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.

Fix mac issue. (#12250)

comfyanonymous85fc35e8fa44
diff --git a/comfy/text_encoders/llama.py b/comfy/text_encoders/llama.py
index d1c628d20..f4ac224c6 100644
--- a/comfy/text_encoders/llama.py
+++ b/comfy/text_encoders/llama.py
@@ -628,10 +628,10 @@ class Llama2_(nn.Module):
mask = None
if attention_mask is not None:
mask = 1.0 - attention_mask.to(x.dtype).reshape((attention_mask.shape[0], 1, -1, attention_mask.shape[-1])).expand(attention_mask.shape[0], 1, seq_len, attention_mask.shape[-1])
- mask = mask.masked_fill(mask.to(torch.bool), float("-inf"))
+ mask = mask.masked_fill(mask.to(torch.bool), torch.finfo(x.dtype).min)
if seq_len > 1:
- causal_mask = torch.empty(past_len + seq_len, past_len + seq_len, dtype=x.dtype, device=x.device).fill_(float("-inf")).triu_(1)
+ causal_mask = torch.empty(past_len + seq_len, past_len + seq_len, dtype=x.dtype, device=x.device).fill_(torch.finfo(x.dtype).min).triu_(1)
if mask is not None:
mask += causal_mask
else: