normalize keys_to_ignore

src/transformers/models/bart/modeling_bart.py

@@ -497,7 +497,7 @@ class BartPretrainedModel(PreTrainedModel):
     config_class = BartConfig
     base_model_prefix = "model"
     supports_gradient_checkpointing = True
-    _keys_to_ignore_on_load_unexpected = [r"encoder\.version", r"decoder\.version"]
+    _keys_to_ignore_on_load_unexpected = [r"encoder.version", r"decoder.version"]
 
     def _init_weights(self, module):
         std = self.config.init_std
@@ -1272,7 +1272,7 @@ def forward(
 )
 class BartForConditionalGeneration(BartPretrainedModel):
     base_model_prefix = "model"
-    _keys_to_ignore_on_load_missing = [r"final_logits_bias", r"lm_head\.weight"]
+    _keys_to_ignore_on_load_missing = [r"final_logits_bias", r"lm_head.weight"]
 
     def __init__(self, config: BartConfig):
         super().__init__(config)

src/transformers/models/bigbird_pegasus/modeling_bigbird_pegasus.py

@@ -2476,7 +2476,7 @@ def forward(
 # Copied from transformers.models.bart.modeling_bart.BartForConditionalGeneration with Bart->BigBirdPegasus, BART->BIGBIRD_PEGASUS
 class BigBirdPegasusForConditionalGeneration(BigBirdPegasusPreTrainedModel):
     base_model_prefix = "model"
-    _keys_to_ignore_on_load_missing = [r"final_logits_bias", r"lm_head\.weight"]
+    _keys_to_ignore_on_load_missing = [r"final_logits_bias", r"lm_head.weight"]
 
     def __init__(self, config: BigBirdPegasusConfig):
         super().__init__(config)

src/transformers/models/blenderbot/modeling_blenderbot.py

@@ -1226,9 +1226,9 @@ class BlenderbotForConditionalGeneration(BlenderbotPreTrainedModel):
     base_model_prefix = "model"
     _keys_to_ignore_on_load_missing = [
         r"final_logits_bias",
-        r"encoder\.version",
-        r"decoder\.version",
-        r"lm_head\.weight",
+        r"encoder.version",
+        r"decoder.version",
+        r"lm_head.weight",
     ]
 
     def __init__(self, config: BlenderbotConfig):

src/transformers/models/blenderbot_small/modeling_blenderbot_small.py

@@ -1208,9 +1208,9 @@ class BlenderbotSmallForConditionalGeneration(BlenderbotSmallPreTrainedModel):
     base_model_prefix = "model"
     _keys_to_ignore_on_load_missing = [
         r"final_logits_bias",
-        r"encoder\.version",
-        r"decoder\.version",
-        r"lm_head\.weight",
+        r"encoder.version",
+        r"decoder.version",
+        r"lm_head.weight",
     ]
 
     def __init__(self, config: BlenderbotSmallConfig):

src/transformers/models/convbert/modeling_convbert.py

@@ -251,7 +251,7 @@ class ConvBertPreTrainedModel(PreTrainedModel):
     base_model_prefix = "convbert"
     supports_gradient_checkpointing = True
     authorized_missing_keys = [r"position_ids"]
-    authorized_unexpected_keys = [r"convbert\.embeddings_project\.weight", r"convbert\.embeddings_project\.bias"]
+    authorized_unexpected_keys = [r"convbert.embeddings_project.weight", r"convbert.embeddings_project.bias"]
 
     def _init_weights(self, module):
         """Initialize the weights"""

src/transformers/models/electra/modeling_electra.py

@@ -670,7 +670,7 @@ class ElectraPreTrainedModel(PreTrainedModel):
     base_model_prefix = "electra"
     supports_gradient_checkpointing = True
     _keys_to_ignore_on_load_missing = [r"position_ids"]
-    _keys_to_ignore_on_load_unexpected = [r"electra\.embeddings_project\.weight", r"electra\.embeddings_project\.bias"]
+    _keys_to_ignore_on_load_unexpected = [r"electra.embeddings_project.weight", r"electra.embeddings_project.bias"]
 
