some people get errors with view instead of reshape

AlexiaJM · web-flow · commit 5a58cd945aa8 · 2025-12-02T10:11:11.000-05:00
diff --git a/models/layers.py b/models/layers.py
@@ -131,7 +131,7 @@ def forward(self, cos_sin: CosSin, hidden_states: torch.Tensor) -> torch.Tensor:
         query, key, value = map(lambda t: einops.rearrange(t, 'B S H D -> B H S D'), (query, key, value)) # needed for scaled_dot_product_attention but not flash_attn_func
         attn_output = scaled_dot_product_attention(query=query, key=key, value=value, is_causal=self.causal)
         attn_output = einops.rearrange(attn_output, 'B H S D -> B S H D')
-        attn_output = attn_output.view(batch_size, seq_len, self.output_size)  # type: ignore
+        attn_output = attn_output.reshape(batch_size, seq_len, self.output_size)  # type: ignore
         return self.o_proj(attn_output)
 
 class LinearSwish(nn.Module):
