proposed patch for the Tensor size mismatch error

Patch proposed for this error type:
"RuntimeError: The expanded size of the tensor (750) must match the existing size (751) at non-singleton dimension 1. Target sizes: [1, 750]. Tensor sizes: [1, 751]"

Patch issued on GitHub: https://github.com/vikhyat/moondream/issues/50#issuecomment-1971789438, by https://github.com/JohnnyStreet

modeling_phi.py

@@ -315,6 +315,7 @@ class CrossAttention(nn.Module):
                 dtype=scores.dtype,
                 device=scores.device,
             )
+            key_padding_mask = key_padding_mask[:, :seqlen_k]
             padding_mask.masked_fill_(key_padding_mask, 0.0)
             scores = scores + rearrange(padding_mask, "b s -> b 1 1 s")