     # Copied from transformers.models.bert.modeling_bert.BertPreTrainedModel._init_weights
     def _init_weights(self, module):

src/transformers/models/gpt2/modeling_gpt2.py

@@ -1328,7 +1328,7 @@ def _reorder_cache(past: Tuple[Tuple[torch.Tensor]], beam_idx: torch.Tensor) ->
     GPT2_START_DOCSTRING,
 )
 class GPT2ForSequenceClassification(GPT2PreTrainedModel):
-    _keys_to_ignore_on_load_missing = [r"h\.\d+\.attn\.masked_bias", r"lm_head\.weight"]
+    _keys_to_ignore_on_load_missing = [r"h\.\d+\.attn\.masked_bias", r"lm_head.weight"]
 
     def __init__(self, config):
         super().__init__(config)

src/transformers/models/gpt_neo/modeling_gpt_neo.py

@@ -661,7 +661,7 @@ def custom_forward(*inputs):
 class GPTNeoForCausalLM(GPTNeoPreTrainedModel):
     _keys_to_ignore_on_load_missing = [
         r"h\.\d+\.attn\.masked_bias",
-        r"lm_head\.weight",
+        r"lm_head.weight",
         r"h\.\d+\.attn\.attention\.bias",
     ]
     _keys_to_ignore_on_save = [r"lm_head.weight"]
@@ -812,7 +812,7 @@ def _reorder_cache(past: Tuple[Tuple[torch.Tensor]], beam_idx: torch.Tensor) ->
     GPT_NEO_START_DOCSTRING,
 )
 class GPTNeoForSequenceClassification(GPTNeoPreTrainedModel):
-    _keys_to_ignore_on_load_missing = [r"h\.\d+\.attn\.masked_bias", r"lm_head\.weight"]
+    _keys_to_ignore_on_load_missing = [r"h\.\d+\.attn\.masked_bias", r"lm_head.weight"]
 
     def __init__(self, config):
         super().__init__(config)

src/transformers/models/gptj/modeling_gptj.py

@@ -890,7 +890,7 @@ def _reorder_cache(past: Tuple[Tuple[torch.Tensor]], beam_idx: torch.Tensor) ->
     GPTJ_START_DOCSTRING,
 )
 class GPTJForSequenceClassification(GPTJPreTrainedModel):
-    _keys_to_ignore_on_load_missing = [r"h\.\d+\.attn\.masked_bias", r"h\.\d+\.attn\.bias", r"lm_head\.weight"]
+    _keys_to_ignore_on_load_missing = [r"h\.\d+\.attn\.masked_bias", r"h\.\d+\.attn\.bias", r"lm_head.weight"]
 
     def __init__(self, config):
         super().__init__(config)
@@ -1017,7 +1017,7 @@ def forward(
     GPTJ_START_DOCSTRING,
 )
 class GPTJForQuestionAnswering(GPTJPreTrainedModel):
-    _keys_to_ignore_on_load_missing = [r"h\.\d+\.attn\.masked_bias", r"h\.\d+\.attn\.bias", r"lm_head\.weight"]
+    _keys_to_ignore_on_load_missing = [r"h\.\d+\.attn\.masked_bias", r"h\.\d+\.attn\.bias", r"lm_head.weight"]
 
     def __init__(self, config):
         super().__init__(config)

src/transformers/models/imagegpt/modeling_imagegpt.py

@@ -1087,7 +1087,7 @@ def _reorder_cache(past: Tuple[Tuple[torch.Tensor]], beam_idx: torch.Tensor) ->
     IMAGEGPT_START_DOCSTRING,
 )
 class ImageGPTForImageClassification(ImageGPTPreTrainedModel):
-    _keys_to_ignore_on_load_missing = [r"h\.\d+\.attn\.masked_bias", r"lm_head\.weight"]
+    _keys_to_ignore_on_load_missing = [r"h\.\d+\.attn\.masked_bias", r"lm_head.weight"]
 
