Lain 9a3835aaa9 Fix trtllm-gen attention env and add attention sink (#22378) ...

Signed-off-by: Siyuan Fu <siyuanf@nvidia.com>
Signed-off-by: Lain <fusiyuan2000@hotmail.com>
Signed-off-by: Yongye Zhu <zyy1102000@gmail.com>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
Co-authored-by: Michael Goin <mgoin64@gmail.com>
Co-authored-by: Yongye Zhu <zyy1102000@gmail.com>

2025-08-06 18:07:41 -07:00

..

__init__.py

[Core] Move multimodal placeholder from chat utils to model definition (#20355)

2025-07-03 08:18:30 +00:00

adapters.py

[Model][1/N] Support multiple poolers at model level (#21227)

2025-07-21 02:22:21 -07:00

aimv2.py

[Misc] Remove redundant config definitions (#21891)

2025-07-29 23:54:18 -07:00

arcee.py

Fix Arcee model weight loading: Add custom load_weights (#21725)

2025-08-04 04:09:56 -07:00

arctic.py

[Misc] Add SPDX-FileCopyrightText (#19100)

2025-06-03 11:20:17 -07:00

aria.py

Migrate AriaImagePixelInputs to TensorSchema for shape validation (#21620)

2025-07-26 06:08:15 -07:00

aya_vision.py

[Misc] Automatically resolve HF processor init kwargs (#22005)

2025-07-31 22:44:10 -07:00

baichuan.py

[CI] change spell checker from codespell to typos (#18711)

2025-06-11 19:57:10 -07:00

bailing_moe.py

support torch.compile for bailing moe (#21664)

2025-07-26 23:54:32 +00:00

bamba.py

[v1] - Mamba1 Attention Metadata (#21249)

2025-08-06 17:03:42 -07:00

bart.py

[Model] use AutoWeightsLoader for bart (#18299)

2025-07-20 08:15:50 +00:00

bert_with_rope.py

[Model][CI] Let more pooling models support v1 (#21747)

2025-07-31 01:51:15 -07:00

bert.py

Support encoder-only models without KV-Cache (#21270)

2025-07-26 21:09:52 +08:00

blip2.py

Migrate Blip2ImagePixelInputs and Blip2ImageEmbeddingInputs to TensorSchema (#21656)

2025-07-27 10:33:52 +08:00

blip.py

[Misc] Add SPDX-FileCopyrightText (#19100)

2025-06-03 11:20:17 -07:00

bloom.py

[Misc] Add SPDX-FileCopyrightText (#19100)

2025-06-03 11:20:17 -07:00

chameleon.py

Migrate ChameleonImagePixelInputs to TensorSchema (#21657)

2025-07-26 19:34:25 -07:00

chatglm.py

[Misc] Modify the organization of GLM series (#22171)

2025-08-03 23:51:20 -07:00

clip.py

[Misc] Add SPDX-FileCopyrightText (#19100)

2025-06-03 11:20:17 -07:00

commandr.py

[CI/Build] Ensure compatability with Transformers v4.53 (#20541)

2025-07-11 20:53:07 -07:00

config.py

[GptOss] Add GptOss reasoning parser to support structure output (#22322)

2025-08-05 23:39:13 -07:00

constant_size_cache.py

[Misc] Add SPDX-FileCopyrightText (#19100)

2025-06-03 11:20:17 -07:00

dbrx.py

[Misc] Remove redundant config definitions (#21891)

2025-07-29 23:54:18 -07:00

deepseek_mtp.py

[Bugfix] Fix MTP weight loading (#21941)

2025-07-31 16:33:53 -04:00

deepseek_v2.py

[Model] Remove DSV2 unused code (#21903)

2025-07-30 00:55:03 -07:00

deepseek_vl2.py

[Misc] Add tensor schema test coverage for multimodal models (#21754)

2025-08-03 00:52:14 -07:00

deepseek.py

[Misc] Add SPDX-FileCopyrightText (#19100)

2025-06-03 11:20:17 -07:00

dots1.py

[Quantization] Enable BNB support for more MoE models (#21370)

2025-07-25 03:57:34 -07:00

ernie45_moe.py

[Bugfix] Fix Ernie4_5_MoeForCausalLM shared experts (#21717)

2025-07-28 11:02:25 +00:00

ernie45.py

[Ernie 4.5] Name Change for Base 0.3B Model (#21735)

2025-07-28 12:22:32 +00:00

exaone4.py

[Misc] Remove redundant config definitions (#21891)

