Add a Mistral3 multimodal vision-language model to the library. The implementation should include:

Add a Mistral3 multimodal vision-language model to the library. The implementation should include:: a task in HF ML Tasksmith (Harbor dataset). Mistral3Config ( model type="mistral3" ) with sub-configs vision config (default: Pixtral) and text config (default: Mistral), and fields: image token…

The task

**`Mistral3Config`** (`model_type="mistral3"`) with sub-configs `vision_config` (default: Pixtral) and `text_config` (default: Mistral), and fields: `image_token_index` (default 10), `projector_hidden_act` (default `"gelu"`), `vision_feature_layer` (default -1), `multimodal_projector_bias` (default `False`),…

Part of FineEnvs/HF_ML_Tasksmith.