Mistral Small (4 bit weights) Vision language model (Mistral Small 3.1) using 4 bit weight quantization.