2025-07-29 23:54:18 -07:00

exaone.py

[Misc] Remove redundant config definitions (#21891)

2025-07-29 23:54:18 -07:00

fairseq2_llama.py

[Misc] Add SPDX-FileCopyrightText (#19100)

2025-06-03 11:20:17 -07:00

falcon_h1.py

[v1] - Mamba1 Attention Metadata (#21249)

2025-08-06 17:03:42 -07:00

falcon.py

[Misc] Add SPDX-FileCopyrightText (#19100)

2025-06-03 11:20:17 -07:00

florence2.py

[Misc] Automatically resolve HF processor init kwargs (#22005)

2025-07-31 22:44:10 -07:00

fuyu.py

[Misc] Automatically resolve HF processor init kwargs (#22005)

2025-07-31 22:44:10 -07:00

gemma2.py

Mark invariant normalizer in Gemma as non-persistent (#19788)

2025-06-18 22:56:03 -07:00

gemma3_mm.py

Migrate Gemma3ImagePixelInputs to TensorSchema (#21676)

2025-07-27 22:36:05 -07:00

gemma3.py

[CI/Build] Ensure compatability with Transformers v4.53 (#20541)

2025-07-11 20:53:07 -07:00

gemma3n.py

[Model] [Quantization] Support quantization for Gemma3n (#21974)

2025-07-31 22:45:15 -07:00

gemma.py

[Model] Re-add the implicit conversion feature for as_seq_cls_model (#21103)

2025-07-18 07:15:07 +00:00

glm4_1v.py

for glm-4.1V update (#22000)

2025-08-02 01:46:57 -07:00

glm4_moe_mtp.py

GLM-4 Update (#20736)

2025-07-19 22:40:31 +00:00

glm4_moe.py

self.gate dtype update for GLM-4.5 (#22203)

2025-08-04 19:12:38 -07:00

glm4.py

[Misc] Add SPDX-FileCopyrightText (#19100)

2025-06-03 11:20:17 -07:00

glm4v.py

[Misc] Modify the organization of GLM series (#22171)

2025-08-03 23:51:20 -07:00

glm.py

[Misc] Add SPDX-FileCopyrightText (#19100)

2025-06-03 11:20:17 -07:00

gpt2.py

[Model][1/N] Support multiple poolers at model level (#21227)

2025-07-21 02:22:21 -07:00

gpt_bigcode.py

[Misc] Add SPDX-FileCopyrightText (#19100)

2025-06-03 11:20:17 -07:00

gpt_j.py

[Misc] Add SPDX-FileCopyrightText (#19100)

2025-06-03 11:20:17 -07:00

gpt_neox.py

[Misc] Add SPDX-FileCopyrightText (#19100)

2025-06-03 11:20:17 -07:00

gpt_oss.py

Fix trtllm-gen attention env and add attention sink (#22378)

2025-08-06 18:07:41 -07:00

granite_speech.py

[Bugfix] Fix granite speech shape validation (#21762)

2025-07-28 14:19:21 -04:00

granite.py

[Bugfix] Fix Tensor Parallelism Padding Consistency in Granite Models (#20843)

2025-07-12 06:11:30 -07:00

granitemoe.py

Let GraniteMoeAttention use YaRN (#21174)

2025-07-18 05:52:52 -07:00

granitemoehybrid.py

[v1] - Mamba1 Attention Metadata (#21249)

2025-08-06 17:03:42 -07:00

granitemoeshared.py

Let GraniteMoeAttention use YaRN (#21174)

2025-07-18 05:52:52 -07:00

gritlm.py

[V1] Get supported tasks from model runner instead of model config (#21585)

2025-07-25 05:36:45 -07:00

grok1.py

[Quantization] Enable BNB support for more MoE models (#21100)

2025-07-18 17:52:02 -07:00

h2ovl.py

[Misc] Automatically resolve HF processor init kwargs (#22005)

2025-07-31 22:44:10 -07:00

hunyuan_v1.py

[Model] add Hunyuan V1 Dense Model support. (#21368)

2025-07-23 03:54:08 -07:00

hyperclovax_vision.py

[Misc] Automatically resolve HF processor init kwargs (#22005)

2025-07-31 22:44:10 -07:00

idefics2_vision_model.py

[Misc] Add SPDX-FileCopyrightText (#19100)

2025-06-03 11:20:17 -07:00

idefics3.py

[Misc] Automatically resolve HF processor init kwargs (#22005)

2025-07-31 22:44:10 -07:00

interfaces_base.py

Update transformers to v4.55 (#21931)

