patrickvonplaten commited on
Commit
6ebcbaa
1 Parent(s): fcede83

Training in progress, step 500

Browse files
.gitignore ADDED
@@ -0,0 +1 @@
 
 
1
+ checkpoint-*/
added_tokens.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"<s>": 35, "</s>": 36}
config.json ADDED
@@ -0,0 +1,107 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "facebook/wav2vec2-xls-r-300m",
3
+ "activation_dropout": 0.1,
4
+ "adapter_kernel_size": 3,
5
+ "adapter_stride": 2,
6
+ "add_adapter": false,
7
+ "apply_spec_augment": true,
8
+ "architectures": [
9
+ "Wav2Vec2ForCTC"
10
+ ],
11
+ "attention_dropout": 0.0,
12
+ "bos_token_id": 1,
13
+ "classifier_proj_size": 256,
14
+ "codevector_dim": 768,
15
+ "contrastive_logits_temperature": 0.1,
16
+ "conv_bias": true,
17
+ "conv_dim": [
18
+ 512,
19
+ 512,
20
+ 512,
21
+ 512,
22
+ 512,
23
+ 512,
24
+ 512
25
+ ],
26
+ "conv_kernel": [
27
+ 10,
28
+ 3,
29
+ 3,
30
+ 3,
31
+ 3,
32
+ 2,
33
+ 2
34
+ ],
35
+ "conv_stride": [
36
+ 5,
37
+ 2,
38
+ 2,
39
+ 2,
40
+ 2,
41
+ 2,
42
+ 2
43
+ ],
44
+ "ctc_loss_reduction": "mean",
45
+ "ctc_zero_infinity": false,
46
+ "diversity_loss_weight": 0.1,
47
+ "do_stable_layer_norm": true,
48
+ "eos_token_id": 2,
49
+ "feat_extract_activation": "gelu",
50
+ "feat_extract_dropout": 0.0,
51
+ "feat_extract_norm": "layer",
52
+ "feat_proj_dropout": 0.0,
53
+ "feat_quantizer_dropout": 0.0,
54
+ "final_dropout": 0.0,
55
+ "hidden_act": "gelu",
56
+ "hidden_dropout": 0.0,
57
+ "hidden_size": 1024,
58
+ "initializer_range": 0.02,
59
+ "intermediate_size": 4096,
60
+ "layer_norm_eps": 1e-05,
61
+ "layerdrop": 0.0,
62
+ "mask_feature_length": 64,
63
+ "mask_feature_min_masks": 0,
64
+ "mask_feature_prob": 0.25,
65
+ "mask_time_length": 10,
66
+ "mask_time_min_masks": 2,
67
+ "mask_time_prob": 0.75,
68
+ "model_type": "wav2vec2",
69
+ "num_adapter_layers": 3,
70
+ "num_attention_heads": 16,
71
+ "num_codevector_groups": 2,
72
+ "num_codevectors_per_group": 320,
73
+ "num_conv_pos_embedding_groups": 16,
74
+ "num_conv_pos_embeddings": 128,
75
+ "num_feat_extract_layers": 7,
76
+ "num_hidden_layers": 24,
77
+ "num_negatives": 100,
78
+ "output_hidden_size": 1024,
79
+ "pad_token_id": 34,
80
+ "proj_codevector_dim": 768,
81
+ "tdnn_dilation": [
82
+ 1,
83
+ 2,
84
+ 3,
85
+ 1,
86
+ 1
87
+ ],
88
+ "tdnn_dim": [
89
+ 512,
90
+ 512,
91
+ 512,
92
+ 512,
93
+ 1500
94
+ ],
95
+ "tdnn_kernel": [
96
+ 5,
97
+ 3,
98
+ 3,
99
+ 1,
100
+ 1
101
+ ],
102
+ "torch_dtype": "float32",
103
+ "transformers_version": "4.17.0.dev0",
104
+ "use_weighted_layer_sum": false,
105
+ "vocab_size": 37,
106
+ "xvector_output_dim": 512
107
+ }
preprocessor_config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "do_normalize": true,
3
+ "feature_extractor_type": "Wav2Vec2FeatureExtractor",
4
+ "feature_size": 1,
5
+ "padding_side": "right",
6
+ "padding_value": 0,
7
+ "return_attention_mask": true,
8
+ "sampling_rate": 16000
9
+ }
pytorch_model.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c5eeb7897ad6cbe8e3e085ae4a0f42c6a084811bce3a504e25737455e164653a
3
+ size 1262075377
run.sh ADDED
@@ -0,0 +1,34 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ python run_speech_recognition_ctc.py \
2
+ --dataset_name="mozilla-foundation/common_voice_8_0" \
3
+ --model_name_or_path="facebook/wav2vec2-xls-r-300m" \
4
+ --dataset_config_name="sv-SE" \
5
+ --output_dir="./" \
6
+ --overwrite_output_dir \
7
+ --num_train_epochs="50" \
8
+ --per_device_train_batch_size="8" \
9
+ --per_device_eval_batch_size="8" \
10
+ --gradient_accumulation_steps="4" \
11
+ --learning_rate="7.5e-5" \
12
+ --warmup_steps="2000" \
13
+ --length_column_name="input_length" \
14
+ --evaluation_strategy="steps" \
15
+ --text_column_name="sentence" \
16
+ --save_steps="500" \
17
+ --eval_steps="500" \
18
+ --logging_steps="100" \
19
+ --layerdrop="0.0" \
20
+ --activation_dropout="0.1" \
21
+ --save_total_limit="3" \
22
+ --freeze_feature_encoder \
23
+ --feat_proj_dropout="0.0" \
24
+ --mask_time_prob="0.75" \
25
+ --mask_time_length="10" \
26
+ --mask_feature_prob="0.25" \
27
+ --mask_feature_length="64" \
28
+ --chars_to_ignore , ? . ! \- \; \: \" “ % ‘ ” � — ’ … – \
29
+ --gradient_checkpointing \
30
+ --use_auth_token \
31
+ --fp16 \
32
+ --group_by_length \
33
+ --do_train --do_eval \
34
+ --push_to_hub
run_speech_recognition_ctc.py ADDED
@@ -0,0 +1,731 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/usr/bin/env python
2
+ # coding=utf-8
3
+ # Copyright 2021 The HuggingFace Inc. team. All rights reserved.
