Closed HaiyangDING closed 1 week ago
Currently we do not plan to run Phi-3 as serverless API. You can make a photon and run a dedicated deployment on our platform.
Currently we do not plan to run Phi-3 as serverless API. You can make a photon and run a dedicated deployment on our platform.
Hi, thanks for the kind instruction. This is what I am trying to do now. Let me close the issue for now.
Is your feature request related to a problem? Please describe.
Phi-3 is a family of small language models and multi-modal models from Microsoft. Having some Phi-3 models available in Leptonai playground not only enriches the Leptonai ecosystem, but also benefits Phi-3 as more users could be able to access it with ease (thanks to Lepton!).
Describe the solution you'd like
I would like to work on bringing some Phi-3 models to Lepton AI Playground as new options (a new block along side with Llama, Mistral, OpenChat, etc.). The plan is to start with microsoft/Phi-3-vision-128k-instruct.
However I did not find the doc on how to add a new model to the playground. Is there a doc for me to start with? (I am trying to build a
photon
locally now.)Please let me know how you think about this idea. I volunteer to do all the required work.
Describe alternatives you've considered
I could also try to build new Phi-3
photon
using Lepton framework and share with users in some other way. But I really want to make the model(s) available on Lepton playground to make an easy life for all the users.Additional context
Phi-3 models are under MIT license. The license information