mlx-community/SmolVLM2-2.2B-Instruct-mlx

Same weights, converted vs HuggingFaceTB/SmolVLM2-2.2B-Instruct. Same weights stored in another dtype (F32 → BF16).

Tensors changed: other 0/1, Embeddings 0/3, Norms 0/159, MLP 0/180, Attention 0/312, lm_head 0/1

On the Hub · Downloads on Model Pulse