4
+ #
5
+ # Licensed under the Apache License, Version 2.0 (the "License");
6
+ # you may not use this file except in compliance with the License.
7
+ # You may obtain a copy of the License at
8
+ #
9
+ # http://www.apache.org/licenses/LICENSE-2.0
10
+ #
11
+ # Unless required by applicable law or agreed to in writing, software
12
+ # distributed under the License is distributed on an "AS IS" BASIS,
13
+ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14
+ # See the License for the specific language governing permissions and
15
+
16
+ """ Fine-tuning a 🤗 Transformers CTC model for automatic speech recognition"""
17
+
18
+ import functools
19
+ import json
20
+ import logging
21
+ import os
22
+ import re
23
+ import sys
24
+ import warnings
25
+ from dataclasses import dataclass, field
26
+ from typing import Dict, List, Optional, Union
27
+
28
+ import datasets
29
+ import numpy as np
30
+ import torch
31
+ from datasets import DatasetDict, load_dataset, load_metric
32
+
33
+ import transformers
34
+ from transformers import (
35
+ AutoConfig,
36
+ AutoFeatureExtractor,
37
+ AutoModelForCTC,
38
+ AutoProcessor,
39
+ AutoTokenizer,
40
+ HfArgumentParser,
41
+ Trainer,
42
+ TrainingArguments,
43
+ Wav2Vec2Processor,
44
+ set_seed,
45
+ )
46
+ from transformers.trainer_utils import get_last_checkpoint, is_main_process
47
+ from transformers.utils import check_min_version
48
+ from transformers.utils.versions import require_version
49
+
50
+
51
+ # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
52
+ check_min_version("4.16.0.dev0")
53
+
54
+ require_version("datasets>=1.13.3", "To fix: pip install -r examples/pytorch/text-classification/requirements.txt")
55
+
56
+
57
+ logger = logging.getLogger(__name__)
58
+
59
+
60
+ def list_field(default=None, metadata=None):
61
+ return field(default_factory=lambda: default, metadata=metadata)
62
+
63
+
64
+ @dataclass
65
+ class ModelArguments:
66
+ """
67
+ Arguments pertaining to which model/config/tokenizer we are going to fine-tune from.
68
+ """
69
+
70
+ model_name_or_path: str = field(
71
+ metadata={"help": "Path to pretrained model or model identifier from huggingface.co/models"}
72
+ )
73
+ tokenizer_name_or_path: Optional[str] = field(
74
+ default=None,
75
+ metadata={"help": "Path to pretrained tokenizer or tokenizer identifier from huggingface.co/models"},
76
+ )
77
+ cache_dir: Optional[str] = field(
78
+ default=None,
79
+ metadata={"help": "Where do you want to store the pretrained models downloaded from huggingface.co"},
80
+ )
81
+ freeze_feature_encoder: bool = field(
82
+ default=True, metadata={"help": "Whether to freeze the feature encoder layers of the model."}
83
+ )
84
+ attention_dropout: float = field(
85
+ default=0.0, metadata={"help": "The dropout ratio for the attention probabilities."}
86
+ )
87
+ activation_dropout: float = field(
88
+ default=0.0, metadata={"help": "The dropout ratio for activations inside the fully connected layer."}
89
+ )
90
+ feat_proj_dropout: float = field(default=0.0, metadata={"help": "The dropout ratio for the projected features."})
91
+ hidden_dropout: float = field(
92
+ default=0.0,
93
+ metadata={
94
+ "help": "The dropout probability for all fully connected layers in the embeddings, encoder, and pooler."
95
+ },
96
+ )
97
+ final_dropout: float = field(
98
+ default=0.0,
99
+ metadata={"help": "The dropout probability for the final projection layer."},
100
+ )
101
+ mask_time_prob: float = field(
102
+ default=0.05,
103
+ metadata={
104
+ "help": "Probability of each feature vector along the time axis to be chosen as the start of the vector"
105
+ "span to be masked. Approximately ``mask_time_prob * sequence_length // mask_time_length`` feature"
106
+ "vectors will be masked along the time axis."
107
+ },
108
+ )
109
+ mask_time_length: int = field(
110
+ default=10,
111
+ metadata={"help": "Length of vector span to mask along the time axis."},
112
+ )
113
+ mask_feature_prob: float = field(
114
+ default=0.0,
115
+ metadata={
116
+ "help": "Probability of each feature vector along the feature axis to be chosen as the start of the vector"
117
+ "span to be masked. Approximately ``mask_feature_prob * sequence_length // mask_feature_length`` feature bins will be masked along the time axis."
118
+ },
119
+ )
120
+ mask_feature_length: int = field(
121
+ default=10,
122
+ metadata={"help": "Length of vector span to mask along the feature axis."},
123
+ )
124
+ layerdrop: float = field(default=0.0, metadata={"help": "The LayerDrop probability."})
125
+ ctc_loss_reduction: Optional[str] = field(
126
+ default="mean", metadata={"help": "The way the ctc loss should be reduced. Should be one of 'mean' or 'sum'."}
127
+ )
128
+
129
+
130
+ @dataclass
131
+ class DataTrainingArguments:
132
+ """
133
+ Arguments pertaining to what data we are going to input our model for training and eval.
