[BLOOM] Deprecate `position_ids` (#18342)

src/transformers/models/bloom/modeling_bloom.py

@@ -15,6 +15,7 @@
 """PyTorch BLOOM model."""
 
 import math
+import warnings
 from typing import Tuple, Union
 
 import torch
@@ -522,11 +523,6 @@ BLOOM_INPUTS_DOCSTRING = r"""
             - 0 for tokens that are **masked**.
 
             [What are attention masks?](../glossary#attention-mask)
-        position_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
-            Indices of positions of each input sequence tokens in the position embeddings. Selected in the range `[0,
-            config.max_position_embeddings - 1]`.
-
-            [What are position IDs?](../glossary#position-ids)
         head_mask (`torch.FloatTensor` of shape `(num_heads,)` or `(num_layers, num_heads)`, *optional*):
             Mask to nullify selected heads of the self-attention modules. Mask values selected in `[0, 1]`:
 
@@ -617,14 +613,24 @@ class BloomModel(BloomPreTrainedModel):
         input_ids=None,
         past_key_values=None,
         attention_mask=None,
-        position_ids=None,
         head_mask=None,
         inputs_embeds=None,
         use_cache=None,
         output_attentions=None,
         output_hidden_states=None,
         return_dict=None,
+        **deprecated_arguments
     ) -> Union[Tuple[torch.Tensor], BaseModelOutputWithPastAndCrossAttentions]:
+        if deprecated_arguments.pop("position_ids", False) is not False:
+            # `position_ids` could have been `torch.Tensor` or `None` so defaulting pop to `False` allows to detect if users were passing explicitly `None`
+            warnings.warn(
+                "`position_ids` have no functionality in BLOOM and will be removed in v5.0.0. You can safely ignore"
+                " passing `position_ids`.",
+                FutureWarning,
+            )
+        if len(deprecated_arguments) > 0:
+            raise ValueError(f"Got unexpected arguments: {deprecated_arguments}")
+
         output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
         output_hidden_states = (
             output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
@@ -772,16 +778,7 @@ class BloomForCausalLM(BloomPreTrainedModel):
             input_ids = input_ids[:, -1].unsqueeze(-1)
 
         attention_mask = kwargs.get("attention_mask", None)
-        position_ids = kwargs.get("position_ids", None)
-
-        if attention_mask is not None and position_ids is None:
-            # create position_ids on the fly for batch generation
-            position_ids = attention_mask.long().cumsum(-1) - 1
-            position_ids.masked_fill_(attention_mask == 0, 1)
-            if past:
-                position_ids = position_ids[:, -1].unsqueeze(-1)
-        else:
-            position_ids = None
+
         return {
             "input_ids": input_ids,
             "past_key_values": past,
@@ -801,7 +798,6 @@ class BloomForCausalLM(BloomPreTrainedModel):
         input_ids=None,
         past_key_values=None,
         attention_mask=None,
-        position_ids=None,
         head_mask=None,
         inputs_embeds=None,
         labels=None,
@@ -809,6 +805,7 @@ class BloomForCausalLM(BloomPreTrainedModel):
         output_attentions=None,
         output_hidden_states=None,
         return_dict=None,
+        **deprecated_arguments
     ) -> Union[Tuple[torch.Tensor], CausalLMOutputWithCrossAttentions]:
         r"""
         labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
@@ -816,13 +813,22 @@ class BloomForCausalLM(BloomPreTrainedModel):
             `labels = input_ids` Indices are selected in `[-100, 0, ..., config.vocab_size]` All labels set to `-100`
             are ignored (masked), the loss is only computed for labels in `[0, ..., config.vocab_size]`
         """
+        if deprecated_arguments.pop("position_ids", False) is not False:
+            # `position_ids` could have been `torch.Tensor` or `None` so defaulting pop to `False` allows to detect if users were passing explicitly `None`
+            warnings.warn(
+                "`position_ids` have no functionality in BLOOM and will be removed in v5.0.0. You can safely ignore"
+                " passing `position_ids`.",
+                FutureWarning,
+            )
+        if len(deprecated_arguments) > 0:
+            raise ValueError(f"Got unexpected arguments: {deprecated_arguments}")
+
         return_dict = return_dict if return_dict is not None else self.config.use_return_dict
 
