introduce `logger.warning_once` and use it for grad checkpointing code (#21804)

* logger.warning_once

* style

src/transformers/models/git/modeling_git.py

@@ -444,7 +444,7 @@ class GitEncoder(nn.Module):
 
             if self.gradient_checkpointing and self.training:
                 if use_cache:
-                    logger.warning(
+                    logger.warning_once(
                         "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
                     )
                     use_cache = False

src/transformers/models/gpt2/modeling_gpt2.py

@@ -853,7 +853,7 @@ class GPT2Model(GPT2PreTrainedModel):
 
         if self.gradient_checkpointing and self.training:
             if use_cache:
-                logger.warning(
+                logger.warning_once(
                     "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
                 )
                 use_cache = False

src/transformers/models/gpt_neo/modeling_gpt_neo.py

@@ -589,7 +589,7 @@ class GPTNeoModel(GPTNeoPreTrainedModel):
 
         if self.gradient_checkpointing and self.training:
             if use_cache:
-                logger.warning(
+                logger.warning_once(
                     "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
                 )
                 use_cache = False

src/transformers/models/gptj/modeling_gptj.py

@@ -653,7 +653,7 @@ class GPTJModel(GPTJPreTrainedModel):
 
             if self.gradient_checkpointing and self.training:
                 if use_cache:
-                    logger.warning(
+                    logger.warning_once(
                         "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
                     )
                     use_cache = False

src/transformers/models/imagegpt/modeling_imagegpt.py

@@ -812,7 +812,7 @@ class ImageGPTModel(ImageGPTPreTrainedModel):
 
             if self.gradient_checkpointing and self.training:
                 if use_cache:
-                    logger.warning(
+                    logger.warning_once(
                         "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
                     )
                     use_cache = False

src/transformers/models/layoutlm/modeling_layoutlm.py

@@ -479,7 +479,7 @@ class LayoutLMEncoder(nn.Module):
 
             if self.gradient_checkpointing and self.training:
                 if use_cache:
-                    logger.warning(
+                    logger.warning_once(
                         "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
                     )
                     use_cache = False

src/transformers/models/led/modeling_led.py

@@ -2136,7 +2136,7 @@ class LEDDecoder(LEDPreTrainedModel):
 
             if self.gradient_checkpointing and self.training:
                 if use_cache:
-                    logger.warning(
+                    logger.warning_once(
                         "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
                     )
                     use_cache = False

src/transformers/models/m2m_100/modeling_m2m_100.py

@@ -1055,7 +1055,7 @@ class M2M100Decoder(M2M100PreTrainedModel):
 
                 if self.gradient_checkpointing and self.training:
                     if use_cache:
-                        logger.warning(
+                        logger.warning_once(
                             "`use_cache=True` is incompatible with gradient checkpointing. Setting"
                             " `use_cache=False`..."
                         )

src/transformers/models/marian/modeling_marian.py

@@ -1020,7 +1020,7 @@ class MarianDecoder(MarianPreTrainedModel):
 
             if self.gradient_checkpointing and self.training:
                 if use_cache:
-                    logger.warning(
+                    logger.warning_once(
                         "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
                     )
                     use_cache = False

src/transformers/models/markuplm/modeling_markuplm.py

@@ -641,7 +641,7 @@ class MarkupLMEncoder(nn.Module):
 
             if self.gradient_checkpointing and self.training:
                 if use_cache:
-                    logger.warning(
+                    logger.warning_once(
                         "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
                     )
                     use_cache = False

src/transformers/models/mbart/modeling_mbart.py

@@ -1069,7 +1069,7 @@ class MBartDecoder(MBartPreTrainedModel):
 
             if self.gradient_checkpointing and self.training:
                 if use_cache:
-                    logger.warning(
+                    logger.warning_once(
                         "`use_cache=True` is incompatible with gradient checkpointing`. Setting `use_cache=False`..."
                     )
                     use_cache = False

src/transformers/models/megatron_bert/modeling_megatron_bert.py

@@ -544,7 +544,7 @@ class MegatronBertEncoder(nn.Module):
 
             if self.gradient_checkpointing and self.training:
                 if use_cache:
-                    logger.warning(
+                    logger.warning_once(
                         "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
                     )
                     use_cache = False

src/transformers/models/mt5/modeling_mt5.py

@@ -1008,7 +1008,7 @@ class MT5Stack(MT5PreTrainedModel):
 
             if self.gradient_checkpointing and self.training:
                 if use_cache:
-                    logger.warning(
+                    logger.warning_once(
                         "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
                     )
                     use_cache = False

src/transformers/models/mvp/modeling_mvp.py

@@ -1212,7 +1212,7 @@ class MvpDecoder(MvpPreTrainedModel):
 
             if self.gradient_checkpointing and self.training:
                 if use_cache:
-                    logger.warning(
+                    logger.warning_once(
                         "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
                     )
                     use_cache = False

src/transformers/models/nezha/modeling_nezha.py

@@ -571,7 +571,7 @@ class NezhaEncoder(nn.Module):
 
             if self.gradient_checkpointing and self.training:
                 if use_cache:
-                    logger.warning(
+                    logger.warning_once(
                         "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
                     )
                     use_cache = False

src/transformers/models/opt/modeling_opt.py

@@ -671,7 +671,7 @@ class OPTDecoder(OPTPreTrainedModel):
 
             if self.gradient_checkpointing and self.training:
                 if use_cache:
-                    logger.warning(
+                    logger.warning_once(
                         "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
                     )
                     use_cache = False

src/transformers/models/pegasus/modeling_pegasus.py

@@ -1070,7 +1070,7 @@ class PegasusDecoder(PegasusPreTrainedModel):
 
             if self.gradient_checkpointing and self.training:
                 if use_cache:
-                    logger.warning(
+                    logger.warning_once(
                         "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
                     )
                     use_cache = False

src/transformers/models/pegasus_x/modeling_pegasus_x.py

@@ -1311,7 +1311,7 @@ class PegasusXDecoder(PegasusXPreTrainedModel):
 
             if self.gradient_checkpointing and self.training:
                 if use_cache:
-                    logger.warning(
+                    logger.warning_once(
                         "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
                     )
                     use_cache = False

src/transformers/models/plbart/modeling_plbart.py

@@ -1048,7 +1048,7 @@ class PLBartDecoder(PLBartPreTrainedModel):
 
             if self.gradient_checkpointing and self.training:
                 if use_cache:
-                    logger.warning(
+                    logger.warning_once(
                         "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
                     )
                     use_cache = False

src/transformers/models/prophetnet/modeling_prophetnet.py

@@ -1572,7 +1572,7 @@ class ProphetNetDecoder(ProphetNetPreTrainedModel):
 
         if self.gradient_checkpointing and self.training:
             if use_cache:
-                logger.warning(
+                logger.warning_once(
                     "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
                 )
                 use_cache = False