134
+
135
+ Using `HfArgumentParser` we can turn this class
136
+ into argparse arguments to be able to specify them on
137
+ the command line.
138
+ """
139
+
140
+ dataset_name: str = field(
141
+ metadata={"help": "The configuration name of the dataset to use (via the datasets library)."}
142
+ )
143
+ dataset_config_name: str = field(
144
+ default=None, metadata={"help": "The configuration name of the dataset to use (via the datasets library)."}
145
+ )
146
+ train_split_name: str = field(
147
+ default="train+validation",
148
+ metadata={
149
+ "help": "The name of the training data set split to use (via the datasets library). Defaults to 'train'"
150
+ },
151
+ )
152
+ eval_split_name: str = field(
153
+ default="test",
154
+ metadata={
155
+ "help": "The name of the training data set split to use (via the datasets library). Defaults to 'train'"
156
+ },
157
+ )
158
+ audio_column_name: str = field(
159
+ default="audio",
160
+ metadata={"help": "The name of the dataset column containing the audio data. Defaults to 'audio'"},
161
+ )
162
+ text_column_name: str = field(
163
+ default="text",
164
+ metadata={"help": "The name of the dataset column containing the text data. Defaults to 'text'"},
165
+ )
166
+ overwrite_cache: bool = field(
167
+ default=False, metadata={"help": "Overwrite the cached preprocessed datasets or not."}
168
+ )
169
+ preprocessing_num_workers: Optional[int] = field(
170
+ default=None,
171
+ metadata={"help": "The number of processes to use for the preprocessing."},
172
+ )
173
+ max_train_samples: Optional[int] = field(
174
+ default=None,
175
+ metadata={
176
+ "help": "For debugging purposes or quicker training, truncate the number of training examples to this "
177
+ "value if set."
178
+ },
179
+ )
180
+ max_eval_samples: Optional[int] = field(
181
+ default=None,
182
+ metadata={
183
+ "help": "For debugging purposes or quicker training, truncate the number of validation examples to this "
184
+ "value if set."
185
+ },
186
+ )
187
+ chars_to_ignore: Optional[List[str]] = list_field(
188
+ default=None,
189
+ metadata={"help": "A list of characters to remove from the transcripts."},
190
+ )
191
+ eval_metrics: List[str] = list_field(
192
+ default=["wer"],
193
+ metadata={"help": "A list of metrics the model should be evaluated on. E.g. `'wer cer'`"},
194
+ )
195
+ max_duration_in_seconds: float = field(
196
+ default=20.0,
197
+ metadata={
198
+ "help": "Filter audio files that are longer than `max_duration_in_seconds` seconds to 'max_duration_in_seconds`"
199
+ },
200
+ )
201
+ min_duration_in_seconds: float = field(
202
+ default=0.0, metadata={"help": "Filter audio files that are shorter than `min_duration_in_seconds` seconds"}
203
+ )
204
+ preprocessing_only: bool = field(
205
+ default=False,
206
+ metadata={
207
+ "help": "Whether to only do data preprocessing and skip training. "
208
+ "This is especially useful when data preprocessing errors out in distributed training due to timeout. "
209
+ "In this case, one should run the preprocessing in a non-distributed setup with `preprocessing_only=True` "
210
+ "so that the cached datasets can consequently be loaded in distributed training"
211
+ },
212
+ )
213
+ use_auth_token: bool = field(
214
+ default=False,
215
+ metadata={
216
+ "help": "If :obj:`True`, will use the token generated when running"
217
+ ":obj:`transformers-cli login` as HTTP bearer authorization for remote files."
218
+ },
219
+ )
220
+ unk_token: str = field(
221
+ default="[UNK]",
222
+ metadata={"help": "The unk token for the tokenizer"},
223
+ )
224
+ pad_token: str = field(
225
+ default="[PAD]",
226
+ metadata={"help": "The padding token for the tokenizer"},
227
+ )
228
+ word_delimiter_token: str = field(
229
+ default="|",
230
+ metadata={"help": "The word delimiter token for the tokenizer"},
231
+ )
232
+ phoneme_language: Optional[str] = field(
233
+ default=None,
234
+ metadata={
235
+ "help": "The target language that should be used be"
236
+ " passed to the tokenizer for tokenization. Note that"
237
+ " this is only relevant if the model classifies the"
238
+ " input audio to a sequence of phoneme sequences."
239
+ },
240
+ )
241
+
242
+
243
+ @dataclass
244
+ class DataCollatorCTCWithPadding:
245
+ """
246
+ Data collator that will dynamically pad the inputs received.
247
+ Args:
248
+ processor (:class:`~transformers.AutoProcessor`)
249
+ The processor used for proccessing the data.
250
+ padding (:obj:`bool`, :obj:`str` or :class:`~transformers.tokenization_utils_base.PaddingStrategy`, `optional`, defaults to :obj:`True`):
251
+ Select a strategy to pad the returned sequences (according to the model's padding side and padding index)
252
+ among:
253
+ * :obj:`True` or :obj:`'longest'`: Pad to the longest sequence in the batch (or no padding if only a single
254
+ sequence if provided).
255
+ * :obj:`'max_length'`: Pad to a maximum length specified with the argument :obj:`max_length` or to the
256
+ maximum acceptable input length for the model if that argument is not provided.
257
+ * :obj:`False` or :obj:`'do_not_pad'` (default): No padding (i.e., can output a batch with sequences of
258
+ different lengths).
259
+ max_length (:obj:`int`, `optional`):
260
+ Maximum length of the ``input_values`` of the returned list and optionally padding length (see above).
261
+ max_length_labels (:obj:`int`, `optional`):
262
+ Maximum length of the ``labels`` returned list and optionally padding length (see above).
