[MODEL REQUEST] Phi-3

[EwoutH]

Is your feature request related to a problem? Please describe.
The famous Phi-3 series models offer SOTA performance, especially in reasoning, math and coding. Microsoft released the models under the very permissive MIT license.

Details of model being requested

• Model name:
  • Phi-3-mini
  • Phi-3-small
  • Phi-3-medium
• Source repo link: https://huggingface.co/collections/microsoft/phi-3-6626e15e9585a200d2d761e3
• Research paper link: https://arxiv.org/abs/2404.14219
• Model use case: Standard SOTA LLM use cases.

Additional context for requested model
ONNX and GGUF formats are also available on HuggingFace, which should make the implementation easier. Quantized would be recommended, good Phi-3 quants are available from bartowski.

[mestrona-3]

Hi @EwoutH, thank you for the model request. We've received quite a few requests for Phi-3. It is on our backlog, though we are currently focused on Llama-3 🦙😊

[kory]

Phi3.5 is now in active development on our end. Stay tuned.