deepspeedai / deepspeedai/DeepSpeed
List Of Supported Models For DeepSpeed Inference; Add MPT And Falcon Inference Support[REQUEST]
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 43.1k
- Forks
- 5k
- Avg merge
- 4d 15h
- Merged PRs (30d)
- 112
Description
Is your feature request related to a problem? Please describe.
DeepSpeed inference is one of the best and or easiest ways to speed up LLM inference. However, with all the models coming out, it's not clear what models are supported. It would be great to either have documentation or some other tool that lists what models are supported.
Additionally, at this time, it seems to me that Falcon and MPT models are the best open-source commercial-friendly LLMs. Falcon for short context, MPT for long context(thanks to Alibi)
Describe the solution you'd like
Clear documentation of the method within DeepSpeed inference that lists what models are supported.
Additionally, if not already supported, add support for the best LLMs at this time, that being MPT and Falcon
Describe alternatives you've considered
If I understand the documentation, I think I can manually map some parts of these models with some of the kernel optimizations. Some of the kernels are likely compatible with Falcon and MPT. However, it would be nice if it was automatically supported like GPTJ.
I also thought about reading the source code to determine what models are supported but that's not ideal.
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start with the DeepSpeed inference documentation and source code, the entry points identified in the issue, to determine how supported models are currently represented. Define the supported-model list and assess what MPT and Falcon support would require; done means the documentation is clear and any added support is covered consistently.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python, pytorch
- Domain
- documentation, machine-learning
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 25/100