263
+ pad_to_multiple_of (:obj:`int`, `optional`):
264
+ If set will pad the sequence to a multiple of the provided value.
265
+ This is especially useful to enable the use of Tensor Cores on NVIDIA hardware with compute capability >=
266
+ 7.5 (Volta).
267
+ """
268
+
269
+ processor: AutoProcessor
270
+ padding: Union[bool, str] = "longest"
271
+ pad_to_multiple_of: Optional[int] = None
272
+ pad_to_multiple_of_labels: Optional[int] = None
273
+
274
+ def __call__(self, features: List[Dict[str, Union[List[int], torch.Tensor]]]) -> Dict[str, torch.Tensor]:
275
+ # split inputs and labels since they have to be of different lenghts and need
276
+ # different padding methods
277
+ input_features = [{"input_values": feature["input_values"]} for feature in features]
278
+ label_features = [{"input_ids": feature["labels"]} for feature in features]
279
+
280
+ batch = self.processor.pad(
281
+ input_features,
282
+ padding=self.padding,
283
+ pad_to_multiple_of=self.pad_to_multiple_of,
284
+ return_tensors="pt",
285
+ )
286
+
287
+ with self.processor.as_target_processor():
288
+ labels_batch = self.processor.pad(
289
+ label_features,
290
+ padding=self.padding,
291
+ pad_to_multiple_of=self.pad_to_multiple_of_labels,
292
+ return_tensors="pt",
293
+ )
294
+
295
+ # replace padding with -100 to ignore loss correctly
296
+ labels = labels_batch["input_ids"].masked_fill(labels_batch.attention_mask.ne(1), -100)
297
+
298
+ batch["labels"] = labels
299
+
300
+ return batch
301
+
302
+
303
+ def create_vocabulary_from_data(
304
+ datasets: DatasetDict,
305
+ word_delimiter_token: Optional[str] = None,
306
+ unk_token: Optional[str] = None,
307
+ pad_token: Optional[str] = None,
308
+ ):
309
+ # Given training and test labels create vocabulary
310
+ def extract_all_chars(batch):
311
+ all_text = " ".join(batch["target_text"])
312
+ vocab = list(set(all_text))
313
+ return {"vocab": [vocab], "all_text": [all_text]}
314
+
315
+ vocabs = datasets.map(
316
+ extract_all_chars,
317
+ batched=True,
318
+ batch_size=-1,
319
+ keep_in_memory=True,
320
+ remove_columns=datasets["train"].column_names,
321
+ )
322
+
323
+ # take union of all unique characters in each dataset
324
+ vocab_set = functools.reduce(
325
+ lambda vocab_1, vocab_2: set(vocab_1["vocab"][0]) | set(vocab_2["vocab"][0]), vocabs.values()
326
+ )
327
+
328
+ vocab_dict = {v: k for k, v in enumerate(sorted(list(vocab_set)))}
329
+
330
+ # replace white space with delimiter token
331
+ if word_delimiter_token is not None:
332
+ vocab_dict[word_delimiter_token] = vocab_dict[" "]
333
+ del vocab_dict[" "]
334
+
335
+ # add unk and pad token
336
+ if unk_token is not None:
337
+ vocab_dict[unk_token] = len(vocab_dict)
338
+
339
+ if pad_token is not None:
340
+ vocab_dict[pad_token] = len(vocab_dict)
341
+
342
+ return vocab_dict
343
+
344
+
345
+ def main():
346
+ # See all possible arguments in src/transformers/training_args.py
347
+ # or by passing the --help flag to this script.
348
+ # We now keep distinct sets of args, for a cleaner separation of concerns.
349
+
350
+ parser = HfArgumentParser((ModelArguments, DataTrainingArguments, TrainingArguments))
351
+ if len(sys.argv) == 2 and sys.argv[1].endswith(".json"):
352
+ # If we pass only one argument to the script and it's the path to a json file,
353
+ # let's parse it to get our arguments.
354
+ model_args, data_args, training_args = parser.parse_json_file(json_file=os.path.abspath(sys.argv[1]))
355
+ else:
356
+ model_args, data_args, training_args = parser.parse_args_into_dataclasses()
357
+
358
+ # Detecting last checkpoint.
359
+ last_checkpoint = None
360
+ if os.path.isdir(training_args.output_dir) and training_args.do_train and not training_args.overwrite_output_dir:
361
+ last_checkpoint = get_last_checkpoint(training_args.output_dir)
362
+ if last_checkpoint is None and len(os.listdir(training_args.output_dir)) > 0:
363
+ raise ValueError(
364
+ f"Output directory ({training_args.output_dir}) already exists and is not empty. "
365
+ "Use --overwrite_output_dir to overcome."
366
+ )
367
+ elif last_checkpoint is not None:
368
+ logger.info(
369
+ f"Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change "
370
+ "the `--output_dir` or add `--overwrite_output_dir` to train from scratch."
371
+ )
372
+
373
+ # Setup logging
374
+ logging.basicConfig(
375
+ format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
376
+ datefmt="%m/%d/%Y %H:%M:%S",
377
+ handlers=[logging.StreamHandler(sys.stdout)],
378
+ )
379
+ logger.setLevel(logging.INFO if is_main_process(training_args.local_rank) else logging.WARN)
380
+
381
+ # Log on each process the small summary:
382
+ logger.warning(
383
+ f"Process rank: {training_args.local_rank}, device: {training_args.device}, n_gpu: {training_args.n_gpu}"
384
+ f"distributed training: {bool(training_args.local_rank != -1)}, 16-bits training: {training_args.fp16}"
385
+ )
386
+ # Set the verbosity to info of the Transformers logger (on main process only):
387
+ if is_main_process(training_args.local_rank):
388
+ transformers.utils.logging.set_verbosity_info()
389
+ logger.info("Training/evaluation parameters %s", training_args)
390
+
391
+ # Set seed before initializing model.
