[Wav2Vec2] Rename model's feature extractor to feature encoder (huggi…

…ngface#14959)

* rename classes

* clean up more namings

* remove bogus file

* Apply suggestions from code review

* Apply suggestions from code review

* replace more names

* more regex replace

* make style

* correct

* correct more

* make style

* finish

* correct more in wav2vec2

* make style

* improve freeze_extractor

* add aliases

* add tf aliases

examples/pytorch/audio-classification/run_audio_classification.py

@@ -17,6 +17,7 @@
 import logging
 import os
 import sys
+import warnings
 from dataclasses import dataclass, field
 from random import randint
 from typing import Optional
@@ -76,24 +77,24 @@ class DataTrainingArguments:
     eval_file: Optional[str] = field(
         default=None, metadata={"help": "A file containing the validation audio paths and labels."}
     )
-    train_split_name: Optional[str] = field(
+    train_split_name: str = field(
         default="train",
         metadata={
             "help": "The name of the training data set split to use (via the datasets library). Defaults to 'train'"
         },
     )
-    eval_split_name: Optional[str] = field(
+    eval_split_name: str = field(
         default="validation",
         metadata={
             "help": "The name of the training data set split to use (via the datasets library). Defaults to "
             "'validation'"
         },
     )
-    audio_column_name: Optional[str] = field(
+    audio_column_name: str = field(
         default="audio",
         metadata={"help": "The name of the dataset column containing the audio data. Defaults to 'audio'"},
     )
-    label_column_name: Optional[str] = field(
+    label_column_name: str = field(
         default="label", metadata={"help": "The name of the dataset column containing the labels. Defaults to 'label'"}
     )
     max_train_samples: Optional[int] = field(
@@ -110,7 +111,7 @@ class DataTrainingArguments:
             "value if set."
         },
     )
-    max_length_seconds: Optional[float] = field(
+    max_length_seconds: float = field(
         default=20,
         metadata={"help": "Audio clips will be randomly cut to this length during training if the value is set."},
     )
@@ -136,11 +137,13 @@ class ModelArguments:
         default="main",
         metadata={"help": "The specific model version to use (can be a branch name, tag name or commit id)."},
     )
-    feature_extractor_name: str = field(default=None, metadata={"help": "Name or path of preprocessor config."})
-    freeze_feature_extractor: Optional[bool] = field(
-        default=True, metadata={"help": "Whether to freeze the feature extractor layers of the model."}
+    feature_extractor_name: Optional[str] = field(
+        default=None, metadata={"help": "Name or path of preprocessor config."}
+    )
+    freeze_feature_encoder: bool = field(
+        default=True, metadata={"help": "Whether to freeze the feature encoder layers of the model."}
     )
-    attention_mask: Optional[bool] = field(
+    attention_mask: bool = field(
         default=True, metadata={"help": "Whether to generate an attention mask in the feature extractor."}
     )
     use_auth_token: bool = field(
@@ -150,6 +153,24 @@ class ModelArguments:
             "with private models)."
         },
     )
+    freeze_feature_extractor: Optional[bool] = field(
+        default=None, metadata={"help": "Whether to freeze the feature extractor layers of the model."}
+    )
+
+    def __post_init__(self):
+        if not self.freeze_feature_extractor and self.freeze_feature_encoder:
+            warnings.warn(
+                "The argument `--freeze_feature_extractor` is deprecated and "
+                "will be removed in a future version. Use `--freeze_feature_encoder`"
+                "instead. Setting `freeze_feature_encoder==True`.",
+                FutureWarning,
+            )
+        if self.freeze_feature_extractor and not self.freeze_feature_encoder:
+            raise ValueError(
+                "The argument `--freeze_feature_extractor` is deprecated and "
+                "should not be used in combination with `--freeze_feature_encoder`."
+                "Only make use of `--freeze_feature_encoder`."
+            )
 
 
 def main():
@@ -302,8 +323,8 @@ def compute_metrics(eval_pred):
     )
 