2025-08-05 22:56:14 -07:00

interfaces.py

[Bugfix] Relax lang pin for voxtral (#21833)

2025-07-30 20:38:52 -07:00

intern_vit.py

[Misc] Add SPDX-FileCopyrightText (#19100)

2025-06-03 11:20:17 -07:00

internlm2_ve.py

[Misc] Add SPDX-FileCopyrightText (#19100)

2025-06-03 11:20:17 -07:00

internlm2.py

[Model][1/N] Support multiple poolers at model level (#21227)

2025-07-21 02:22:21 -07:00

interns1_vit.py

Support Intern-S1 (#21628)

2025-07-26 19:14:04 +08:00

interns1.py

[VLM] Add video support for Intern-S1 (#21671)

2025-07-27 11:49:43 +00:00

internvl.py

[Misc] Automatically resolve HF processor init kwargs (#22005)

2025-07-31 22:44:10 -07:00

jais.py

[Misc] Add SPDX-FileCopyrightText (#19100)

2025-06-03 11:20:17 -07:00

jamba.py

[v1] - Mamba1 Attention Metadata (#21249)

2025-08-06 17:03:42 -07:00

jina_vl.py

[Model] Pooling model activation supports per request control by PoolingParams (#20538)

2025-08-05 00:37:00 -07:00

keye.py

[Misc] Add tensor schema test coverage for multimodal models (#21754)

2025-08-03 00:52:14 -07:00

kimi_vl.py

Migrate KimiVLImagePixelInputs to TensorSchema (#21769)

2025-08-05 02:36:18 -07:00

llama4_eagle.py

[Meta] Official Eagle mm support, first enablement on llama4 (#20788)

2025-07-31 10:35:07 -07:00

llama4.py

[Meta] Official Eagle mm support, first enablement on llama4 (#20788)

2025-07-31 10:35:07 -07:00

llama_eagle3.py

[Speculative Decoding] Add speculators config support (#21345)

2025-08-01 08:25:18 -04:00

llama_eagle.py

[Meta] Official Eagle mm support, first enablement on llama4 (#20788)

2025-07-31 10:35:07 -07:00

llama.py

[Model] Re-add the implicit conversion feature for as_seq_cls_model (#21103)

2025-07-18 07:15:07 +00:00

llava_next_video.py

[Core] Move multimodal placeholder from chat utils to model definition (#20355)

2025-07-03 08:18:30 +00:00

llava_next.py

[Core] Move multimodal placeholder from chat utils to model definition (#20355)

2025-07-03 08:18:30 +00:00

llava_onevision.py

[Core] Move multimodal placeholder from chat utils to model definition (#20355)

2025-07-03 08:18:30 +00:00

llava.py

[Misc] Automatically resolve HF processor init kwargs (#22005)

2025-07-31 22:44:10 -07:00

mamba2.py

[v1] - Mamba1 Attention Metadata (#21249)

2025-08-06 17:03:42 -07:00

mamba_cache.py

[Kernel] Triton implementation of causal-conv1d for Mamba-based models (#18218)

2025-07-09 12:53:55 -07:00

mamba.py

[v1] - Mamba1 Attention Metadata (#21249)

2025-08-06 17:03:42 -07:00

medusa.py

[Misc] Add SPDX-FileCopyrightText (#19100)

2025-06-03 11:20:17 -07:00

mimo_mtp.py

[Model] Remove model sampler (#21059)

2025-07-16 19:03:37 +00:00

mimo.py

[Model] Remove model sampler (#21059)

2025-07-16 19:03:37 +00:00

minicpm3.py

[Misc] Add SPDX-FileCopyrightText (#19100)

2025-06-03 11:20:17 -07:00

minicpm_eagle.py

[Misc] Add SPDX-FileCopyrightText (#19100)

2025-06-03 11:20:17 -07:00

minicpm.py

[Misc] Add SPDX-FileCopyrightText (#19100)

2025-06-03 11:20:17 -07:00

minicpmo.py

[CI/Build] Ensure compatability with Transformers v4.53 (#20541)

2025-07-11 20:53:07 -07:00

minicpmv.py

[Misc] Automatically resolve HF processor init kwargs (#22005)

2025-07-31 22:44:10 -07:00

minimax_cache.py

[Misc] Add SPDX-FileCopyrightText (#19100)

2025-06-03 11:20:17 -07:00

minimax_text_01.py

[Model] Support HF format of minimax (#20211)

