kanshan/TensorRT-LLMs
1
0
Fork 0
mirror of https://github.com/NVIDIA/TensorRT-LLM.git synced 2026-01-24 04:33:04 +08:00
Code Issues Actions 1 Packages Projects Releases Wiki Activity
3,725 Commits 28 Branches 62 Tags 2.2 GiB
ee6944bfa2
Commit Graph

171 Commits

Author SHA1 Message Date
Neta Zmora
028fc877a5
[#9096][feature] Auto Deploy: configurable fused MoE backend (#9194) 
Allow configuring Auto Deploy's MoE/FP8-MoE backend from external yaml config file.

Signed-off-by: Neta Zmora <96238833+nzmora-nvidia@users.noreply.github.com>
 2025-11-19 21:50:22 -08:00
NVShreyas
1eae941d77
[#9237][feat] enable iter stats in autodeploy (#9278) 
Signed-off-by: Shreyas Misra <shreyasm@nvidia.com>
 2025-11-19 19:29:29 +01:00
Grzegorz Kwasniewski
7905d6c0da
[#9098][feat] Simple sharding latent experts (#9099) 
Signed-off-by: greg-kwasniewski1 <213329731+greg-kwasniewski1@users.noreply.github.com>
 2025-11-18 21:14:22 -05:00
Grzegorz Kwasniewski
92f86a50d4
[#9137][feat] Factory sharding as default (#9144) 
Signed-off-by: greg-kwasniewski1 <213329731+greg-kwasniewski1@users.noreply.github.com>
 2025-11-18 21:12:03 -05:00
Eran Geva
3ac11a6180
[#9152][fix] AutoDeploy fused_allreduce_residual_rmsnorm to support demollm mode (#9197) 
Signed-off-by: Eran Geva <19514940+MrGeva@users.noreply.github.com>
 2025-11-18 22:15:29 +02:00
Chenghao Zhang
f0b68e4c66
[None][feat] AutoDeploy: Perf improvement for small batch size (#9163) 
Signed-off-by: Chenghao Zhang <211069071+nvchenghaoz@users.noreply.github.com>
Co-authored-by: Suyog Gupta <41447211+suyoggupta@users.noreply.github.com>
 2025-11-18 12:11:12 -08:00
Robin Kobus
df41f220a2
[TRTLLM-8831][feat] Enable early exit with overlap scheduler (#8587) 
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
 2025-11-17 18:07:13 +01:00
Kaiyu Xie
04be5a704e
[None] [fix] Fix missing ActivationType issue (#9171) 
Signed-off-by: Kaiyu Xie <26294424+kaiyux@users.noreply.github.com>
Signed-off-by: Enwei Zhu <21126786+syuoni@users.noreply.github.com>
Signed-off-by: Neta Zmora <96238833+nzmora-nvidia@users.noreply.github.com>
Co-authored-by: Enwei Zhu <21126786+syuoni@users.noreply.github.com>
Co-authored-by: Neta Zmora <96238833+nzmora-nvidia@users.noreply.github.com>
 2025-11-17 10:43:25 +08:00
Chenghao Zhang
f6f6e1f25d
[#9102][feat] AutoDeploy: Support fp8 kv cache (#9107) 
Signed-off-by: Chenghao Zhang <211069071+nvchenghaoz@users.noreply.github.com>
 2025-11-13 23:55:45 -08:00
Suyog Gupta
d12cb9436d
[None][feat] Autodeploy add triton configs and optimize mamba prefill (#9083) 
Signed-off-by: Suyog Gupta <41447211+suyoggupta@users.noreply.github.com>
 2025-11-13 19:15:43 -08:00
Neta Zmora
34dc6869f3
[#8732][feat] Update TRTLLM Cutlass MoE kernels with ReLU2 (#9011) 
Update TRTLLM Cutlass MoE kernels with ReLU2 activation.

Nemotron-6 requires ReLU2 (i.e. squared ReLU) MoE activation function.
The PR adds this and adds an API to set the activation function, in general.
The ReLU2 changes are based on this FlashInfer PR: https://github.com/flashinfer-ai/flashinfer/pull/1954.