     # freeze the convolutional waveform encoder
-    if model_args.freeze_feature_extractor:
-        model.freeze_feature_extractor()
+    if model_args.freeze_feature_encoder:
+        model.freeze_feature_encoder()
 
     if training_args.do_train:
         if data_args.max_train_samples is not None:

examples/pytorch/speech-recognition/README.md

@@ -78,7 +78,7 @@ python run_speech_recognition_ctc.py \
 	--eval_steps="100" \
 	--layerdrop="0.0" \
 	--save_total_limit="3" \
-	--freeze_feature_extractor \
+	--freeze_feature_encoder \
 	--gradient_checkpointing \
 	--chars_to_ignore , ? . ! - \; \: \" “ % ‘ ” � \
 	--fp16 \
@@ -113,7 +113,7 @@ python -m torch.distributed.launch \
 	--logging_steps="1" \
 	--layerdrop="0.0" \
 	--save_total_limit="3" \
-	--freeze_feature_extractor \
+	--freeze_feature_encoder \
 	--gradient_checkpointing \
 	--chars_to_ignore , ? . ! - \; \: \" “ % ‘ ” � \
 	--fp16 \
@@ -304,7 +304,7 @@ python run_speech_recognition_seq2seq.py \
 	--eval_steps="400" \
 	--logging_steps="10" \
 	--save_total_limit="1" \
-	--freeze_feature_extractor \
+	--freeze_feature_encoder \
 	--gradient_checkpointing \
 	--fp16 \
 	--group_by_length \
@@ -346,7 +346,7 @@ python -m torch.distributed.launch \
 	--eval_steps="400" \
 	--logging_steps="10" \
 	--save_total_limit="1" \
-	--freeze_feature_extractor \
+	--freeze_feature_encoder \
 	--gradient_checkpointing \
 	--fp16 \
 	--group_by_length \

