Open Source LLMs Score Again! Introducing MPT-7B, Commercially Usable and Free of Charge

preview_player
Показать описание
In this video, I introduce MPT-7B, the latest development in the MosaicML Foundation Series. It is a commercially-usable, open-source LLM that was trained on 1 trillion tokens of text and code.

Watch as we compare the MPT-7B to popular open-source LLMs like LLaMA series from Meta, the Pythia series from EleutherAI, the StableLM series from StabilityAI, and the OpenLLaMA model from Berkeley AI Research. See how MPT-7B surpasses these models in terms of commercial usability, training volume, input handling, training speed, and inference optimization.

The blog post can be found here:
Рекомендации по теме
Комментарии
Автор

I've experimented with it and it's not amazing, but Alibi sure is.

Looking forward to seeing that tech make it into a better 13b model

priestesslucy
Автор

Thanks!! it seems like the model is introduced in May 5. Why in the video I get the impression like it is introduced today?:) Maybe the upload date doesn’t match with the date the video was prepared?

nctamer
Автор

I wish someone would discuss the cost to run this model vs it being “free”. From what I can tell cloud based GPUs that are not going to be painfully slow are about $1.66/hr USD. Who of us that are just trying to get something started can afford or are willing to spend $1200/mth USD for 1 GPU. I’m a bit of a noob … am I wrong?

dbriand
Автор

how do I make this.. work? like if I wanted to use mpt-7-chat model like chatgpt, how do I set that up? sorry for being a braindead

pawekozowski
Автор

I am sorry to say mpt is horrible. I tried to use it, trained it using LoRA and it was still garbage. Do not waste your time.

timothymaggenti