@users.noreply.github.com>

* fix: set k_pe head_num to 1 for kernel 2 and kernel 2V2

Signed-off-by: Mingyang Jiang <13463932+jmydurant@users.noreply.github.com>

* resolve comments

Signed-off-by: Zhen Huang <145532724+zhhuang-nv@users.noreply.github.com>

* use GPTJ style RoPE for MLA

Signed-off-by: Zhen Huang <145532724+zhhuang-nv@users.noreply.github.com>

* fix rebase error and some docs

Signed-off-by: Zhen Huang <145532724+zhhuang-nv@users.noreply.github.com>

* fix kv_lens

Signed-off-by: Zhen Huang <145532724+zhhuang-nv@users.noreply.github.com>

* tiny fix

Signed-off-by: Zhen Huang <145532724+zhhuang-nv@users.noreply.github.com>

* fix torch compile

Signed-off-by: Zhen Huang <145532724+zhhuang-nv@users.noreply.github.com>

* fix: use normal device memory instead of pinned memory for unit test

Signed-off-by: Mingyang Jiang <13463932+jmydurant@users.noreply.github.com>

* fix L0 tests

Signed-off-by: Zhen Huang <145532724+zhhuang-nv@users.noreply.github.com>

* fix torch compile after rebase

Signed-off-by: Zhen Huang <145532724+zhhuang-nv@users.noreply.github.com>

* resolve comments

Signed-off-by: Zhen Huang <145532724+zhhuang-nv@users.noreply.github.com>

* resolve comments again

Signed-off-by: Zhen Huang <145532724+zhhuang-nv@users.noreply.github.com>

---------

Signed-off-by: Zhen Huang <145532724+zhhuang-nv@users.noreply.github.com>
Signed-off-by: Mingyang Jiang <13463932+jmydurant@users.noreply.github.com>
Signed-off-by: zhhuang-nv <145532724+zhhuang-nv@users.noreply.github.com>
Co-authored-by: Mingyang Jiang <13463932+jmydurant@users.noreply.github.com>
 2025-05-15 15:22:21 +08:00
Kaiyu Xie
b4e5df0ee0
Breaking change: perf: Enable scheduling overlap by default (#4174) 
Signed-off-by: Kaiyu Xie <26294424+kaiyux@users.noreply.github.com>
 2025-05-15 14:27:36 +08:00
Fridah-nv
d008d6412f
feat:[AutoDeploy] Update MoE pattern matcher to drop expert selection logic (#3283) 
* update matcher to match expert compute first, then extract other args with LCA

Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>

* support 3D and 2D input in torch.ops.moe.trtllm_fused_moe

Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>

* update custom ops to support 3D and 2D inputs

Signed-off-by: Ubuntu <201670829+Fridah-nv@users.noreply.github.com>

* update deepseek patch

Signed-off-by: Ubuntu <201670829+Fridah-nv@users.noreply.github.com>

---------

Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
 2025-05-15 13:53:09 +08:00
nv-guomingz
e76cf9d9fe
fix:https://nvbugs/5234033 enable starcoder trt-flow with transforme… (#3909) 
fix:https://nvbugs/5234033 enable startcoder trt-flow with transformer 4.51.3.

Signed-off-by: nv-guomingz <137257613+nv-guomingz@users.noreply.github.com>
 2025-05-15 11:16:45 +08:00
Zeyu WANG
2681b26e48
[TRTLLM-2795] feat: Add yarn support for other models in trt-flow (#3840) 
Add yarn support for general models(e.g. llama, qwen) other than deepseek in trt-flow.

