Text Generation
Transformers
Safetensors
minimax_m2
conversational
custom_code
fp8

Actual guide for fine-tuning?

#41
by openSourcerer9000 - opened

Can we get an actual guide for training a qlora in mlx with this architecture? It says possible with 256gb of vram, but this may just be AI generated and not actually supported yet for q lora. https://huggingface.co/MiniMaxAI/MiniMax-M2/blob/main/docs/mlx_deploy_guide.md

Thanks for the question.

Inference requires 256GB of memory, and fine-tuning will require even more. This document was completed with the help of community contributor Prince Canuma.

@prince-canuma , would you mind taking a look?

Sign up or log in to comment