pyvene.models.llava.modelings_intervenable_llava.llava_lm_type_to_dimension_mapping

llava_lm_type_to_dimension_mapping = {'attention_input': ('text_config.hidden_size',), 'attention_output': ('text_config.hidden_size',), 'attention_value_output': ('text_config.hidden_size',), 'block_input': ('text_config.hidden_size',), 'block_output': ('text_config.hidden_size',), 'head_attention_value_output': ('text_config.hidden_size/text_config.num_attention_heads',), 'head_key_output': ('text_config.hidden_size/text_config.num_attention_heads',), 'head_query_output': ('text_config.hidden_size/text_config.num_attention_heads',), 'head_value_output': ('text_config.hidden_size/text_config.num_attention_heads',), 'key_output': ('text_config.num_key_value_heads*text_config.hidden_size/text_config.num_attention_heads',), 'mlp_activation': ('text_config.intermediate_size',), 'mlp_input': ('text_config.hidden_size',), 'mlp_output': ('text_config.hidden_size',), 'n_head': ('text_config.num_attention_heads',), 'n_kv_head': ('text_config.num_key_value_heads',), 'query_output': ('text_config.hidden_size',), 'value_output': ('text_config.num_key_value_heads*text_config.hidden_size/text_config.num_attention_heads',)}

llava model with classifier head