The PR also updates the Auto Deploy MoE backend for 16-bit and FP8 from
Triton (`torch.ops.auto_deploy.triton_moe_fused`, `torch.ops.auto_deploy.triton_quant_fp8_moe`) to TRTLLM/Cutlass (`torch.ops.auto_deploy.trtllm_moe_fused`, `torch.ops.auto_deploy.trtllm_quant_fp8_moe_fused`).

Signed-off-by: Neta Zmora <96238833+nzmora-nvidia@users.noreply.github.com>
Signed-off-by: Chenghao Zhang <211069071+nvchenghaoz@users.noreply.github.com>
Co-authored-by: Chenghao Zhang <211069071+nvchenghaoz@users.noreply.github.com>
 2025-11-13 16:54:45 -08:00
Frida Hou
b51258acdd
[None][autodeploy] fix weight extraction for graph based quantized checkpoints (#9109) 
Signed-off-by: Fridah-nv <201670829+Fridah-nv@users.noreply.github.com>
 2025-11-13 13:14:24 -08:00
Frida Hou
e96a3d294d
[None][autodeploy] minor refactor to rmsnorm transforms (#8657) 
Signed-off-by: Fridah-nv <201670829+Fridah-nv@users.noreply.github.com>
 2025-11-13 13:13:58 -08:00
Chenghao Zhang
f1d637ec69
[None][fix] AutoDeploy: Use tmp folder for the load_moe_align (#9101) 
Signed-off-by: Chenghao Zhang <211069071+nvchenghaoz@users.noreply.github.com>
 2025-11-12 14:59:49 -08:00
Neta Zmora
53491ffdb1
[#9023][feat] reduce AD graph optimization time for non-participating passes (#9024) 
Shorten AD graph optimization by 30% (measured on Nemotron-6):

A bug in the transformation interface marked all passes as not clean, regardless of what was reported by the transformation
Fix how the optimization passes report the results of their actions. Many passes report that the graph is not clean even when they didn't participate in the optimization. Each graph cleaning invocation can take several seconds.

