Update transformer config to use joint_attention_dim and make it so that pos_embed_max_size is configurable in order to deal with multiple model sizes.
joint_attention_dim
pos_embed_max_size
· Sign up or log in to comment