The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface. The fastest local inference engine in the world.
Disable sage attention on ace step 1.5 (#12297)
comfyanonymousa50c32d63fe5
diff --git a/comfy/ldm/ace/ace_step15.py b/comfy/ldm/ace/ace_step15.pyindex 17a37e573..f2b130bc1 100644--- a/comfy/ldm/ace/ace_step15.py+++ b/comfy/ldm/ace/ace_step15.py@@ -183,7 +183,7 @@ class AceStepAttention(nn.Module):else:attn_bias = window_bias- attn_output = optimized_attention(query_states, key_states, value_states, self.num_heads, attn_bias, skip_reshape=True)+ attn_output = optimized_attention(query_states, key_states, value_states, self.num_heads, attn_bias, skip_reshape=True, low_precision_attention=False)attn_output = self.o_proj(attn_output)return attn_outputdiff --git a/comfy/ldm/modules/attention.py b/comfy/ldm/modules/attention.pyindex ccf690945..10d051325 100644--- a/comfy/ldm/modules/attention.py+++ b/comfy/ldm/modules/attention.py@@ -524,6 +524,9 @@ def attention_pytorch(q, k, v, heads, mask=None, attn_precision=None, skip_resha@wrap_attndef attention_sage(q, k, v, heads, mask=None, attn_precision=None, skip_reshape=False, skip_output_reshape=False, **kwargs):+ if kwargs.get("low_precision_attention", True) is False:+ return attention_pytorch(q, k, v, heads, mask=mask, skip_reshape=skip_reshape, skip_output_reshape=skip_output_reshape, **kwargs)+exception_fallback = Falseif skip_reshape:b, _, _, dim_head = q.shape