ethanolivertroy/HackIDLE-NIST-Coder-MLX-4bit

Partial fine-tune vs mlx-community/Qwen2.5-Coder-7B-Instruct-4bit. Partial change: 49% of the weights changed (attention, mlp), in 16 of 28 layers.

Tensors changed: lm_head 0/3, Embeddings 0/3, Norms 0/57, MLP 144/252, Attention 192/420

On the Hub · Downloads on Model Pulse