2025-07-11 02:55:21 +00:00

minimax_vl_01.py

[Misc] Remove redundant config definitions (#21891)

2025-07-29 23:54:18 -07:00

mistral3.py

[Core] Move multimodal placeholder from chat utils to model definition (#20355)

2025-07-03 08:18:30 +00:00

mixtral_quant.py

[CI] change spell checker from codespell to typos (#18711)

2025-06-11 19:57:10 -07:00

mixtral.py

[Quantization] add BNB for MixtralForCausalLM (#20893)

2025-07-14 07:34:34 +00:00

mllama4.py

[Fix] Fix llama4 modelopt weight loading error (#22107)

2025-08-03 00:50:34 -07:00

mllama.py

[Core] Move multimodal placeholder from chat utils to model definition (#20355)

2025-07-03 08:18:30 +00:00

mlp_speculator.py

[Misc] Add SPDX-FileCopyrightText (#19100)

2025-06-03 11:20:17 -07:00

modernbert.py

[Model][CI] Let more pooling models support v1 (#21747)

2025-07-31 01:51:15 -07:00

module_mapping.py

[Misc] Add SPDX-FileCopyrightText (#19100)

2025-06-03 11:20:17 -07:00

molmo.py

[Core] Move multimodal placeholder from chat utils to model definition (#20355)

2025-07-03 08:18:30 +00:00

moonvit.py

[Misc] Add SPDX-FileCopyrightText (#19100)

2025-06-03 11:20:17 -07:00

mpt.py

[CI/Build] Fix registry tests (#21934)

2025-07-30 09:10:41 -07:00

nemotron_h.py

[v1] - Mamba1 Attention Metadata (#21249)

2025-08-06 17:03:42 -07:00

nemotron_nas.py

[Misc] Add SPDX-FileCopyrightText (#19100)

2025-06-03 11:20:17 -07:00

nemotron_vl.py

[Misc] Automatically resolve HF processor init kwargs (#22005)

2025-07-31 22:44:10 -07:00

nemotron.py

[Misc] Add SPDX-FileCopyrightText (#19100)

2025-06-03 11:20:17 -07:00

nvlm_d.py

[Misc] Automatically resolve HF processor init kwargs (#22005)

2025-07-31 22:44:10 -07:00

olmo2.py

[Misc] Add SPDX-FileCopyrightText (#19100)

2025-06-03 11:20:17 -07:00

olmo.py

[Misc] Add SPDX-FileCopyrightText (#19100)

2025-06-03 11:20:17 -07:00

olmoe.py

[Quantization] add BNB for MixtralForCausalLM (#20893)

2025-07-14 07:34:34 +00:00

opt.py

[Misc] Add SPDX-FileCopyrightText (#19100)

2025-06-03 11:20:17 -07:00

orion.py

[Misc] Add SPDX-FileCopyrightText (#19100)

2025-06-03 11:20:17 -07:00

ovis.py

[Misc] Automatically resolve HF processor init kwargs (#22005)

2025-07-31 22:44:10 -07:00

paligemma.py

[CI/Build] Ensure compatability with Transformers v4.53 (#20541)

2025-07-11 20:53:07 -07:00

persimmon.py

[Misc] Add SPDX-FileCopyrightText (#19100)

2025-06-03 11:20:17 -07:00

phi3.py

[Misc] Add SPDX-FileCopyrightText (#19100)

2025-06-03 11:20:17 -07:00

phi3v.py

[Misc] Automatically resolve HF processor init kwargs (#22005)

2025-07-31 22:44:10 -07:00

phi4_multimodal.py

[Misc] Automatically resolve HF processor init kwargs (#22005)

2025-07-31 22:44:10 -07:00

phi4flash.py

[Model] Mamba2 preallocate SSM output tensor to avoid d2d copy overhead (#21075)

2025-08-02 01:59:34 -07:00

phi4mm_audio.py

[CI] change spell checker from codespell to typos (#18711)

2025-06-11 19:57:10 -07:00

phi4mm_utils.py

[Misc] Add SPDX-FileCopyrightText (#19100)

2025-06-03 11:20:17 -07:00

phi4mm.py

[Misc] Automatically resolve HF processor init kwargs (#22005)

2025-07-31 22:44:10 -07:00

phi.py

[Misc] Add SPDX-FileCopyrightText (#19100)

2025-06-03 11:20:17 -07:00

phimoe.py

[Misc] Fix PhiMoE expert mapping (#21085)

2025-07-17 05:47:49 +00:00

pixtral.py

Remove xformers requirement for Mistral-format Pixtral and Mistral3 (#21154)