392
+ set_seed(training_args.seed)
393
+
394
+ # 1. First, let's load the dataset
395
+ raw_datasets = DatasetDict()
396
+
397
+ if training_args.do_train:
398
+ raw_datasets["train"] = load_dataset(
399
+ data_args.dataset_name, data_args.dataset_config_name, split=data_args.train_split_name, use_auth_token=data_args.use_auth_token
400
+ )
401
+
402
+ if data_args.audio_column_name not in raw_datasets["train"].column_names:
403
+ raise ValueError(
404
+ f"--audio_column_name '{data_args.audio_column_name}' not found in dataset '{data_args.dataset_name}'. "
405
+ "Make sure to set `--audio_column_name` to the correct audio column - one of "
406
+ f"{', '.join(raw_datasets['train'].column_names)}."
407
+ )
408
+
409
+ if data_args.text_column_name not in raw_datasets["train"].column_names:
410
+ raise ValueError(
411
+ f"--text_column_name {data_args.text_column_name} not found in dataset '{data_args.dataset_name}'. "
412
+ "Make sure to set `--text_column_name` to the correct text column - one of "
413
+ f"{', '.join(raw_datasets['train'].column_names)}."
414
+ )
415
+
416
+ if data_args.max_train_samples is not None:
417
+ raw_datasets["train"] = raw_datasets["train"].select(range(data_args.max_train_samples))
418
+
419
+ if training_args.do_eval:
420
+ raw_datasets["eval"] = load_dataset(
421
+ data_args.dataset_name, data_args.dataset_config_name, split=data_args.eval_split_name, use_auth_token=data_args.use_auth_token
422
+ )
423
+
424
+ if data_args.max_eval_samples is not None:
425
+ raw_datasets["eval"] = raw_datasets["eval"].select(range(data_args.max_eval_samples))
426
+
427
+ # 2. We remove some special characters from the datasets
428
+ # that make training complicated and do not help in transcribing the speech
429
+ # E.g. characters, such as `,` and `.` do not really have an acoustic characteristic
430
+ # that could be easily picked up by the model
431
+ chars_to_ignore_regex = (
432
+ f'[{"".join(data_args.chars_to_ignore)}]' if data_args.chars_to_ignore is not None else None
433
+ )
434
+ text_column_name = data_args.text_column_name
435
+
436
+ def remove_special_characters(batch):
437
+ if chars_to_ignore_regex is not None:
438
+ batch["target_text"] = re.sub(chars_to_ignore_regex, "", batch[text_column_name]).lower() + " "
439
+ else:
440
+ batch["target_text"] = batch[text_column_name].lower() + " "
441
+ return batch
442
+
443
+ with training_args.main_process_first(desc="dataset map special characters removal"):
444
+ raw_datasets = raw_datasets.map(
445
+ remove_special_characters,
446
+ remove_columns=[text_column_name],
447
+ desc="remove special characters from datasets",
448
+ )
449
+
450
+ # save special tokens for tokenizer
451
+ word_delimiter_token = data_args.word_delimiter_token
452
+ unk_token = data_args.unk_token
453
+ pad_token = data_args.pad_token
454
+
455
+ # 3. Next, let's load the config as we might need it to create
456
+ # the tokenizer
457
+ # load config
458
+ config = AutoConfig.from_pretrained(
459
+ model_args.model_name_or_path, cache_dir=model_args.cache_dir, use_auth_token=data_args.use_auth_token
460
+ )
461
+
462
+ # 4. Next, if no tokenizer file is defined,
463
+ # we create the vocabulary of the model by extracting all unique characters from
464
+ # the training and evaluation datasets
465
+ # We need to make sure that only first rank saves vocabulary
466
+ # make sure all processes wait until vocab is created
467
+ tokenizer_name_or_path = model_args.tokenizer_name_or_path
468
+ tokenizer_kwargs = {}
469
+ if tokenizer_name_or_path is None:
470
+ # save vocab in training output dir
471
+ tokenizer_name_or_path = training_args.output_dir
472
+
473
+ vocab_file = os.path.join(tokenizer_name_or_path, "vocab.json")
474
+
475
+ with training_args.main_process_first():
476
+ if training_args.overwrite_output_dir and os.path.isfile(vocab_file):
477
+ os.remove(vocab_file)
478
+
479
+ with training_args.main_process_first(desc="dataset map vocabulary creation"):
480
+ if not os.path.isfile(vocab_file):
481
+ os.makedirs(tokenizer_name_or_path, exist_ok=True)
482
+ vocab_dict = create_vocabulary_from_data(
483
+ raw_datasets,
484
+ word_delimiter_token=word_delimiter_token,
485
+ unk_token=unk_token,
486
+ pad_token=pad_token,
487
+ )
488
+
489
+ # save vocab dict to be loaded into tokenizer
490
+ with open(vocab_file, "w") as file:
491
+ json.dump(vocab_dict, file)
492
+
493
+ # if tokenizer has just been created
494
+ # it is defined by `tokenizer_class` if present in config else by `model_type`
495
+ tokenizer_kwargs = {
496
+ "config": config if config.tokenizer_class is not None else None,
497
+ "tokenizer_type": config.model_type if config.tokenizer_class is None else None,
498
+ "unk_token": unk_token,
499
+ "pad_token": pad_token,
500
+ "word_delimiter_token": word_delimiter_token,
501
+ }
502
+
503
+ # 5. Now we can instantiate the feature extractor, tokenizer and model
504
+ # Note for distributed training, the .from_pretrained methods guarantee that only
505
+ # one local process can concurrently download model & vocab.
