[baichuan] model_name = Baichuan2_7B head_num = 32 size_per_head = 128 inter_size = 11008 num_layer = 32 rotary_embedding = 128 layernorm_eps = 1e-06 vocab_size = 125696 start_id = 1 end_id = 2 tensor_para_size = 1 weight_data_type = fp16