[Feature]: Contribute T5 model to vLLM - Githubissues

vllm-project / vllm

A high-throughput and memory-efficient inference and serving engine for LLMs

https://docs.vllm.ai

Apache License 2.0

26.76k stars 3.92k forks source link

[Feature]: Contribute T5 model to vLLM #8036

Open shivance opened 2 weeks ago

shivance commented 2 weeks ago

🚀 The feature, motivation and pitch

Hi vLLM team, As you already know, T5 is a perfect model and as mentioned in #7366, the project intends to add support for T5. I want to help the cause and contribute T5 to the project.

As this is my first contribution to vLLM project I am quite pumped up as well as eager to learn from you all.

Alternatives

No response

Additional context

No response

Before submitting a new issue...

[X] Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.

robertgshaw2-neuralmagic commented 2 weeks ago

@shivance thanks for your interest!

See: https://github.com/vllm-project/vllm/issues/7366 for a detailed discussion of enc-dec support. We would definitely welcome help on T5. For T5, the key is that we need an attention implementation with the custom bias

https://github.com/vllm-project/vllm/issues/7366#user-content-add-t5-model