Signed-off-by: Zeyu Wang <zeyuw@nvidia.com>
 2025-05-15 11:03:57 +08:00
Mike Iovine
f9adac3dea
[feat] Enable chunked context for flashinfer (#4132) 
Signed-off-by: Mike Iovine <6158008+mikeiovine@users.noreply.github.com>
 2025-05-15 10:59:38 +08:00
QI JUN
498ce8a056
Revert "feat: Low Precision Allreduce for PCIe based GPU" (#4340) 
Revert "feat: Low Precision Allreduce for PCIe based GPU (#3851)"

This reverts commit 5e634dd1bd.
 2025-05-15 09:52:39 +08:00
sugunav14
7c828d767f
feat: [AutoDeploy] DSV3 mla attn ref op (#4272) 
* raw ref op + new patch untested

Signed-off-by: Suguna Velury <178320438+sugunav14@users.noreply.github.com>

* Added mla attn ref op and unit tests for attn + module patches

Signed-off-by: Suguna Velury <178320438+sugunav14@users.noreply.github.com>

* update stray changes in deepseek.py

Signed-off-by: Suguna Velury <178320438+sugunav14@users.noreply.github.com>

* Updated stale documentation

Signed-off-by: Suguna Velury <178320438+sugunav14@users.noreply.github.com>

* removed stray update in sdpa return shapes

Signed-off-by: Suguna Velury <178320438+sugunav14@users.noreply.github.com>

---------

Signed-off-by: Suguna Velury <178320438+sugunav14@users.noreply.github.com>
 2025-05-15 01:58:20 +08:00
HuiGao-NV
f4059c6e2e
Add test case for kv memory estimation (#4158) 
* Add test case for kv memory estimation
* Dump running log into file and parse kv cache memory size from file
* Set bigger peak memory size for mixed percision case and test_ptp_quickstart_advanced_eagle3 case
* Revert change to usage of fraction
* use context manager to guard temp files

Signed-off-by: Hui Gao <huig@nvidia.com>
 2025-05-14 18:39:25 +08:00
kanghui0204
5e634dd1bd
feat: Low Precision Allreduce for PCIe based GPU (#3851) 
This PR adds a customized allreduce to TensorRT-LLM. The new allreduce is used for communication on PCIe-based GPUs via low-precision quantization, which can accelerate the PCIe allreduce process.

Signed-off-by: Hui Kang <hkang@nvidia.com>
Co-authored-by: Hui Kang <hkang@nvidia.com>
 2025-05-14 16:45:43 +08:00
Barry Kang
20b42912ce
[TRTLLM-3330][feat] Support DeepSeek-R1 W4A8 on Hopper (#4123) 
Support DeepSeek-R1 W4A8 on Hopper

Co-authored-by: Barry Kang <43644113+Barry-Delaney@users.noreply.github.com>
Co-authored-by: Jiang Shao <91270701+StudyingShao@users.noreply.github.com>
Signed-off-by: Barry Kang <43644113+Barry-Delaney@users.noreply.github.com>
 2025-05-14 15:48:07 +08:00
brb-nv
8280c3d4f2
feat: Support Gemma3-1b-it in Pytorch workflow (#3999) 
Signed-off-by: Balaram Buddharaju <169953907+brb-nv@users.noreply.github.com>
 2025-05-14 14:02:44 +08:00
Fridah-nv
21dbd163a7
[TRTLLM-5188] fix: [AutoDeploy] unwaive AD build test (#4273) 
* unwaive small build test

Signed-off-by: Ubuntu <201670829+Fridah-nv@users.noreply.github.com>

* unwaive mutigpu/integration tests

Signed-off-by: Ubuntu <201670829+Fridah-nv@users.noreply.github.com>

* fix for torch.compile+flashinfer attention

Signed-off-by: Ubuntu <201670829+Fridah-nv@users.noreply.github.com>

---------

Signed-off-by: Ubuntu <201670829+Fridah-nv@users.noreply.github.com>
 2025-05-14 10:40:12 +08:00
Zhanrui Sun
23b9705bf4
chore: bump version to 0.20.0rc3 (#4261) 
Signed-off-by: ZhanruiSunCh <184402041+ZhanruiSunCh@users.noreply.github.com>
 2025-05-14 10:15:25 +08:00
Anurag Mukkara
b0a03a289c
fix: Merge PP overlap and non-overlap executor loop (#3878) 
Signed-off-by: Anurag Mukkara <134339030+amukkara@users.noreply.github.com>
 2025-05-14 06:04:36 +08:00
brb-nv
cd5b3d21a0
feat: Support Mistral Small 3.1 24B VLM in TRT workflow (#4183) 
Signed-off-by: Balaram Buddharaju <169953907+brb-nv@users.noreply.github.com>
 2025-05-14 03:47:22 +08:00
Frank
c0c3c7f68c
[TRTLLM-5233][feat]: Add chunking to PyT heuristic for trtllm-bench. (#4133) 
* Add chunking to PyT heuristic.

