Fix onnx non-expotable inplace aten op (#34376)

* fix onnx non-expotable inplace op * mistral, qwen2, qwen2_vl, starcoder2 * fixup copies
huggingface · Oct 29, 2024 · 565f0e9 · 565f0e9
1 parent dcfe3c7
commit 565f0e9
Show file tree

Hide file tree

Showing 10 changed files with 11 additions and 11 deletions.
diff --git a/src/transformers/models/mimi/modeling_mimi.py b/src/transformers/models/mimi/modeling_mimi.py
@@ -1156,7 +1156,7 @@ def _prepare_4d_causal_attention_mask_with_cache_position(
                     sliding_attend_mask = torch.arange(target_length, device=device) <= (
                         cache_position.reshape(-1, 1) - config.sliding_window
                     )
-                    diagonal_attend_mask |= sliding_attend_mask
+                    diagonal_attend_mask.bitwise_or_(sliding_attend_mask)
             causal_mask *= diagonal_attend_mask
             causal_mask = causal_mask[None, None, :, :].expand(batch_size, 1, -1, -1)
             if attention_mask is not None:

diff --git a/src/transformers/models/mistral/modeling_mistral.py b/src/transformers/models/mistral/modeling_mistral.py
@@ -961,7 +961,7 @@ def _prepare_4d_causal_attention_mask_with_cache_position(
                     sliding_attend_mask = torch.arange(target_length, device=device) <= (
                         cache_position.reshape(-1, 1) - config.sliding_window
                     )
-                    diagonal_attend_mask |= sliding_attend_mask
+                    diagonal_attend_mask.bitwise_or_(sliding_attend_mask)
             causal_mask *= diagonal_attend_mask
             causal_mask = causal_mask[None, None, :, :].expand(batch_size, 1, -1, -1)
             if attention_mask is not None:

diff --git a/src/transformers/models/mixtral/modeling_mixtral.py b/src/transformers/models/mixtral/modeling_mixtral.py
@@ -1174,7 +1174,7 @@ def _prepare_4d_causal_attention_mask_with_cache_position(
                     sliding_attend_mask = torch.arange(target_length, device=device) <= (
                         cache_position.reshape(-1, 1) - config.sliding_window
                     )
-                    diagonal_attend_mask |= sliding_attend_mask
+                    diagonal_attend_mask.bitwise_or_(sliding_attend_mask)
             causal_mask *= diagonal_attend_mask
             causal_mask = causal_mask[None, None, :, :].expand(batch_size, 1, -1, -1)
             if attention_mask is not None:

diff --git a/src/transformers/models/moshi/modeling_moshi.py b/src/transformers/models/moshi/modeling_moshi.py
@@ -1385,7 +1385,7 @@ def _prepare_4d_causal_attention_mask_with_cache_position(
                     sliding_attend_mask = torch.arange(target_length, device=device) <= (
                         cache_position.reshape(-1, 1) - config.sliding_window
                     )
-                    diagonal_attend_mask |= sliding_attend_mask
+                    diagonal_attend_mask.bitwise_or_(sliding_attend_mask)
             causal_mask *= diagonal_attend_mask
             causal_mask = causal_mask[None, None, :, :].expand(batch_size, 1, -1, -1)
             if attention_mask is not None:
@@ -1689,7 +1689,7 @@ def _prepare_4d_causal_attention_mask_with_cache_position(
                     sliding_attend_mask = torch.arange(target_length, device=device) <= (
                         cache_position.reshape(-1, 1) - config.sliding_window
                     )
-                    diagonal_attend_mask |= sliding_attend_mask
+                    diagonal_attend_mask.bitwise_or_(sliding_attend_mask)
             causal_mask *= diagonal_attend_mask
             causal_mask = causal_mask[None, None, :, :].expand(batch_size, 1, -1, -1)
             if attention_mask is not None:

