Do you plan on finetuning smaller model?

by - opened

Running 175B model is not possible on consumer hardware so a lower parameters model would be appreciate.

wonderful, i really really love this model, it is upto gpt 3.5 size and even in capacity as i test it

Can I use this with normal API call for making instruction based inferences in Python ? If yes, then how can someone guide me ?

well, the model seems to be instruction tuned model, so it seems to follow instructions and to perform infrence you have to host it on AWS sage maker because now i don't think hugging faces has deployed any inference API.

SambaNova Systems org

@AlaminI is right, BLOOMChat has OCK data in the instruction tuning phase. BLOOMChat is mostly targeted for multilingual chat, but it can perform instruction following in normal NLP tasks in English.

For the online live demo, it has certain system prompt in the backend optimized for chat experience. Thus if you want to experiment conventional NLP task capabilities, using the baremetal weights without the conversation system prompt will be recommended.

We are also excited to take feedbacks from developers using different language on the evaluation for conventional instruction following in non-English scenarios.

it is actually good, i tried translating english to nigerian specific pidgin english, and it seems to be following the instructions pretty good, but it seems there is no enough fine tuning on nigerian pidgin english, and i also tested it with arabic, it performs well following the instructions.

Sign up or log in to comment