Skip to content

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 sageattention for SAM3 (#13529)

Jukka Seppänen084e08c6e2d1
diff --git a/comfy/ldm/sam3/detector.py b/comfy/ldm/sam3/detector.py
index 6ae919a79..12d3a01ab 100644
--- a/comfy/ldm/sam3/detector.py
+++ b/comfy/ldm/sam3/detector.py
@@ -54,7 +54,7 @@ class SplitMHA(nn.Module):
if mask is not None and mask.ndim == 2:
mask = mask[:, None, None, :] # [B, T] -> [B, 1, 1, T] for SDPA broadcast
dtype = q.dtype # manual_cast may produce mixed dtypes
- out = optimized_attention(q, k.to(dtype), v.to(dtype), self.num_heads, mask=mask)
+ out = optimized_attention(q, k.to(dtype), v.to(dtype), self.num_heads, mask=mask, low_precision_attention=False)
return self.out_proj(out)
diff --git a/comfy/ldm/sam3/sam.py b/comfy/ldm/sam3/sam.py
index 272781d45..75cb457cf 100644
--- a/comfy/ldm/sam3/sam.py
+++ b/comfy/ldm/sam3/sam.py
@@ -40,7 +40,7 @@ class SAMAttention(nn.Module):
q = self.q_proj(q)
k = self.k_proj(k)
v = self.v_proj(v)
- return self.out_proj(optimized_attention(q, k, v, self.num_heads))
+ return self.out_proj(optimized_attention(q, k, v, self.num_heads, low_precision_attention=False))
class TwoWayAttentionBlock(nn.Module):
@@ -179,7 +179,7 @@ class Attention(nn.Module):
q, k, v = qkv.permute(2, 0, 3, 1, 4).unbind(dim=0)
if self.use_rope and freqs_cis is not None:
q, k = apply_rope(q, k, freqs_cis)
- return self.proj(optimized_attention(q, k, v, self.num_heads, skip_reshape=True))
+ return self.proj(optimized_attention(q, k, v, self.num_heads, skip_reshape=True, low_precision_attention=False))
class Block(nn.Module):
diff --git a/comfy/ldm/sam3/tracker.py b/comfy/ldm/sam3/tracker.py
index 6ff6369d1..8f7481003 100644
--- a/comfy/ldm/sam3/tracker.py
+++ b/comfy/ldm/sam3/tracker.py
@@ -364,7 +364,7 @@ class SplitAttn(nn.Module):
v = self.v_proj(v)
if rope is not None:
q, k = apply_rope_memory(q, k, rope, self.num_heads, num_k_exclude_rope)
- out = optimized_attention(q, k, v, self.num_heads)
+ out = optimized_attention(q, k, v, self.num_heads, low_precision_attention=False)
return self.out_proj(out)
@@ -657,7 +657,7 @@ class DecoupledMemoryAttnLayer(nn.Module):
v = self.self_attn_v_proj(normed)
if rope is not None:
q, k = apply_rope_memory(q, k, rope, self.num_heads, 0)
- x = x + self.self_attn_out_proj(optimized_attention(q, k, v, self.num_heads))
+ x = x + self.self_attn_out_proj(optimized_attention(q, k, v, self.num_heads, low_precision_attention=False))
# Decoupled cross-attention: fuse image and memory projections
normed = self.norm2(x)
@@ -668,7 +668,7 @@ class DecoupledMemoryAttnLayer(nn.Module):
v = self.cross_attn_v_proj(memory)
if rope is not None:
q, k = apply_rope_memory(q, k, rope, self.num_heads, num_k_exclude_rope)
- x = x + self.cross_attn_out_proj(optimized_attention(q, k, v, self.num_heads))
+ x = x + self.cross_attn_out_proj(optimized_attention(q, k, v, self.num_heads, low_precision_attention=False))
# FFN
x = x + self.linear2(F.gelu(self.linear1(self.norm3(x))))