506
+
507
+ # load feature_extractor and tokenizer
508
+ tokenizer = AutoTokenizer.from_pretrained(
509
+ tokenizer_name_or_path,
510
+ use_auth_token=data_args.use_auth_token,
511
+ **tokenizer_kwargs,
512
+ )
513
+ feature_extractor = AutoFeatureExtractor.from_pretrained(
514
+ model_args.model_name_or_path, cache_dir=model_args.cache_dir, use_auth_token=data_args.use_auth_token
515
+ )
516
+
517
+ # adapt config
518
+ config.update(
519
+ {
520
+ "feat_proj_dropout": model_args.feat_proj_dropout,
521
+ "attention_dropout": model_args.attention_dropout,
522
+ "hidden_dropout": model_args.hidden_dropout,
523
+ "final_dropout": model_args.final_dropout,
524
+ "mask_time_prob": model_args.mask_time_prob,
525
+ "mask_time_length": model_args.mask_time_length,
526
+ "mask_feature_prob": model_args.mask_feature_prob,
527
+ "mask_feature_length": model_args.mask_feature_length,
528
+ "gradient_checkpointing": training_args.gradient_checkpointing,
529
+ "layerdrop": model_args.layerdrop,
530
+ "ctc_loss_reduction": model_args.ctc_loss_reduction,
531
+ "pad_token_id": tokenizer.pad_token_id,
532
+ "vocab_size": len(tokenizer),
533
+ "activation_dropout": model_args.activation_dropout,
534
+ }
535
+ )
536
+
537
+ # create model
538
+ model = AutoModelForCTC.from_pretrained(
539
+ model_args.model_name_or_path,
540
+ cache_dir=model_args.cache_dir,
541
+ config=config,
542
+ use_auth_token=data_args.use_auth_token,
543
+ )
544
+
545
+ # freeze encoder
546
+ if model_args.freeze_feature_encoder:
547
+ model.freeze_feature_encoder()
548
+
549
+ # 6. Now we preprocess the datasets including loading the audio, resampling and normalization
550
+ # Thankfully, `datasets` takes care of automatically loading and resampling the audio,
551
+ # so that we just need to set the correct target sampling rate and normalize the input
552
+ # via the `feature_extractor`
553
+
554
+ # make sure that dataset decodes audio with correct sampling rate
555
+ dataset_sampling_rate = next(iter(raw_datasets.values())).features[data_args.audio_column_name].sampling_rate
556
+ if dataset_sampling_rate != feature_extractor.sampling_rate:
557
+ raw_datasets = raw_datasets.cast_column(
558
+ data_args.audio_column_name, datasets.features.Audio(sampling_rate=feature_extractor.sampling_rate)
559
+ )
560
+
561
+ # derive max & min input length for sample rate & max duration
562
+ max_input_length = data_args.max_duration_in_seconds * feature_extractor.sampling_rate
563
+ min_input_length = data_args.min_duration_in_seconds * feature_extractor.sampling_rate
564
+ audio_column_name = data_args.audio_column_name
565
+ num_workers = data_args.preprocessing_num_workers
566
+
567
+ # `phoneme_language` is only relevant if the model is fine-tuned on phoneme classification
568
+ phoneme_language = data_args.phoneme_language
569
+
570
+ # Preprocessing the datasets.
571
+ # We need to read the audio files as arrays and tokenize the targets.
572
+ def prepare_dataset(batch):
573
+ # load audio
574
+ sample = batch[audio_column_name]
575
+
576
+ inputs = feature_extractor(sample["array"], sampling_rate=sample["sampling_rate"])
577
+ batch["input_values"] = inputs.input_values[0]
578
+ batch["input_length"] = len(batch["input_values"])
579
+
580
+ # encode targets
581
+ additional_kwargs = {}
582
+ if phoneme_language is not None:
583
+ additional_kwargs["phonemizer_lang"] = phoneme_language
584
+
585
+ batch["labels"] = tokenizer(batch["target_text"], **additional_kwargs).input_ids
586
+ return batch
587
+
588
+ with training_args.main_process_first(desc="dataset map preprocessing"):
589
+ vectorized_datasets = raw_datasets.map(
590
+ prepare_dataset,
591
+ remove_columns=next(iter(raw_datasets.values())).column_names,
592
+ num_proc=num_workers,
593
+ desc="preprocess datasets",
594
+ )
595
+
596
+ def is_audio_in_length_range(length):
597
+ return length > min_input_length and length < max_input_length
598
+
599
+ # filter data that is shorter than min_input_length
600
+ vectorized_datasets = vectorized_datasets.filter(
601
+ is_audio_in_length_range,
602
+ num_proc=num_workers,
603
+ input_columns=["input_length"],
604
+ )
605
+
606
+ # 7. Next, we can prepare the training.
607
+ # Let's use word error rate (WER) as our evaluation metric,
608
+ # instantiate a data collator and the trainer
609
+
610
+ # Define evaluation metrics during training, *i.e.* word error rate, character error rate
611
+ eval_metrics = {metric: load_metric(metric) for metric in data_args.eval_metrics}
612
+
613
+ # for large datasets it is advised to run the preprocessing on a
614
+ # single machine first with ``args.preprocessing_only`` since there will mostly likely
615
+ # be a timeout when running the script in distributed mode.