Signed-off-by: Frank Di Natale <3429989+FrankD412@users.noreply.github.com>

* Cast tokens and batch size to ints.

Signed-off-by: Frank Di Natale <3429989+FrankD412@users.noreply.github.com>

---------

Signed-off-by: Frank Di Natale <3429989+FrankD412@users.noreply.github.com>
 2025-05-13 21:47:06 +08:00
Yukun He
cbca6505ff
[nvbugs/5268808][fix] Fix the list-out-of-range access issue of AllReduce workspace on multi-node. (#4159) 
This issue is found for tp=ep=8 on the multi-node machine due to the inconsistent PP sizes.
* Reform the workspace allocation implementation to avoid the list-out-of-range issues.
* Disable min_latency_mode under the multi-node scenario to avoid the illegal memory access issue.

Signed-off-by: Yukun He <23156053+hyukn@users.noreply.github.com>
 2025-05-13 17:17:25 +08:00
Perkz Zheng
e8d7834c50
fix: [https://nvbugspro.nvidia.com/bug/5238626] illegal memory address when running llama 4 with cuda graph enabled (#4101) 
Signed-off-by: Perkz Zheng <67892460+PerkzZheng@users.noreply.github.com>
 2025-05-13 14:58:54 +08:00
v-shobhit
1770dd96d8
Fix Pipeline Parallelism in Llama4 (#4106) 
Signed-off-by: Shobhit Verma <shobhitv@nvidia.com>
 2025-05-12 22:54:37 -07:00
nvpohanh
13c8e5a8a8
feat: Prefetch safetensors files before loading them (#4140) 
Prefetching safetensors files so that they are stored in the system file
cache. This significantly speeds up the model weight loading for the
very first run after entering the docker container.

This is beneficial because model weight loading is done layer-by-layer,
which means reading from the safetensors chunk-by-chunk, and that cannot
utilize the internet bandwidth very well, assuming that these files are
stored in some network drives. Instead, loading the whole files in bulk
can achieve higher internet bandwidth utilization.

When running with world_size>1, all ranks collaboratedly prefetch these
files.

In theory, we should add heuristics to decide whether to prefetch the
files or not, but that is beyond the scope of this commit.

For example, when the CPU memory is small, doing prefetching may result
in file cache thrashing, resulting in slower weight loading time.

Signed-off-by: Po-Han Huang <pohanh@nvidia.com>
 2025-05-13 13:35:30 +08:00
pcastonguay
9643be5f20
[TRTLLM-5050][feat] Enable per-request stats with PyT backend (#4156) 
* feat: Add per-request stats support with PyT backend

Signed-off-by: Patrice Castonguay <55748270+pcastonguay@users.noreply.github.com>

* Adding unit test

Signed-off-by: Patrice Castonguay <55748270+pcastonguay@users.noreply.github.com>

* Fixing stats unit test

Signed-off-by: Patrice Castonguay <55748270+pcastonguay@users.noreply.github.com>

* Fixing test with overlap

Signed-off-by: Patrice Castonguay <55748270+pcastonguay@users.noreply.github.com>

---------

Signed-off-by: Patrice Castonguay <55748270+pcastonguay@users.noreply.github.com>
 2025-05-12 21:35:15 -04:00
Simeng Liu
286a789549
feat: Add heuristic for GroupRMSNorm kernel selection. (#4047) 
* feat: Add heuristic for GroupRMSNorm kernel selection.

