mirror of
https://github.com/comfyanonymous/ComfyUI.git
synced 2025-09-11 20:17:30 +00:00
Properly fix attention masks in CLIP with batches.
This commit is contained in:
@@ -97,7 +97,7 @@ class CLIPTextModel_(torch.nn.Module):
|
||||
x = self.embeddings(input_tokens)
|
||||
mask = None
|
||||
if attention_mask is not None:
|
||||
mask = 1.0 - attention_mask.to(x.dtype).reshape((attention_mask.shape[0], -1, attention_mask.shape[-1])).expand(attention_mask.shape[0], attention_mask.shape[-1], attention_mask.shape[-1])
|
||||
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, attention_mask.shape[-1], attention_mask.shape[-1])
|
||||
mask = mask.masked_fill(mask.to(torch.bool), float("-inf"))
|
||||
|
||||
causal_mask = torch.empty(x.shape[1], x.shape[1], dtype=x.dtype, device=x.device).fill_(float("-inf")).triu_(1)
|
||||
|
Reference in New Issue
Block a user