Is there any feature related to GPT-like models that can be applied to BERT-like models?

NVIDIA / TensorRT-LLM

TensorRT-LLM provides users with an easy-to-use Python API to define Large Language Models (LLMs) and build TensorRT engines that contain state-of-the-art optimizations to perform inference efficiently on NVIDIA GPUs. TensorRT-LLM also contains components to create Python and C++ runtimes that execute those TensorRT engines.

https://nvidia.github.io/TensorRT-LLM

Apache License 2.0

8.61k stars 978 forks source link

Is there any feature related to GPT-like models that can be applied to BERT-like models? #1517

Open zhangxin81 opened 6 months ago

zhangxin81 commented 6 months ago

Is there any fesature related to GPT-like models that can be applied to BERT-like models?

byshiue commented 6 months ago

They have some common optimization idea, like fusing the multi head attention kernel, quantizing the model to int8 or fp8.

Ashwin-Ramesh2607 commented 5 months ago

hI @byshiue, a related question. Does BertAttentionPlugin also use FlashAttention2 that GptAttention uses?

byshiue commented 5 months ago

Yes.