xiaomi-research/MiLMMT-46-4B-v1.0
Full fine-tune vs xiaomi-research/MiLMMT-46-4B-v0.1. Fine-tune: 90% of the weights changed, across 34 of 34 layers.
Tensors changed: Embeddings 1/4, Norms 87/316, MLP 102/210, Attention 136/352, other 0/1
2 derivatives checked
- mradermacher/MiLMMT-46-4B-v1.0-GGUF: Same weights, converted (fewer layers)
- translate-studio/MiLMMT-46-4B-v1.0-4bit-MLX: Quantization
1 other repos hold byte-for-byte identical weight files.