MPT-7B

lastmod 2023-05-10created 2023-05-101 edits

A commercially usable open-source LLM

Features

• A transformer trained on 1 trillion tokens
• Quality similar to LLaMA-7B, but open source and commercially usable
• Three additional fine-tuned models were also released
  • MPT-7B-Instruct, MPT-7B-Chat, MPT-7B-StoryWriter-65K+ (supports 65K tokens)
• Trained on the MosaicML platform in 9.5 days at a cost of about ~$200K; you can train, fine-tune, and deploy your own MPT models
• MPT: Mosaic Pretrained Transformers

See also

• LLM

Favorite site

• Introducing MPT-7B: A New Standard for Open-Source, Commercially Usable LLMs