Signed-off-by: Neta Zmora <96238833+nzmora-nvidia@users.noreply.github.com>
 2025-11-12 09:05:53 -08:00
Lucas Liebenwein
aca56097cb
[None][fix] AutoDeploy: update nano3 accuracy test (#9061) 
Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>
 2025-11-11 12:26:31 -08:00
Chenghao Zhang
ec9cf715a2
[None][feat] AutoDeploy: Perf improvement for mamba layers (#8991) 
Signed-off-by: Chenghao Zhang <211069071+nvchenghaoz@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>
 2025-11-11 08:27:07 -08:00
Lucas Liebenwein
6bf4e59267
[#8763][feature] AutoDeploy: configurable dtype for caching (#8812) 
Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>
 2025-11-10 22:17:14 -08:00
Frida Hou
f40e1f7496
[https://nvbugs/5625972][fix] Add context manager to fix FakeTensorProp (#9047) 
Signed-off-by: Fridah-nv <201670829+Fridah-nv@users.noreply.github.com>
 2025-11-10 16:25:58 -08:00
QI JUN
1c6e490894
[TRTLLM-9065][chore] remove PyTorchConfig completely (#8856) 
Signed-off-by: junq <22017000+QiJune@users.noreply.github.com>
 2025-11-06 22:37:03 -08:00
Eran Geva
990e674b71
[None][fix] Switch AD AllReduce strategy to NCCL (#8979) 
Signed-off-by: Eran Geva <19514940+MrGeva@users.noreply.github.com>
 2025-11-07 06:49:44 +02:00
Chenghao Zhang
1a78e7a3d6
[None][feat] AutoDeploy: Support Latent MOE for Nemotron (#8955) 
Signed-off-by: Chenghao Zhang <211069071+nvchenghaoz@users.noreply.github.com>
 2025-11-06 12:40:19 -08:00
Chenghao Zhang
ddf2d010e2
[TRTLLM-8814][feat] AutoDeploy: Use TRTLLM kernels for FP8 linear (#8820) 
Signed-off-by: Chenghao Zhang <211069071+nvchenghaoz@users.noreply.github.com>
Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>
Signed-off-by: nvchenghaoz <211069071+nvchenghaoz@users.noreply.github.com>
Co-authored-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>
 2025-11-06 11:00:10 -08:00
Frida Hou
fb7f9831d3
[#8924][fix] Fix AutoDeploy pattern matcher for torch 2.9 (#8920) 
Signed-off-by: Fridah-nv <201670829+Fridah-nv@users.noreply.github.com>
 2025-11-05 13:29:20 -08:00
Lucas Liebenwein
b181568d6f
[TRTLLM-8201][feat] Nemotron H MoE Sharding (#8744) 
Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>
Signed-off-by: greg-kwasniewski1 <213329731+greg-kwasniewski1@users.noreply.github.com>
Co-authored-by: greg-kwasniewski1 <213329731+greg-kwasniewski1@users.noreply.github.com>
Co-authored-by: Suyog Gupta <41447211+suyoggupta@users.noreply.github.com>
 2025-11-05 12:35:29 -08:00
Frida Hou
11ded113cd
[#8389][fix] Update group attention matching to first map to custom torch attention (#8638) 
Signed-off-by: Fridah-nv <201670829+Fridah-nv@users.noreply.github.com>
 2025-11-04 12:00:43 -08:00
Yechan Kim
67208f1512
[None][fix] InputProcessor config naming convention fix (#8705) 
Signed-off-by: yechank <161688079+yechank-nvidia@users.noreply.github.com>
 2025-11-03 22:29:21 -08:00
Eran Geva
f8778230e3
[#8781][fix] Cache the AllReduce wrapper to avoid re-allocating workspace which caused a hang (#8803) 
Signed-off-by: Eran Geva <19514940+MrGeva@users.noreply.github.com>
 2025-11-02 15:30:39 +02:00
QI JUN
89e0117097
[TRTLLM-8836][chore] Create ModelEngine from LlmArgs (#8600) 
Signed-off-by: junq <22017000+QiJune@users.noreply.github.com>
 2025-11-01 05:26:06 -07:00
Suyog Gupta
3d0e38e074
[None][perf] AutoDeploy optimize _get_unique_value (#8822) 
Signed-off-by: Suyog Gupta <41447211+suyoggupta@users.noreply.github.com>
 2025-10-31 04:57:10 -07:00
Chenghao Zhang
71c5576a44
[TRTLLM-8734][feat] AutoDeploy: Enable the nvfp4 for Nemotron MOE (#8737) 
Signed-off-by: nvchenghaoz <211069071+nvchenghaoz@users.noreply.github.com>
Co-authored-by: Suyog Gupta <41447211+suyoggupta@users.noreply.github.com>
 2025-10-30 12:33:08 -07:00
Lucas Liebenwein
0ee71d95ec
[https://nvbugs/5606166][fix] AutoDeploy: use tuples for cudagraph shape lookup (#8658) 
Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>
 2025-10-28 10:52:43 -07:00
Anish Shanbhag
a09b38a862
[TRTLLM-8684][chore] Migrate BuildConfig to Pydantic, add a Python wrapper for KVCacheType enum (#8330) 
Signed-off-by: Anish Shanbhag <ashanbhag@nvidia.com>
 2025-10-28 09:17:26 -07:00
Eran Geva
e051a05e6c
