History

Anchen 362e88a744 feat: move lora into mlx-lm (#337 ) * feat: Add lora and qlora training to mlx-lm --------- Co-authored-by: Awni Hannun <awni@apple.com>		2024-01-23 08:44:37 -08:00
..
generate.py	Phixtral (#290 )	2024-01-13 08:35:03 -08:00
phixtral.py	feat: move lora into mlx-lm (#337 )	2024-01-23 08:44:37 -08:00
README.md	Phixtral (#290 )	2024-01-13 08:35:03 -08:00
requirements.txt	Phixtral (#290 )	2024-01-13 08:35:03 -08:00

Phixtral

Phixtral is a Mixture of Experts (MoE) architecture inspired by Mixtral but made by combinding fine-tuned versions of Phi-2.¹²

Install the dependencies:

pip install -r requirements.txt

python generate.py \
  --model mlabonne/phixtral-4x2_8 \
  --prompt "write a quick sort in Python"

Run python generate.py --help to see all the options.

For more details on Phixtral, see the Hugging Face repo. ↩︎
For more details on Phi-2 see Microsoft's blog post and the Hugging Face repo. ↩︎