pyvene.models.gpt_neox.modelings_intervenable_gpt_neox.gpt_neox_type_to_dimension_mapping

gpt_neox_type_to_dimension_mapping = {'attention_input': ('hidden_size',), 'attention_output': ('hidden_size',), 'attention_value_output': ('hidden_size',), 'block_input': ('hidden_size',), 'block_output': ('hidden_size',), 'head_attention_value_output': ('hidden_size/num_attention_heads',), 'mlp_activation': ('intermediate_size', 'hidden_size*4'), 'mlp_input': ('hidden_size',), 'mlp_output': ('hidden_size',), 'n_head': ('num_attention_heads',)}

gpt_neox model with LM head