Skip to content
This repository has been archived by the owner on Sep 27, 2024. It is now read-only.

Commit

Permalink
Merge pull request #59 from lmstudio-ai/phi2-mit
Browse files Browse the repository at this point in the history
Update phi-2.json (MIT License)
  • Loading branch information
yagil authored Jan 6, 2024
2 parents 9aafb53 + 205a130 commit fa41a37
Showing 1 changed file with 2 additions and 2 deletions.
4 changes: 2 additions & 2 deletions models/phi-2.json
Original file line number Diff line number Diff line change
Expand Up @@ -2,7 +2,7 @@
"_descriptorVersion": "0.0.1",
"datePublished": "2023-12-13T21:22:37",
"name": "Phi 2",
"description": "Phi-2 is a 2.7 billion parameter Transformer model, an extension of Phi-1.5, with additional training data including synthetic NLP texts and curated web content. It demonstrates near state-of-the-art performance in benchmarks for common sense, language understanding, and logical reasoning within its parameter class. Phi-2 has not undergone reinforcement learning fine-tuning and is open-source, aimed at enabling safety research like toxicity reduction and bias understanding. It is designed for QA, chat, and code formats and has a context length of 2048 tokens. The model was trained on 250 billion tokens from a dataset combining AOAI GPT-3.5 synthetic data and filtered web data, using 1.4 trillion training tokens. It utilized 96xA100-80G GPUs over a span of 14 days. Phi-2 is intended solely for research use.",
"description": "Phi-2 is a 2.7 billion parameter Transformer model, an extension of Phi-1.5, with additional training data including synthetic NLP texts and curated web content. It demonstrates near state-of-the-art performance in benchmarks for common sense, language understanding, and logical reasoning within its parameter class. Phi-2 has not undergone reinforcement learning fine-tuning and is open-source, aimed at enabling safety research like toxicity reduction and bias understanding. It is designed for QA, chat, and code formats and has a context length of 2048 tokens. The model was trained on 250 billion tokens from a dataset combining AOAI GPT-3.5 synthetic data and filtered web data, using 1.4 trillion training tokens. It utilized 96xA100-80G GPUs over a span of 14 days. Phi-2 is released under the MIT license.",
"author": {
"name": "Microsoft Research",
"url": "https://www.microsoft.com/en-us/research/",
Expand Down Expand Up @@ -56,4 +56,4 @@
}
]
}
}
}

0 comments on commit fa41a37

Please sign in to comment.