2025-07-26 17:20:29 -06:00

plamo2.py

[Model] Mamba2 preallocate SSM output tensor to avoid d2d copy overhead (#21075)

2025-08-02 01:59:34 -07:00

prithvi_geospatial_mae.py

Add support for Prithvi in Online serving mode (#21518)

2025-07-25 07:01:27 -07:00

qwen2_5_omni_thinker.py

[Misc] Automatically resolve HF processor init kwargs (#22005)

2025-07-31 22:44:10 -07:00

qwen2_5_vl.py

[FEAT][ROCm] Enable running Flash Attention as ViT attn backend for Qwen-VL models on ROCm platform. (#22069)

2025-08-01 23:53:18 -07:00

qwen2_audio.py

[Misc] Automatically resolve HF processor init kwargs (#22005)

2025-07-31 22:44:10 -07:00

qwen2_moe.py

[Misc] Qwen MoE model supports LoRA (#20932)

2025-07-17 18:32:52 +00:00

qwen2_rm.py

[Model] Pooling model activation supports per request control by PoolingParams (#20538)

2025-08-05 00:37:00 -07:00

qwen2_vl.py

Update transformers to v4.55 (#21931)

2025-08-05 22:56:14 -07:00

qwen2.py

[Speculators][Speculative Decoding] Add Qwen Eagle3 Support (#21835)

2025-08-01 19:43:37 -07:00

qwen3_moe.py

[Feature][EPLB] Add eplb support for Qwen3 (#20815)

2025-07-30 06:27:57 -07:00

qwen3.py

[Speculators][Speculative Decoding] Add Qwen Eagle3 Support (#21835)

2025-08-01 19:43:37 -07:00

qwen_vl.py

[Core] Move multimodal placeholder from chat utils to model definition (#20355)

2025-07-03 08:18:30 +00:00

qwen.py

[Misc] Add SPDX-FileCopyrightText (#19100)

2025-06-03 11:20:17 -07:00

registry.py

Add GPT-OSS model code and config [1/N] (#22327)

2025-08-05 23:26:00 -07:00

roberta.py

Support encoder-only models without KV-Cache (#21270)

2025-07-26 21:09:52 +08:00

siglip.py

[Misc] Add SPDX-FileCopyrightText (#19100)

2025-06-03 11:20:17 -07:00

skyworkr1v.py

[Misc] Automatically resolve HF processor init kwargs (#22005)

2025-07-31 22:44:10 -07:00

smolvlm.py

[Misc] Automatically resolve HF processor init kwargs (#22005)

2025-07-31 22:44:10 -07:00

solar.py

[Misc] Add SPDX-FileCopyrightText (#19100)

2025-06-03 11:20:17 -07:00

stablelm.py

[Misc] Add SPDX-FileCopyrightText (#19100)

2025-06-03 11:20:17 -07:00

starcoder2.py

[Misc] Add SPDX-FileCopyrightText (#19100)

2025-06-03 11:20:17 -07:00

step3_text.py

[Model] Add step3 vl (#21998)

2025-07-31 23:19:06 +08:00

step3_vl.py

[Model] Add step3 vl (#21998)

2025-07-31 23:19:06 +08:00

tarsier.py

[Misc] Automatically resolve HF processor init kwargs (#22005)

2025-07-31 22:44:10 -07:00

telechat2.py

[CI/Build] Fix registry tests (#21934)

2025-07-30 09:10:41 -07:00

teleflm.py

[Misc] Add SPDX-FileCopyrightText (#19100)

2025-06-03 11:20:17 -07:00

transformers.py

Update transformers to v4.55 (#21931)

2025-08-05 22:56:14 -07:00

ultravox.py

[Misc] Automatically resolve HF processor init kwargs (#22005)

2025-07-31 22:44:10 -07:00

utils.py

Update transformers to v4.55 (#21931)

2025-08-05 22:56:14 -07:00

vision.py

[FEAT][ROCm] Enable running Flash Attention as ViT attn backend for Qwen-VL models on ROCm platform. (#22069)

2025-08-01 23:53:18 -07:00

voxtral.py

[Bugfix] Relax lang pin for voxtral (#21833)

2025-07-30 20:38:52 -07:00

whisper.py

[Misc] Automatically resolve HF processor init kwargs (#22005)

2025-07-31 22:44:10 -07:00

zamba2.py

[v1] - Mamba1 Attention Metadata (#21249)

2025-08-06 17:03:42 -07:00