upstream-archive/ComfyUI
1
0
Fork 0
mirror of https://github.com/Comfy-Org/ComfyUI.git synced 2026-09-21 05:27:57 -05:00
Code Issues Packages Projects Releases Wiki Activity

Disable ck attention when fp32 + low_precision_attention False. (#16029)

Browse Source
This commit is contained in:
comfyanonymous
2026-09-01 19:59:29 -04:00
committed by GitHub
parent 02aa70781a
commit 3258eb3a4d
1 changed files with 4 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
comfy/ldm/modules/attention.py
+4
View File
@@ -603,6 +603,8 @@ def _comfy_kitchen_int8_inputs(q, k, v, heads, mask, skip_reshape, enable_gqa):
@wrap_attn
def attention_comfy_kitchen_int8(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 and q.dtype == torch.float32:
return attention_pytorch(q, k, v, heads, mask=mask, attn_precision=attn_precision, skip_reshape=skip_reshape, skip_output_reshape=skip_output_reshape, **kwargs)
q, k, v, mask, b, dim_head = _comfy_kitchen_int8_inputs(
q, k, v, heads, mask, skip_reshape, kwargs.get("enable_gqa", False)
)
@@ -622,6 +624,8 @@ def _attention_comfy_kitchen_int8_containers(q, k, v, heads, mask=None, attn_pre
q = q.take()
k = k.take()
v = v.take()
if kwargs.get("low_precision_attention", True) is False and q.dtype == torch.float32:
return attention_pytorch(q, k, v, heads, mask=mask, attn_precision=attn_precision, skip_reshape=skip_reshape, skip_output_reshape=skip_output_reshape, **kwargs)
q, k, v, mask, b, dim_head = _comfy_kitchen_int8_inputs(
q, k, v, heads, mask, skip_reshape, kwargs.get("enable_gqa", False)
)