Added an argument llm_type to choose the llm model class that will be used in both training and inference.
The valid values for llm_type are taken from the existing 'model_type' variable from the corresponding models config class. More specifically the choices are: ['llava', 'llava_mistral','llava_neox']
When adding a new llm class, as long as the class is defined, the method __init_subclass__ will be invoked in the LlavaMetaModel base class and the Enum class variable ModelType will automatically be updated.
Added an argument llm_type to choose the llm model class that will be used in both training and inference.
The valid values for llm_type are taken from the existing 'model_type' variable from the corresponding models config class. More specifically the choices are:
['llava', 'llava_mistral','llava_neox']When adding a new llm class, as long as the class is defined, the method __init_subclass__ will be invoked in the LlavaMetaModel base class and the Enum class variable ModelType will automatically be updated.