     def __init__(self, config: ImageGPTConfig):
         super().__init__(config)

src/transformers/models/led/modeling_led.py

@@ -2298,9 +2298,9 @@ class LEDForConditionalGeneration(LEDPreTrainedModel):
     base_model_prefix = "led"
     _keys_to_ignore_on_load_missing = [
         r"final_logits_bias",
-        r"encoder\.version",
-        r"decoder\.version",
-        r"lm_head\.weight",
+        r"encoder.version",
+        r"decoder.version",
+        r"lm_head.weight",
     ]
 
     def __init__(self, config: LEDConfig):

src/transformers/models/longt5/modeling_longt5.py

@@ -1721,11 +1721,11 @@ def custom_forward(*inputs):
 )
 class LongT5Model(LongT5PreTrainedModel):
     _keys_to_ignore_on_load_missing = [
-        r"encoder\.embed_tokens\.weight",
-        r"decoder\.embed_tokens\.weight",
+        r"encoder.embed_tokens.weight",
+        r"decoder.embed_tokens.weight",
     ]
     _keys_to_ignore_on_load_unexpected = [
-        r"decoder\.block\.0\.layer\.1\.EncDecAttention\.relative_attention_bias\.weight",
+        r"decoder.block.0.layer.1.EncDecAttention.relative_attention_bias.weight",
     ]
 
     def __init__(self, config: LongT5Config):
@@ -1874,12 +1874,12 @@ def forward(
 @add_start_docstrings("""LONGT5 Model with a `language modeling` head on top.""", LONGT5_START_DOCSTRING)
 class LongT5ForConditionalGeneration(LongT5PreTrainedModel):
     _keys_to_ignore_on_load_missing = [
-        r"encoder\.embed_tokens\.weight",
-        r"decoder\.embed_tokens\.weight",
-        r"lm_head\.weight",
+        r"encoder.embed_tokens.weight",
+        r"decoder.embed_tokens.weight",
+        r"lm_head.weight",
     ]
     _keys_to_ignore_on_load_unexpected = [
-        r"decoder\.block\.0\.layer\.1\.EncDecAttention\.relative_attention_bias\.weight",
+        r"decoder.block.0.layer.1.EncDecAttention.relative_attention_bias.weight",
     ]
 
     def __init__(self, config: LongT5Config):
@@ -2116,7 +2116,7 @@ def _reorder_cache(self, past, beam_idx):
 )
 class LongT5EncoderModel(LongT5PreTrainedModel):
     authorized_missing_keys = [
-        r"encoder\.embed_tokens\.weight",
+        r"encoder.embed_tokens.weight",
     ]
 
     def __init__(self, config: LongT5Config):

src/transformers/models/m2m_100/modeling_m2m_100.py

@@ -1241,9 +1241,9 @@ def forward(
 class M2M100ForConditionalGeneration(M2M100PreTrainedModel):
     base_model_prefix = "model"
     _keys_to_ignore_on_load_missing = [
-        r"encoder\.version",
-        r"decoder\.version",
-        r"lm_head\.weight",
+        r"encoder.version",
+        r"decoder.version",
+        r"lm_head.weight",
         r"model.encoder.embed_positions.weights",
         r"model.decoder.embed_positions.weights",
     ]

src/transformers/models/marian/modeling_marian.py

@@ -1272,9 +1272,9 @@ class MarianMTModel(MarianPreTrainedModel):
     base_model_prefix = "model"
     _keys_to_ignore_on_load_missing = [
         r"final_logits_bias",
-        r"encoder\.version",
-        r"decoder\.version",
-        r"lm_head\.weight",
+        r"encoder.version",
+        r"decoder.version",
+        r"lm_head.weight",
         r"embed_positions",
     ]
 

