diff --git a/src/cache_dit/distributed/transformers/boogu_image.py b/src/cache_dit/distributed/transformers/boogu_image.py index 87e46fb6..e4233091 100644 --- a/src/cache_dit/distributed/transformers/boogu_image.py +++ b/src/cache_dit/distributed/transformers/boogu_image.py @@ -228,6 +228,7 @@ def _patched_sdpa_call( scale=softmax_scale, enable_gqa=enable_gqa, cp_gqa_strategy=cp_gqa_strategy, + backend=getattr(self, '_attention_backend', None), cp_config=cp_config, ) hidden_states = hidden_states.flatten(-2) @@ -293,6 +294,7 @@ def _patched_varlen_call( scale=softmax_scale, enable_gqa=enable_gqa, cp_gqa_strategy=cp_gqa_strategy, + backend=getattr(self, '_attention_backend', None), cp_config=cp_config, ) hidden_states = hidden_states.flatten(-2)