Implements a logistic regression model to dynamically select between:
- GroupRMSNormBaseKernel: Allocates warps proportional to sum of dimensions
  (better SM occupancy in most cases)
- GroupRMSNormLargeBatch: Allocates warps proportional to max dimension
  (better block scheduling in large batch scenarios)

Selection heuristic considers batch size, allocated warps, and scheduling
efficiency on the current GPU architecture. Models for Compute Capability
9.x and 10.x are trained base on nsys kernel runtime data.
The default kernel selection is the base kernel.

The python operator group_rms_norm will use the heuristic by default.
User can pick to use the base or large batch kernels as well.

Signed-off-by: Simeng Liu <simengl@nvidia.com>

* Address the comments.

Signed-off-by: Simeng Liu <simengl@nvidia.com>

---------

Signed-off-by: Simeng Liu <simengl@nvidia.com>
 2025-05-13 08:52:53 +08:00
Erin
4becf32360
fix: reshape token_ids for lp in torch backend (#4239) 
reshape token_ids

Signed-off-by: Erin Ho <14718778+hchings@users.noreply.github.com>
 2025-05-13 08:43:47 +08:00
wili
eba3623a54
Feat: Variable-Beam-Width-Search (VBWS) part4 (#3979) 
* feat/vbws-part4-v1.8: rebase

Signed-off-by: wili-65535 <wili-65535@users.noreply.github.com>

* feat/vbws-part4-v1.9: fix incorrect output when using short output length

Signed-off-by: wili-65535 <wili-65535@users.noreply.github.com>

* v1.9.1: remove useless variables

Signed-off-by: wili-65535 <wili-65535@users.noreply.github.com>

* v1.9.2:fix incorrect output when using short output length

Signed-off-by: wili-65535 <wili-65535@users.noreply.github.com>

* v1.9.3: rebase

Signed-off-by: wili-65535 <wili-65535@users.noreply.github.com>

* v1.9.4: rebase

Signed-off-by: wili-65535 <wili-65535@users.noreply.github.com>

* v1.9.5: remove API change

Signed-off-by: wili-65535 <wili-65535@users.noreply.github.com>

---------

Signed-off-by: wili-65535 <wili-65535@users.noreply.github.com>
Co-authored-by: wili-65535 <wili-65535@users.noreply.github.com>
 2025-05-12 22:32:29 +02:00
yuxianq
a4c3359513
fix: Reset planned states to avoid memory leak in TrtllmAttentionWrapper (#4227) 
Signed-off-by: Yuxian Qiu <142763828+yuxianq@users.noreply.github.com>
 2025-05-12 23:25:54 +08:00
Fridah-nv
3dbb087292
[TRTLLM-5188] fix: [AutoDeploy] update output shape of prepare_fused_mha_metadata_fake (#4199) 
* update output shape of fake kernel prepare_fused_mha_metadata_fake

Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>

* minor

Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>

---------

Signed-off-by: Frida Hou <201670829+Fridah-nv@users.noreply.github.com>
 2025-05-12 11:11:40 -04:00
yuxianq
b35f9a67f9
refactor: Allow models to override apply_qk_norm. (#4078) 
Signed-off-by: Yuxian Qiu <142763828+yuxianq@users.noreply.github.com>
 2025-05-12 19:38:24 +08:00
Zheng Duan
c9e2a963e0
feat: add kv cache aware router (#3831) 
* kv cache aware router

Signed-off-by: Zheng Duan <200704041+zhengd-nv@users.noreply.github.com>

* add tests

Signed-off-by: Zheng Duan <200704041+zhengd-nv@users.noreply.github.com>

* router config

Signed-off-by: Zheng Duan <200704041+zhengd-nv@users.noreply.github.com>

* eviction test

Signed-off-by: Zheng Duan <200704041+zhengd-nv@users.noreply.github.com>

add test

Signed-off-by: Zheng Duan <200704041+zhengd-nv@users.noreply.github.com>

* eviction detect in worker test

Signed-off-by: Zheng Duan <200704041+zhengd-nv@users.noreply.github.com>

* move worker tests to single gpu

Signed-off-by: Zheng Duan <200704041+zhengd-nv@users.noreply.github.com>

* reduce memory fraction

Signed-off-by: Zheng Duan <200704041+zhengd-nv@users.noreply.github.com>

* fix partial block

Signed-off-by: Zheng Duan <200704041+zhengd-nv@users.noreply.github.com>

---------

Signed-off-by: Zheng Duan <200704041+zhengd-nv@users.noreply.github.com>
 2025-05-12 07:23:57 -04:00
Yixin Dong
c90ebadd84
feat: Support the Structural Tag in guided decoding (#4066) 
* finish

