moonshotai/Kimi-VL-A3B-Thinking

Full fine-tune vs moonshotai/Kimi-VL-A3B-Instruct. Fine-tune: 100% of the weights changed, across 27 of 27 layers.

Tensors changed: Embeddings 4/4, Norms 194/194, MLP 2554/2554, Attention 162/189, other 58/58, lm_head 1/1

On the Hub · Downloads on Model Pulse