fix code quality issues

huggingface · Dec 17, 2024 · 6ac713e · 6ac713e
1 parent 38ca81d
commit 6ac713e
Show file tree

Hide file tree

Showing 3 changed files with 37 additions and 22 deletions.
diff --git a/src/transformers/integrations/deepspeed.py b/src/transformers/integrations/deepspeed.py
@@ -494,12 +494,28 @@ def support_deepspeed_ulysses(module):
 
     original_forward = module.forward
 
-    def wrapped_forward(*args, **kwargs):
+    def wrapped_forward(
+        query_states: torch.Tensor,
+        key_states: torch.Tensor,
+        value_states: torch.Tensor,
+        attention_mask: torch.Tensor,
+        query_length: int,
+        is_causal: bool,
+        **kwargs,
+    ):
         # lazily set if sequence parallelism is enabled to ensure deepspeed is initialized first
         if is_deepspeed_sp_enabled():
             module.sp_group_size = deepspeed_groups._get_sequence_parallel_world_size()
 
-        return original_forward(*args, **kwargs)
+        return original_forward(
+            query_states,
+            key_states,
+            value_states,
+            attention_mask,
+            query_length,
+            is_causal,
+            **kwargs,
+        )
 
     module.forward = wrapped_forward
 

diff --git a/src/transformers/trainer.py b/src/transformers/trainer.py
@@ -3644,21 +3644,20 @@ def _finalize_inputs(
         labels: Optional[torch.LongTensor] = None,
         **model_kwargs,
     ):
+        inputs = {
+            "input_ids": input_ids,
+            "attention_mask": attention_mask,
+            "loss_mask": loss_mask,
+            "position_ids": position_ids,
+            "input_embeds": input_embeds,
+            "labels": labels,
+            **model_kwargs,
+        }
         if is_deepspeed_sp_enabled():
             ds_plugin = self.accelerator.state.deepspeed_plugin
             num_shards = ds_plugin.sequence_parallel_size
             rank = ds_plugin.sequence_parallel_rank
-            inputs = shard_inputs(
-                num_shards,
-                rank,
-                input_ids=input_ids,
-                attention_mask=attention_mask,
-                loss_mask=loss_mask,
-                position_ids=position_ids,
-                inputs_embeds=input_embeds,
-                labels=labels,
-                **model_kwargs,
-            )
+            inputs = shard_inputs(num_shards, rank, **inputs)
         return inputs
 
     def compute_loss_context_manager(self):

diff --git a/src/transformers/trainer_utils.py b/src/transformers/trainer_utils.py
@@ -898,12 +898,12 @@ def shard_tensor(tensor, num_shards, rank, dim=1):
 def shard_inputs(
     num_shards=1,
     rank=0,
-    input_ids: torch.LongTensor = None,
-    attention_mask: Optional[torch.Tensor] = None,
-    loss_mask: Optional[torch.Tensor] = None,
-    position_ids: Optional[torch.LongTensor] = None,
-    inputs_embeds: Optional[torch.FloatTensor] = None,
-    labels: Optional[torch.LongTensor] = None,
+    input_ids=None,
+    attention_mask=None,
+    loss_mask=None,
+    position_ids=None,
+    input_embeds=None,
+    labels=None,
     **kwargs,
 ):
     if num_shards == 1:
@@ -912,7 +912,7 @@ def shard_inputs(
             "attention_mask": attention_mask,
             "loss_mask": loss_mask,
             "position_ids": position_ids,
-            "inputs_embeds": inputs_embeds,
+            "input_embeds": input_embeds,
             "labels": labels,
             **kwargs,
         }
@@ -922,8 +922,8 @@ def shard_inputs(
         position_ids = position_ids.unsqueeze(0).expand(input_ids.shape[0], -1)
     result = kwargs
     for key, value in zip(
-        ["input_ids", "attention_mask", "loss_mask", "position_ids", "inputs_embeds", "labels"],
-        [input_ids, attention_mask, loss_mask, position_ids, inputs_embeds, labels],
+        ["input_ids", "attention_mask", "loss_mask", "position_ids", "input_embeds", "labels"],
+        [input_ids, attention_mask, loss_mask, position_ids, input_embeds, labels],
     ):
         if value is not None:
             result[key] = shard_tensor(value, num_shards, rank)