         transformer_outputs = self.transformer(
             input_ids,
             past_key_values=past_key_values,
             attention_mask=attention_mask,
-            position_ids=position_ids,
             head_mask=head_mask,
             inputs_embeds=inputs_embeds,
             use_cache=use_cache,
@@ -907,7 +913,6 @@ class BloomForSequenceClassification(BloomPreTrainedModel):
         input_ids=None,
         past_key_values=None,
         attention_mask=None,
-        position_ids=None,
         head_mask=None,
         inputs_embeds=None,
         labels=None,
@@ -915,6 +920,7 @@ class BloomForSequenceClassification(BloomPreTrainedModel):
         output_attentions=None,
         output_hidden_states=None,
         return_dict=None,
+        **deprecated_arguments
     ) -> Union[Tuple[torch.Tensor], SequenceClassifierOutputWithPast]:
         r"""
         labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
@@ -922,6 +928,15 @@ class BloomForSequenceClassification(BloomPreTrainedModel):
             config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If
             `config.num_labels > 1` a classification loss is computed (Cross-Entropy).
         """
+        if deprecated_arguments.pop("position_ids", False) is not False:
+            # `position_ids` could have been `torch.Tensor` or `None` so defaulting pop to `False` allows to detect if users were passing explicitly `None`
+            warnings.warn(
+                "`position_ids` have no functionality in BLOOM and will be removed in v5.0.0. You can safely ignore"
+                " passing `position_ids`.",
+                FutureWarning,
+            )
+        if len(deprecated_arguments) > 0:
+            raise ValueError(f"Got unexpected arguments: {deprecated_arguments}")
 
         return_dict = return_dict if return_dict is not None else self.config.use_return_dict
 
@@ -929,7 +944,6 @@ class BloomForSequenceClassification(BloomPreTrainedModel):
             input_ids,
             past_key_values=past_key_values,
             attention_mask=attention_mask,
-            position_ids=position_ids,
             head_mask=head_mask,
             inputs_embeds=inputs_embeds,
             use_cache=use_cache,
@@ -1036,7 +1050,6 @@ class BloomForTokenClassification(BloomPreTrainedModel):
         input_ids=None,
         past_key_values=None,
         attention_mask=None,
-        position_ids=None,
         head_mask=None,
         inputs_embeds=None,
         labels=None,
@@ -1044,6 +1057,7 @@ class BloomForTokenClassification(BloomPreTrainedModel):
         output_attentions=None,
         output_hidden_states=None,
         return_dict=None,
+        **deprecated_arguments
     ) -> Union[Tuple[torch.Tensor], TokenClassifierOutput]:
         r"""
         labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
@@ -1051,6 +1065,15 @@ class BloomForTokenClassification(BloomPreTrainedModel):
             config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If
             `config.num_labels > 1` a classification loss is computed (Cross-Entropy).
         """
+        if deprecated_arguments.pop("position_ids", False) is not False:
+            # `position_ids` could have been `torch.Tensor` or `None` so defaulting pop to `False` allows to detect if users were passing explicitly `None`
+            warnings.warn(
+                "`position_ids` have no functionality in BLOOM and will be removed in v5.0.0. You can safely ignore"
+                " passing `position_ids`.",
+                FutureWarning,
+            )
+        if len(deprecated_arguments) > 0:
+            raise ValueError(f"Got unexpected arguments: {deprecated_arguments}")
 
         return_dict = return_dict if return_dict is not None else self.config.use_return_dict
 
@@ -1058,7 +1081,6 @@ class BloomForTokenClassification(BloomPreTrainedModel):
             input_ids,
             past_key_values=past_key_values,
             attention_mask=attention_mask,
-            position_ids=position_ids,
             head_mask=head_mask,
             inputs_embeds=inputs_embeds,
             use_cache=use_cache,