[#8694][fix] fix AutoDeploy cuda memory access failure in nvidia/NVIDIA-Nemotron-Nano-31B-A3-v3 (#8696) 
Signed-off-by: Eran Geva <19514940+MrGeva@users.noreply.github.com>
 2025-10-28 13:21:43 +02:00
gramnarayan
88b0fbc8ff
[#8245][feat] Autodeploy: Guided Decoding Support (#8551) 
Signed-off-by: William Zhang <133824995+2ez4bz@users.noreply.github.com>
Signed-off-by: Govind Ramnarayan <105831528+govind-ramnarayan@users.noreply.github.com>
Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>
Co-authored-by: William Zhang <133824995+2ez4bz@users.noreply.github.com>
Co-authored-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>
 2025-10-28 09:29:57 +08:00
Chenghao Zhang
b9b2802599
[None][feat] Autodeploy: Update the ssm to use slice (#8667) 
Signed-off-by: nvchenghaoz <211069071+nvchenghaoz@users.noreply.github.com>
 2025-10-27 09:45:20 -07:00
Chenghao Zhang
a6d20f6f9b
[None][feat] AutoDeploy: Add FP8 MOE for Nemotron (#8599) 
Signed-off-by: Chenghao Zhang <211069071+nvchenghaoz@users.noreply.github.com>
Signed-off-by: Fridah-nv <201670829+Fridah-nv@users.noreply.github.com>
Signed-off-by: nvchenghaoz <211069071+nvchenghaoz@users.noreply.github.com>
Co-authored-by: Suyog Gupta <41447211+suyoggupta@users.noreply.github.com>
Co-authored-by: Fridah-nv <201670829+Fridah-nv@users.noreply.github.com>
 2025-10-25 15:26:45 -04:00
Suyog Gupta
f512ddaeef
[None][feat] add skip condition in AutoDeploy's triton fused moe kernel (#8632) 
Signed-off-by: Suyog Gupta <41447211+suyoggupta@users.noreply.github.com>
 2025-10-24 08:46:17 -04:00
h-guo18
23920223ab
[#4585][feat] Replace unified attention before export (#8303) 
Signed-off-by: h-guo18 <67671475+h-guo18@users.noreply.github.com>
 2025-10-23 18:02:04 -04:00
Suyog Gupta
2956978da3
[None][feat] Enable rms norm fusion for Nemotron MOE (#8563) 
Signed-off-by: Chenghao Zhang <211069071+nvchenghaoz@users.noreply.github.com>
Signed-off-by: junq <22017000+QiJune@users.noreply.github.com>
Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>
Signed-off-by: Suyog Gupta <41447211+suyoggupta@users.noreply.github.com>
Co-authored-by: Chenghao Zhang <211069071+nvchenghaoz@users.noreply.github.com>
Co-authored-by: QI JUN <22017000+QiJune@users.noreply.github.com>
Co-authored-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>
 2025-10-23 00:09:42 -04:00
Anish Shanbhag
15de45d782
[TRTLLM-8682][chore] Remove auto_parallel module (#8329) 
Signed-off-by: Anish Shanbhag <ashanbhag@nvidia.com>
 2025-10-22 20:53:08 -04:00
Leslie Fang
50d4e5bc06
[TRTLLM-8483][chore] Refine scheduler_config and peft_cache_config in create_py_executor (#8451) 
Signed-off-by: leslie-fang25 <leslief@nvidia.com>
 2025-10-22 08:33:48 +08:00
Chenghao Zhang
bac9e8c2ad
[None][feat] AutoDeploy: Add Nemotron MOE support for AutoDeploy (#8469) 2025-10-21 15:32:01 -07:00
Lucas Liebenwein
9b54b3bfaf
[None][chore] AutoDeploy: replace HF's deprecated keyword torch_dtype --> dtype (#8510) 
Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>
 2025-10-21 17:07:06 -04:00
YueWeng
8dc4aac5b6
[TRTLLM-8160][feat] Add max_total_draft_tokens (#8366) 
Signed-off-by: Yue Weng <25103990+yweng0828@users.noreply.github.com>
 2025-10-21 11:11:04 -04:00
Suyog Gupta
7050b1ea49
[#8272][feat] Enable chunked prefill for SSMs in AutoDeploy (#8477) 
Signed-off-by: Suyog Gupta <41447211+suyoggupta@users.noreply.github.com>
 2025-10-20 15:31:52 -07:00
Lucas Liebenwein
55c468b218
[#8461][feat] AutoDeploy: trtllm-serve bug fix + unit test (#8462) 
Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>
 2025-10-20 16:06:39 -04:00
mpikulski
97ce0ecefe
[TRTLLM-8436][feat] batched sampling and top-k logprobs improvements (#8398) 
Signed-off-by: ixlmar <206748156+ixlmar@users.noreply.github.com>
 2025-10-20 11:15:41 +02:00
Lucas Liebenwein
41169fb20c
[None][feat] AutoDeploy: chunked prefill support (#8158) 
Signed-off-by: Lucas Liebenwein <11156568+lucaslie@users.noreply.github.com>
 2025-10-18 00:47:35 -07:00
h-guo18
55fed1873c
[None][chore] AutoDeploy: cleanup old inference optimizer configs (#8039) 
Signed-off-by: h-guo18 <67671475+h-guo18@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>
 2025-10-17 15:55:57 -04:00