sageattention in hunyuanvideo meet attention_mask

[Andy0422]

How to handle the attention_mask in your sage kernel? e.g. hunyuanvideo calls AttnProcessor2.0. It seems the kernel does not have the attention_mask interface. Many thanks.

class HunyuanVideoAttnProcessor2_0:

       hidden_states = F.scaled_dot_product_attention(
            query, key, value, attn_mask=attention_mask, dropout_p=0.0, is_causal=False
        )

[jt-zhang]

For HunyuanVideo, you can modify the source code in their Github Repository rather than using diffuser.