examples/pytorch/speech-recognition/run_speech_recognition_ctc.py

@@ -78,52 +78,50 @@ class ModelArguments:
         default=None,
         metadata={"help": "Where do you want to store the pretrained models downloaded from huggingface.co"},
     )
-    freeze_feature_extractor: Optional[bool] = field(
-        default=True, metadata={"help": "Whether to freeze the feature extractor layers of the model."}
+    freeze_feature_encoder: bool = field(
+        default=True, metadata={"help": "Whether to freeze the feature encoder layers of the model."}
     )
-    attention_dropout: Optional[float] = field(
+    attention_dropout: float = field(
         default=0.0, metadata={"help": "The dropout ratio for the attention probabilities."}
     )
-    activation_dropout: Optional[float] = field(
+    activation_dropout: float = field(
         default=0.0, metadata={"help": "The dropout ratio for activations inside the fully connected layer."}
     )
-    feat_proj_dropout: Optional[float] = field(
-        default=0.0, metadata={"help": "The dropout ratio for the projected features."}
-    )
-    hidden_dropout: Optional[float] = field(
+    feat_proj_dropout: float = field(default=0.0, metadata={"help": "The dropout ratio for the projected features."})
+    hidden_dropout: float = field(
         default=0.0,
         metadata={
             "help": "The dropout probability for all fully connected layers in the embeddings, encoder, and pooler."
         },
     )
-    final_dropout: Optional[float] = field(
+    final_dropout: float = field(
         default=0.0,
         metadata={"help": "The dropout probability for the final projection layer."},
     )
-    mask_time_prob: Optional[float] = field(
+    mask_time_prob: float = field(
         default=0.05,
         metadata={
             "help": "Probability of each feature vector along the time axis to be chosen as the start of the vector"
             "span to be masked. Approximately ``mask_time_prob * sequence_length // mask_time_length`` feature"
             "vectors will be masked along the time axis."
         },
     )
-    mask_time_length: Optional[int] = field(
+    mask_time_length: int = field(
         default=10,
         metadata={"help": "Length of vector span to mask along the time axis."},
     )
-    mask_feature_prob: Optional[float] = field(
+    mask_feature_prob: float = field(
         default=0.0,
         metadata={
             "help": "Probability of each feature vector along the feature axis to be chosen as the start of the vector"
             "span to be masked. Approximately ``mask_feature_prob * sequence_length // mask_feature_length`` feature bins will be masked along the time axis."
         },
     )
-    mask_feature_length: Optional[int] = field(
+    mask_feature_length: int = field(
         default=10,
         metadata={"help": "Length of vector span to mask along the feature axis."},
     )
-    layerdrop: Optional[float] = field(default=0.0, metadata={"help": "The LayerDrop probability."})
+    layerdrop: float = field(default=0.0, metadata={"help": "The LayerDrop probability."})
     ctc_loss_reduction: Optional[str] = field(
         default="mean", metadata={"help": "The way the ctc loss should be reduced. Should be one of 'mean' or 'sum'."}
     )
@@ -142,26 +140,26 @@ class DataTrainingArguments:
     dataset_name: str = field(
         metadata={"help": "The configuration name of the dataset to use (via the datasets library)."}
     )
-    dataset_config_name: Optional[str] = field(
+    dataset_config_name: str = field(
         default=None, metadata={"help": "The configuration name of the dataset to use (via the datasets library)."}
     )
-    train_split_name: Optional[str] = field(
+    train_split_name: str = field(
         default="train+validation",
         metadata={
             "help": "The name of the training data set split to use (via the datasets library). Defaults to 'train'"
         },
     )
-    eval_split_name: Optional[str] = field(
+    eval_split_name: str = field(
         default="test",
         metadata={
             "help": "The name of the training data set split to use (via the datasets library). Defaults to 'train'"
         },
     )
-    audio_column_name: Optional[str] = field(
+    audio_column_name: str = field(
         default="audio",
         metadata={"help": "The name of the dataset column containing the audio data. Defaults to 'audio'"},
     )
-    text_column_name: Optional[str] = field(
+    text_column_name: str = field(
         default="text",
         metadata={"help": "The name of the dataset column containing the text data. Defaults to 'text'"},
     )
@@ -190,20 +188,20 @@ class DataTrainingArguments:
         default=None,
         metadata={"help": "A list of characters to remove from the transcripts."},
     )
-    eval_metrics: Optional[List[str]] = list_field(
+    eval_metrics: List[str] = list_field(
         default=["wer"],
         metadata={"help": "A list of metrics the model should be evaluated on. E.g. `'wer cer'`"},
     )
-    max_duration_in_seconds: Optional[float] = field(
+    max_duration_in_seconds: float = field(
         default=20.0,
         metadata={
             "help": "Filter audio files that are longer than `max_duration_in_seconds` seconds to 'max_duration_in_seconds`"
         },
     )
-    min_duration_in_seconds: Optional[float] = field(
+    min_duration_in_seconds: float = field(
         default=0.0, metadata={"help": "Filter audio files that are shorter than `min_duration_in_seconds` seconds"}
     )
-    preprocessing_only: Optional[bool] = field(
+    preprocessing_only: bool = field(
         default=False,
         metadata={
             "help": "Whether to only do data preprocessing and skip training. "
@@ -212,22 +210,22 @@ class DataTrainingArguments:
             "so that the cached datasets can consequently be loaded in distributed training"
         },
     )
-    use_auth_token: Optional[bool] = field(
+    use_auth_token: bool = field(
         default=False,
         metadata={
             "help": "If :obj:`True`, will use the token generated when running"
             ":obj:`transformers-cli login` as HTTP bearer authorization for remote files."
         },
     )
-    unk_token: Optional[str] = field(
+    unk_token: str = field(
         default="[UNK]",
         metadata={"help": "The unk token for the tokenizer"},
     )
-    pad_token: Optional[str] = field(
+    pad_token: str = field(
         default="[PAD]",
         metadata={"help": "The padding token for the tokenizer"},
     )
-    word_delimiter_token: Optional[str] = field(
+    word_delimiter_token: str = field(
         default="|",
         metadata={"help": "The word delimiter token for the tokenizer"},
     )
@@ -545,8 +543,8 @@ def remove_special_characters(batch):
     )
 
