xiaomi-research/MiLMMT-46-4B-v0.1

Full fine-tune vs xiaomi-research/MiLMMT-46-4B-Pretrain. Fine-tune: 92% of the weights changed, across 34 of 34 layers.

Tensors changed: Embeddings 1/4, Norms 140/316, MLP 102/210, Attention 136/352, other 0/1, lm_head 1/1

3 derivatives checked

• xiaomi-research/MiLMMT-46-4B-v1.0: Full fine-tune
• mradermacher/MiLMMT-46-4B-v0.1-i1-GGUF: Quantization (fewer layers)
• mradermacher/MiLMMT-46-4B-v0.1-GGUF: Same weights, converted (fewer layers)

On the Hub · Downloads on Model Pulse