microsoft/MagenticBrain

Full fine-tune vs Qwen/Qwen3-14B. Fine-tune: 95% of the weights changed, across 40 of 40 layers.

Tensors changed: Embeddings 0/1, Norms 103/161, MLP 120/120, Attention 160/160, lm_head 1/1

2 derivatives checked

• devehz/MagenticBrain-GGUF: Same weights, converted
• mlx-community/MagenticBrain-4bit: Quantization

On the Hub · Downloads on Model Pulse