src/transformers/models/mbart/modeling_mbart.py

@@ -1263,9 +1263,9 @@ class MBartForConditionalGeneration(MBartPreTrainedModel):
     base_model_prefix = "model"
     _keys_to_ignore_on_load_missing = [
         r"final_logits_bias",
-        r"encoder\.version",
-        r"decoder\.version",
-        r"lm_head\.weight",
+        r"encoder.version",
+        r"decoder.version",
+        r"lm_head.weight",
     ]
 
     def __init__(self, config: MBartConfig):

src/transformers/models/mt5/modeling_mt5.py

@@ -49,13 +49,13 @@ class MT5Model(T5Model):
     model_type = "mt5"
     config_class = MT5Config
     _keys_to_ignore_on_load_missing = [
-        r"encoder\.embed_tokens\.weight",
-        r"decoder\.embed_tokens\.weight",
-        r"decoder\.block\.0\.layer\.1\.EncDecAttention\.relative_attention_bias\.weight",
+        r"encoder.embed_tokens.weight",
+        r"decoder.embed_tokens.weight",
+        r"decoder.block.0.layer.1.EncDecAttention.relative_attention_bias.weight",
     ]
     _keys_to_ignore_on_save = [
-        r"encoder\.embed_tokens\.weight",
-        r"decoder\.embed_tokens\.weight",
+        r"encoder.embed_tokens.weight",
+        r"decoder.embed_tokens.weight",
     ]
 
 
@@ -84,10 +84,10 @@ class MT5ForConditionalGeneration(T5ForConditionalGeneration):
     model_type = "mt5"
     config_class = MT5Config
     _keys_to_ignore_on_load_missing = [
-        r"encoder\.embed_tokens\.weight",
+        r"encoder.embed_tokens.weight",
     ]
     _keys_to_ignore_on_save = [
-        r"encoder\.embed_tokens\.weight",
+        r"encoder.embed_tokens.weight",
     ]
 
 
@@ -112,8 +112,8 @@ class MT5EncoderModel(T5EncoderModel):
     model_type = "mt5"
     config_class = MT5Config
     _keys_to_ignore_on_load_missing = [
-        r"encoder\.embed_tokens\.weight",
+        r"encoder.embed_tokens.weight",
     ]
     _keys_to_ignore_on_save = [
-        r"encoder\.embed_tokens\.weight",
+        r"encoder.embed_tokens.weight",
     ]

src/transformers/models/opt/modeling_opt.py

@@ -382,7 +382,7 @@ class OPTPreTrainedModel(PreTrainedModel):
     base_model_prefix = "model"
     supports_gradient_checkpointing = True
     _no_split_modules = ["OPTDecoderLayer"]
-    _keys_to_ignore_on_load_unexpected = [r"decoder\.version"]
+    _keys_to_ignore_on_load_unexpected = [r"decoder.version"]
 
     def _init_weights(self, module):
         std = self.config.init_std
@@ -780,7 +780,7 @@ def forward(
 
 
 class OPTForCausalLM(OPTPreTrainedModel):
-    _keys_to_ignore_on_load_missing = [r"lm_head\.weight"]
+    _keys_to_ignore_on_load_missing = [r"lm_head.weight"]
 
     def __init__(self, config):
         super().__init__(config)

src/transformers/models/pegasus/modeling_pegasus.py

@@ -1290,10 +1290,10 @@ class PegasusForConditionalGeneration(PegasusPreTrainedModel):
     base_model_prefix = "model"
     _keys_to_ignore_on_load_missing = [
         r"final_logits_bias",
-        r"encoder\.version",
-        r"decoder\.version",
-        r"lm_head\.weight",
-        r"embed_positions\.weight",
+        r"encoder.version",
+        r"decoder.version",
+        r"lm_head.weight",
+        r"embed_positions.weight",
     ]
 
     def __init__(self, config: PegasusConfig):

src/transformers/models/plbart/modeling_plbart.py

@@ -1235,9 +1235,9 @@ class PLBartForConditionalGeneration(PLBartPreTrainedModel):
     base_model_prefix = "model"
     _keys_to_ignore_on_load_missing = [
         r"final_logits_bias",
-        r"encoder\.version",
-        r"decoder\.version",
-        r"lm_head\.weight",
+        r"encoder.version",
+        r"decoder.version",
+        r"lm_head.weight",
     ]
 