     # freeze encoder
-    if model_args.freeze_feature_extractor:
-        model.freeze_feature_extractor()
+    if model_args.freeze_feature_encoder:
+        model.freeze_feature_encoder()
 
     # 6. Now we preprocess the datasets including loading the audio, resampling and normalization
     # Thankfully, `datasets` takes care of automatically loading and resampling the audio,

examples/pytorch/speech-recognition/run_speech_recognition_seq2seq.py

@@ -91,8 +91,8 @@ class ModelArguments:
             "with private models)."
         },
     )
-    freeze_feature_extractor: Optional[bool] = field(
-        default=True, metadata={"help": "Whether to freeze the feature extractor layers of the model."}
+    freeze_feature_encoder: bool = field(
+        default=True, metadata={"help": "Whether to freeze the feature encoder layers of the model."}
     )
 
 
@@ -102,7 +102,7 @@ class DataTrainingArguments:
     Arguments pertaining to what data we are going to input our model for training and eval.
     """
 
-    dataset_name: Optional[str] = field(
+    dataset_name: str = field(
         default=None, metadata={"help": "The name of the dataset to use (via the datasets library)."}
     )
     dataset_config_name: Optional[str] = field(
@@ -133,24 +133,24 @@ class DataTrainingArguments:
             "value if set."
         },
     )
-    audio_column_name: Optional[str] = field(
+    audio_column_name: str = field(
         default="audio",
         metadata={"help": "The name of the dataset column containing the audio data. Defaults to 'audio'"},
     )
-    text_column_name: Optional[str] = field(
+    text_column_name: str = field(
         default="text",
         metadata={"help": "The name of the dataset column containing the text data. Defaults to 'text'"},
     )
-    max_duration_in_seconds: Optional[float] = field(
+    max_duration_in_seconds: float = field(
         default=20.0,
         metadata={
             "help": "Truncate audio files that are longer than `max_duration_in_seconds` seconds to 'max_duration_in_seconds`"
         },
     )
-    min_duration_in_seconds: Optional[float] = field(
+    min_duration_in_seconds: float = field(
         default=0.0, metadata={"help": "Filter audio files that are shorter than `min_duration_in_seconds` seconds"}
     )
-    preprocessing_only: Optional[bool] = field(
+    preprocessing_only: bool = field(
         default=False,
         metadata={
             "help": "Whether to only do data preprocessing and skip training. "
@@ -159,19 +159,19 @@ class DataTrainingArguments:
             "so that the cached datasets can consequently be loaded in distributed training"
         },
     )
-    train_split_name: Optional[str] = field(
+    train_split_name: str = field(
         default="train",
         metadata={
             "help": "The name of the training data set split to use (via the datasets library). Defaults to 'train'"
         },
     )
-    eval_split_name: Optional[str] = field(
+    eval_split_name: str = field(
         default="test",
         metadata={
             "help": "The name of the training data set split to use (via the datasets library). Defaults to 'train'"
         },
     )
-    do_lower_case: Optional[bool] = field(
+    do_lower_case: bool = field(
         default=True,
         metadata={"help": "Whether the target text should be lower cased."},
     )
@@ -335,8 +335,8 @@ def main():
     if model.config.decoder_start_token_id is None:
         raise ValueError("Make sure that `config.decoder_start_token_id` is correctly defined")
 
-    if model_args.freeze_feature_extractor:
-        model.freeze_feature_extractor()
+    if model_args.freeze_feature_encoder:
+        model.freeze_feature_encoder()
 
     # 6. Resample speech dataset if necassary
     dataset_sampling_rate = next(iter(raw_datasets.values())).features[data_args.audio_column_name].sampling_rate