616
+ # In a second step ``args.preprocessing_only`` can then be set to `False` to load the
617
+ # cached dataset
618
+ if data_args.preprocessing_only:
619
+ logger.info(f"Data preprocessing finished. Files cached at {vectorized_datasets.cache_files}")
620
+ return
621
+
622
+ def compute_metrics(pred):
623
+ pred_logits = pred.predictions
624
+ pred_ids = np.argmax(pred_logits, axis=-1)
625
+
626
+ pred.label_ids[pred.label_ids == -100] = tokenizer.pad_token_id
627
+
628
+ pred_str = tokenizer.batch_decode(pred_ids)
629
+ # we do not want to group tokens when computing the metrics
630
+ label_str = tokenizer.batch_decode(pred.label_ids, group_tokens=False)
631
+
632
+ metrics = {k: v.compute(predictions=pred_str, references=label_str) for k, v in eval_metrics.items()}
633
+
634
+ return metrics
635
+
636
+ # Now save everything to be able to create a single processor later
637
+ if is_main_process(training_args.local_rank):
638
+ # save feature extractor, tokenizer and config
639
+ feature_extractor.save_pretrained(training_args.output_dir)
640
+ tokenizer.save_pretrained(training_args.output_dir)
641
+ config.save_pretrained(training_args.output_dir)
642
+
643
+ try:
644
+ processor = AutoProcessor.from_pretrained(training_args.output_dir)
645
+ except (OSError, KeyError):
646
+ warnings.warn(
647
+ "Loading a processor from a feature extractor config that does not"
648
+ " include a `processor_class` attribute is deprecated and will be removed in v5. Please add the following "
649
+ " attribute to your `preprocessor_config.json` file to suppress this warning: "
650
+ " `'processor_class': 'Wav2Vec2Processor'`",
651
+ FutureWarning,
652
+ )
653
+ processor = Wav2Vec2Processor.from_pretrained(training_args.output_dir)
654
+
655
+ # Instantiate custom data collator
656
+ data_collator = DataCollatorCTCWithPadding(processor=processor)
657
+
658
+ # Initialize Trainer
659
+ trainer = Trainer(
660
+ model=model,
661
+ data_collator=data_collator,
662
+ args=training_args,
663
+ compute_metrics=compute_metrics,
664
+ train_dataset=vectorized_datasets["train"] if training_args.do_train else None,
665
+ eval_dataset=vectorized_datasets["eval"] if training_args.do_eval else None,
666
+ tokenizer=feature_extractor,
667
+ )
668
+
669
+ # 8. Finally, we can start training
670
+
671
+ # Training
672
+ if training_args.do_train:
673
+
674
+ # use last checkpoint if exist
675
+ if last_checkpoint is not None:
676
+ checkpoint = last_checkpoint
677
+ elif os.path.isdir(model_args.model_name_or_path):
678
+ checkpoint = model_args.model_name_or_path
679
+ else:
680
+ checkpoint = None
681
+
682
+ train_result = trainer.train(resume_from_checkpoint=checkpoint)
683
+ trainer.save_model()
684
+
685
+ metrics = train_result.metrics
686
+ max_train_samples = (
687
+ data_args.max_train_samples
688
+ if data_args.max_train_samples is not None
689
+ else len(vectorized_datasets["train"])
690
+ )
691
+ metrics["train_samples"] = min(max_train_samples, len(vectorized_datasets["train"]))
692
+
693
+ trainer.log_metrics("train", metrics)
694
+ trainer.save_metrics("train", metrics)
695
+ trainer.save_state()
696
+
697
+ # Evaluation
698
+ results = {}
699
+ if training_args.do_eval:
700
+ logger.info("*** Evaluate ***")
701
+ metrics = trainer.evaluate()
702
+ max_eval_samples = (
703
+ data_args.max_eval_samples if data_args.max_eval_samples is not None else len(vectorized_datasets["eval"])
704
+ )
705
+ metrics["eval_samples"] = min(max_eval_samples, len(vectorized_datasets["eval"]))
706
+
707
+ trainer.log_metrics("eval", metrics)
708
+ trainer.save_metrics("eval", metrics)
709
+
710
+ # Write model card and (optionally) push to hub
711
+ config_name = data_args.dataset_config_name if data_args.dataset_config_name is not None else "na"
712
+ kwargs = {
713
+ "finetuned_from": model_args.model_name_or_path,
714
+ "tasks": "speech-recognition",
715
+ "tags": ["automatic-speech-recognition", data_args.dataset_name],
716
+ "dataset_args": f"Config: {config_name}, Training split: {data_args.train_split_name}, Eval split: {data_args.eval_split_name}",
717
+ "dataset": f"{data_args.dataset_name.upper()} - {config_name.upper()}",
718
+ }
719
+ if "common_voice" in data_args.dataset_name:
720
+ kwargs["language"] = config_name
721
+
722
+ if training_args.push_to_hub:
723
+ trainer.push_to_hub(**kwargs)
724
+ else:
725
+ trainer.create_model_card(**kwargs)
726
+
727
+ return results
728
+
729
+
730
+ if __name__ == "__main__":