Signed-off-by: Ubospica <ubospica@gmail.com>

* update

Signed-off-by: Ubospica <ubospica@gmail.com>

* update

Signed-off-by: Ubospica <ubospica@gmail.com>

* fix

Signed-off-by: Enwei Zhu <21126786+syuoni@users.noreply.github.com>

* exc overlap scheduler

Signed-off-by: Enwei Zhu <21126786+syuoni@users.noreply.github.com>

* add test

Signed-off-by: Enwei Zhu <21126786+syuoni@users.noreply.github.com>

* fix api ref

Signed-off-by: Enwei Zhu <21126786+syuoni@users.noreply.github.com>

---------

Signed-off-by: Ubospica <ubospica@gmail.com>
Signed-off-by: Enwei Zhu <21126786+syuoni@users.noreply.github.com>
Co-authored-by: Enwei Zhu <21126786+syuoni@users.noreply.github.com>
 2025-05-12 17:24:50 +08:00
Zhenhuan Chen
9212e9a740
[TRTLLM-4911] feat(scaffolding): make sampling_params only setable by controller (#4151) 
feat(scaffolding): make sampling_params only setable by controller

Signed-off-by: Zhenhuan Chen <chenzhh3671@gmail.com>
 2025-05-12 15:29:09 +08:00
Chuang Zhu
1333f4f5d5
remove cache_transceiver_prealloc_size (#4153) 
Signed-off-by: Chuang Zhu <111838961+chuangz0@users.noreply.github.com>
 2025-05-12 11:53:53 +08:00
Frank
0dcf47f1c2
[TRTLLM-4717][perf] Set CUDA graph max batch size and padding in throughput benchmark. (#3875) 
* Set cuda graph max batch size.

Signed-off-by: Frank Di Natale <3429989+FrankD412@users.noreply.github.com>

* Set padding.

Signed-off-by: Frank Di Natale <3429989+FrankD412@users.noreply.github.com>

---------

Signed-off-by: Frank Di Natale <3429989+FrankD412@users.noreply.github.com>
 2025-05-09 23:20:52 +08:00
Mike Iovine
4b8ba7ad61
[fix][nvbug/5244009] Fix llama 4 test lists/scout accuracy issue (#4069) 
[fix] Fix llama 4 test lists

Signed-off-by: Mike Iovine <6158008+mikeiovine@users.noreply.github.com>
 2025-05-09 22:45:14 +08:00
chenfeiz0326
ffc13bd325
Cherry-pick: Use multi-threading to load MoE expert weights (#4137) 
* Use multi-threading to load MoE expert weights

Signed-off-by: Po-Han Huang <pohanh@nvidia.com>

* Update code formatting

Signed-off-by: Chenfei Zhang <chenfeiz@nvidia.com>

* Update code formatting

Signed-off-by: Chenfei Zhang <chenfeiz@nvidia.com>

---------

Signed-off-by: Po-Han Huang <pohanh@nvidia.com>
Signed-off-by: Chenfei Zhang <chenfeiz@nvidia.com>
Co-authored-by: Po-Han Huang <pohanh@nvidia.com>
 2025-05-09 17:29:24 +08:00
WeiHaocheng
0f01826dde
feat: support task collection for to collect information (#3328) (#3824) 
Signed-off-by: fredw (generated by with_the_same_user script) <20514172+WeiHaocheng@users.noreply.github.com>
 2025-05-09 17:09:01 +08:00