Best Commercial Friendly Open Source LLMs: Falcon-40B And MPT-7B

In this video, we discuss the best large language models available, considering their performance, additional capabilities, and licensing for commercial applications. We take a look at the OpenLM leaderboard on Hugging Face and highlight the top model, Falcon 40B Instruct, which recently changed its license to the Apache license, making it more commercially friendly. We also introduce MPT, a model that performs well, has a commercially friendly license, and can handle longer inputs without retraining. Join us as we explore these powerful language models and their potential applications. Don't forget to like, subscribe, and join our Discord community for more AI discussions.
#ai #chatgpt #gpt4 #llm #llama #machinelearning

Time stamps:

00:00 - Intro
00:26 - Open LLM Leaderboard
00:47 - Falcon 40B
02:09 - Falcon Sequence Length
03:39 - MPT 7B
04:20 - MPT Adjustable Context Window
04:47 - Alibi Paper
06:14 - Additional MPT Details
07:40 - MPT Finetuning
08:00 - Summary
08:44 - Outro

[nupersu]

Can I run any of them with 6 gb vram? Are there models of MPT or Falcon in 4 bit precision?

[Srinivas_Billa]

In my experience falcon is like 4 times slower than OpenLlama