Revert "Error (also in original) model, scaling only q matrix not qk.…

…T dot product (qk.T/sqrt(dim_per_head))" (#22444)

Revert "Error (also in original) model, scaling only q matrix not qk.T dot product (qk.T/sqrt(dim_per_head)) (#21627)"

This reverts commit bad8300.

src/transformers/models/flaubert/modeling_flaubert.py

@@ -172,7 +172,8 @@ def unshape(x):
                     k, v = cache[self.layer_id]
             cache[self.layer_id] = (k, v)
 
-        scores = torch.matmul(q, k.transpose(2, 3)) / math.sqrt(dim_per_head)  # (bs, n_heads, qlen, klen)
+        q = q / math.sqrt(dim_per_head)  # (bs, n_heads, qlen, dim_per_head)
+        scores = torch.matmul(q, k.transpose(2, 3))  # (bs, n_heads, qlen, klen)
         mask = (mask == 0).view(mask_reshape).expand_as(scores)  # (bs, n_heads, qlen, klen)
         scores.masked_fill_(mask, torch.finfo(scores.dtype).min)  # (bs, n_heads, qlen, klen)
 

src/transformers/models/xlm/modeling_xlm.py

@@ -176,7 +176,8 @@ def unshape(x):
                     k, v = cache[self.layer_id]
             cache[self.layer_id] = (k, v)
 
-        scores = torch.matmul(q, k.transpose(2, 3)) / math.sqrt(dim_per_head)  # (bs, n_heads, qlen, klen)
+        q = q / math.sqrt(dim_per_head)  # (bs, n_heads, qlen, dim_per_head)
+        scores = torch.matmul(q, k.transpose(2, 3))  # (bs, n_heads, qlen, klen)
         mask = (mask == 0).view(mask_reshape).expand_as(scores)  # (bs, n_heads, qlen, klen)
         scores.masked_fill_(mask, torch.finfo(scores.dtype).min)  # (bs, n_heads, qlen, klen)