     def __init__(self, config: PLBartConfig):

src/transformers/models/roformer/modeling_roformer.py

@@ -699,8 +699,8 @@ class RoFormerPreTrainedModel(PreTrainedModel):
     supports_gradient_checkpointing = True
     _keys_to_ignore_on_load_missing = []
     _keys_to_ignore_on_load_unexpected = [
-        r"roformer\.embeddings_project\.weight",
-        r"roformer\.embeddings_project\.bias",
+        r"roformer.embeddings_project.weight",
+        r"roformer.embeddings_project.bias",
     ]
 
     def _init_weights(self, module):

src/transformers/models/speech_to_text/modeling_speech_to_text.py

@@ -1252,8 +1252,8 @@ def forward(
 class Speech2TextForConditionalGeneration(Speech2TextPreTrainedModel):
     base_model_prefix = "model"
     _keys_to_ignore_on_load_missing = [
-        r"encoder\.version",
-        r"decoder\.version",
+        r"encoder.version",
+        r"decoder.version",
         r"model.encoder.embed_positions.weights",
         r"model.decoder.embed_positions.weights",
     ]

src/transformers/models/t5/modeling_t5.py

@@ -1266,11 +1266,11 @@ def custom_forward(*inputs):
 )
 class T5Model(T5PreTrainedModel):
     _keys_to_ignore_on_load_missing = [
-        r"encoder\.embed_tokens\.weight",
-        r"decoder\.embed_tokens\.weight",
+        r"encoder.embed_tokens.weight",
+        r"decoder.embed_tokens.weight",
     ]
     _keys_to_ignore_on_load_unexpected = [
-        r"decoder\.block\.0\.layer\.1\.EncDecAttention\.relative_attention_bias\.weight",
+        r"decoder.block.0.layer.1.EncDecAttention.relative_attention_bias.weight",
     ]
 
     def __init__(self, config: T5Config):
@@ -1455,12 +1455,12 @@ def forward(
 @add_start_docstrings("""T5 Model with a `language modeling` head on top.""", T5_START_DOCSTRING)
 class T5ForConditionalGeneration(T5PreTrainedModel):
     _keys_to_ignore_on_load_missing = [
-        r"encoder\.embed_tokens\.weight",
-        r"decoder\.embed_tokens\.weight",
-        r"lm_head\.weight",
+        r"encoder.embed_tokens.weight",
+        r"decoder.embed_tokens.weight",
+        r"lm_head.weight",
     ]
     _keys_to_ignore_on_load_unexpected = [
-        r"decoder\.block\.0\.layer\.1\.EncDecAttention\.relative_attention_bias\.weight",
+        r"decoder.block.0.layer.1.EncDecAttention.relative_attention_bias.weight",
     ]
 
     def __init__(self, config: T5Config):
@@ -1749,7 +1749,7 @@ def _reorder_cache(self, past, beam_idx):
 )
 class T5EncoderModel(T5PreTrainedModel):
     authorized_missing_keys = [
-        r"encoder\.embed_tokens\.weight",
+        r"encoder.embed_tokens.weight",
     ]
 
     def __init__(self, config: T5Config):

src/transformers/models/transfo_xl/modeling_transfo_xl.py

@@ -1198,7 +1198,7 @@ def _reorder_cache(mems: List[torch.Tensor], beam_idx: torch.Tensor) -> List[tor
     TRANSFO_XL_START_DOCSTRING,
 )
 class TransfoXLForSequenceClassification(TransfoXLPreTrainedModel):
-    _keys_to_ignore_on_load_missing = [r"h\.\d+\.attn\.masked_bias", r"lm_head\.weight"]
+    _keys_to_ignore_on_load_missing = [r"h\.\d+\.attn\.masked_bias", r"lm_head.weight"]
 
     def __init__(self, config):
         super().__init__(config)