diff --git a/src/transformers/models/phi3/modeling_phi3.py b/src/transformers/models/phi3/modeling_phi3.py
@@ -1139,7 +1139,7 @@ def _prepare_4d_causal_attention_mask_with_cache_position(
                     sliding_attend_mask = torch.arange(target_length, device=device) <= (
                         cache_position.reshape(-1, 1) - config.sliding_window
                     )
-                    diagonal_attend_mask |= sliding_attend_mask
+                    diagonal_attend_mask.bitwise_or_(sliding_attend_mask)
             causal_mask *= diagonal_attend_mask
             causal_mask = causal_mask[None, None, :, :].expand(batch_size, 1, -1, -1)
             if attention_mask is not None:

diff --git a/src/transformers/models/phimoe/modeling_phimoe.py b/src/transformers/models/phimoe/modeling_phimoe.py
@@ -1305,7 +1305,7 @@ def _prepare_4d_causal_attention_mask_with_cache_position(
                     sliding_attend_mask = torch.arange(target_length, device=device) <= (
                         cache_position.reshape(-1, 1) - config.sliding_window
                     )
-                    diagonal_attend_mask |= sliding_attend_mask
+                    diagonal_attend_mask.bitwise_or_(sliding_attend_mask)
             causal_mask *= diagonal_attend_mask
             causal_mask = causal_mask[None, None, :, :].expand(batch_size, 1, -1, -1)
             if attention_mask is not None:

diff --git a/src/transformers/models/qwen2/modeling_qwen2.py b/src/transformers/models/qwen2/modeling_qwen2.py
@@ -1059,7 +1059,7 @@ def _prepare_4d_causal_attention_mask_with_cache_position(
                     sliding_attend_mask = torch.arange(target_length, device=device) <= (
                         cache_position.reshape(-1, 1) - config.sliding_window
                     )
-                    diagonal_attend_mask |= sliding_attend_mask
+                    diagonal_attend_mask.bitwise_or_(sliding_attend_mask)
             causal_mask *= diagonal_attend_mask
             causal_mask = causal_mask[None, None, :, :].expand(batch_size, 1, -1, -1)
             if attention_mask is not None:

diff --git a/src/transformers/models/qwen2_moe/modeling_qwen2_moe.py b/src/transformers/models/qwen2_moe/modeling_qwen2_moe.py
@@ -1239,7 +1239,7 @@ def _prepare_4d_causal_attention_mask_with_cache_position(
                     sliding_attend_mask = torch.arange(target_length, device=device) <= (
                         cache_position.reshape(-1, 1) - config.sliding_window
                     )
-                    diagonal_attend_mask |= sliding_attend_mask
+                    diagonal_attend_mask.bitwise_or_(sliding_attend_mask)
             causal_mask *= diagonal_attend_mask
             causal_mask = causal_mask[None, None, :, :].expand(batch_size, 1, -1, -1)
             if attention_mask is not None:

diff --git a/src/transformers/models/qwen2_vl/modeling_qwen2_vl.py b/src/transformers/models/qwen2_vl/modeling_qwen2_vl.py
@@ -1321,7 +1321,7 @@ def _prepare_4d_causal_attention_mask_with_cache_position(
                     sliding_attend_mask = torch.arange(target_length, device=device) <= (
                         cache_position.reshape(-1, 1) - config.sliding_window
                     )
-                    diagonal_attend_mask |= sliding_attend_mask
+                    diagonal_attend_mask.bitwise_or_(sliding_attend_mask)
             causal_mask *= diagonal_attend_mask
             causal_mask = causal_mask[None, None, :, :].expand(batch_size, 1, -1, -1)
             if attention_mask is not None:

diff --git a/src/transformers/models/starcoder2/modeling_starcoder2.py b/src/transformers/models/starcoder2/modeling_starcoder2.py
@@ -1033,7 +1033,7 @@ def _prepare_4d_causal_attention_mask_with_cache_position(
                     sliding_attend_mask = torch.arange(target_length, device=device) <= (
                         cache_position.reshape(-1, 1) - config.sliding_window
                     )
-                    diagonal_attend_mask |= sliding_attend_mask
+                    diagonal_attend_mask.bitwise_or_(sliding_attend_mask)
             causal_mask *= diagonal_attend_mask
             causal_mask = causal_mask[None, None, :, :].expand(batch_size, 1, -1, -1)
             if attention_mask is not None: