I love both streamlit and gradio as UI full stack ML python development boot up SDK/LIB supertankers that have maybe best mashup of important imports in 2023.
I find that torch and transformers while large and affecting a boot and compile time, that my ultimate goto 3 ingredient recipes remain:
gradio+torch+transformers
-and-
streamlit+torch+transformers
Those two if blended to a boot docker will immediately simplify the triumvirate of SL, FT, ICL, and RLHF making cool things like hellaswag, qa seq to seq, mlm, and rlhf or some form of episodic memory cognitive style with an emotion sentiment response from a human to tune up…
Consider that next, perhaps gradio first? In my AI classes, about 19 of 30 like the easy button to gradio for two lines of code to run models their first try (and some with no previous development experience which is magic!_)
Want to say too - the Freddy A Boulton kitchen sink was a hit with new students of AI. Partly due to full coverage of cool AI-UI - well done!
Also they love MediaPipe right now… Can we add some native support for it? If you haven’t seen 30FPS wild 3D positioning of human vertices yet check it out here! Spaces - Hugging Face
I think you are the best in AI. I’m also sharing everything I can about BLOOM, the open source multilingual model I think is the next world hit for the model of the year provided we solve the creative needs of community like ROOTS does and maybe InstructGPT, Anthropic, RLHF, ICL methods and dataset theories get better at simulating our neocortex features including human memory types.
Keep up the awesome work!
–Aaron @awacke1