731
+ main()
runs/Feb02_22-30-13_brutasse/1643841025.8737185/events.out.tfevents.1643841025.brutasse.21084.1 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5900f695e767278150f119bbc534f29d95ed6852934cd2c8b6976444c7b517b4
3
+ size 4744
runs/Feb02_22-30-13_brutasse/events.out.tfevents.1643841025.brutasse.21084.0 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:594838aadfc207129a42efdcea909d0eacd2e7a7e6880b7dd38b497707240650
3
+ size 5781
special_tokens_map.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"bos_token": "<s>", "eos_token": "</s>", "unk_token": "[UNK]", "pad_token": "[PAD]", "additional_special_tokens": [{"content": "<s>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true}, {"content": "</s>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true}]}
tokenizer_config.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"unk_token": "[UNK]", "bos_token": "<s>", "eos_token": "</s>", "pad_token": "[PAD]", "do_lower_case": false, "word_delimiter_token": "|", "special_tokens_map_file": null, "tokenizer_file": null, "name_or_path": "./", "tokenizer_class": "Wav2Vec2CTCTokenizer"}
training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d41e44089628708752fc1e2b1e0676d97378d11ff950b5f208852f14cdb9f89b
3
+ size 2991
vocab.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"a": 1, "b": 2, "c": 3, "d": 4, "e": 5, "f": 6, "g": 7, "h": 8, "i": 9, "j": 10, "k": 11, "l": 12, "m": 13, "n": 14, "o": 15, "p": 16, "q": 17, "r": 18, "s": 19, "t": 20, "u": 21, "v": 22, "w": 23, "x": 24, "y": 25, "z": 26, "ä": 27, "å": 28, "é": 29, "ô": 30, "ö": 31, "ü": 32, "|": 0, "[UNK]": 33, "[PAD]": 34}
wandb/debug-internal.log ADDED
@@ -0,0 +1 @@
 
 
1
+ run-20220202_223026-1j0459xm/logs/debug-internal.log
wandb/debug.log ADDED
@@ -0,0 +1 @@
 
 
1
+ run-20220202_223026-1j0459xm/logs/debug.log
wandb/latest-run ADDED
@@ -0,0 +1 @@
 
 
1
+ run-20220202_223026-1j0459xm
wandb/run-20220202_223026-1j0459xm/files/conda-environment.yaml ADDED
@@ -0,0 +1,374 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ name: hugging_face
2
+ channels:
3
+ - pytorch-nightly
4
+ - defaults
5
+ dependencies:
6
+ - _libgcc_mutex=0.1=main
7
+ - _openmp_mutex=4.5=1_gnu
8
+ - blas=1.0=mkl
9
+ - brotlipy=0.7.0=py38h27cfd23_1003
10
+ - bzip2=1.0.8=h7b6447c_0
11
+ - ca-certificates=2021.10.26=h06a4308_2
12
+ - certifi=2021.10.8=py38h06a4308_2
13
+ - charset-normalizer=2.0.4=pyhd3eb1b0_0
14
+ - cryptography=36.0.0=py38h9ce1e76_0
15
+ - cudatoolkit=11.3.1=h2bc3f7f_2
16
+ - ffmpeg=4.2.2=h20bf706_0
17
+ - freetype=2.11.0=h70c0345_0
18
+ - giflib=5.2.1=h7b6447c_0
19
+ - gmp=6.2.1=h2531618_2
20
+ - gnutls=3.6.15=he1e5248_0
21
+ - intel-openmp=2021.4.0=h06a4308_3561
22
+ - jpeg=9b=h024ee3a_2
23
+ - lame=3.100=h7b6447c_0
24
+ - lcms2=2.12=h3be6417_0
25
+ - ld_impl_linux-64=2.35.1=h7274673_9
26
+ - libffi=3.3=he6710b0_2
27
+ - libgcc-ng=9.3.0=h5101ec6_17
28
+ - libgfortran-ng=7.5.0=ha8ba4b0_17
29
+ - libgfortran4=7.5.0=ha8ba4b0_17
30
+ - libgomp=9.3.0=h5101ec6_17
31
+ - libiconv=1.15=h63c8f33_5
32
+ - libidn2=2.3.2=h7f8727e_0
33
+ - libopus=1.3.1=h7b6447c_0
34
+ - libpng=1.6.37=hbc83047_0
35
+ - libstdcxx-ng=9.3.0=hd4cf53a_17
36
+ - libtasn1=4.16.0=h27cfd23_0
37
+ - libtiff=4.2.0=h85742a9_0
38
+ - libunistring=0.9.10=h27cfd23_0
39
+ - libuv=1.40.0=h7b6447c_0
40
+ - libvpx=1.7.0=h439df22_0
41
+ - libwebp=1.2.0=h89dd481_0
42
+ - libwebp-base=1.2.0=h27cfd23_0
43
+ - lz4-c=1.9.3=h295c915_1
44
+ - mkl=2021.4.0=h06a4308_640
45
+ - mkl-service=2.4.0=py38h7f8727e_0
46
+ - mkl_fft=1.3.0=py38h42c9631_2
47
+ - mkl_random=1.2.1=py38ha9443f7_2
48
+ - mpi=1.0=mpich
49
+ - mpi4py=3.0.3=py38h028fd6f_0
50
+ - mpich=3.3.2=hc856adb_0
51
+ - ncurses=6.3=h7f8727e_2
52
+ - nettle=3.7.3=hbbd107a_1
53
+ - numpy-base=1.20.2=py38hfae3a4d_0
54
+ - olefile=0.46=pyhd3eb1b0_0
55
+ - openh264=2.1.0=hd408876_0
56
+ - openssl=1.1.1m=h7f8727e_0
57
+ - pillow=8.4.0=py38h5aabda8_0
58
+ - pyopenssl=21.0.0=pyhd3eb1b0_1
59
+ - pysocks=1.7.1=py38h06a4308_0
60
+ - python=3.8.12=h12debd9_0
61
+ - pytorch-mutex=1.0=cuda
62
+ - readline=8.1=h27cfd23_0
63
+ - setuptools=58.0.4=py38h06a4308_0
64
+ - six=1.16.0=pyhd3eb1b0_0
65
+ - sqlite=3.36.0=hc218d9a_0
66
+ - tk=8.6.11=h1ccaba5_0
67
+ - torchvision=0.12.0.dev20220120=py38_cu113
68
+ - typing_extensions=3.10.0.2=pyh06a4308_0
69
+ - wheel=0.37.0=pyhd3eb1b0_1
70
+ - x264=1!157.20191217=h7b6447c_0
71
+ - xz=5.2.5=h7b6447c_0
72
+ - zlib=1.2.11=h7b6447c_3
73
+ - zstd=1.4.9=haebb681_0
74
+ - pip:
75
+ - absl-py==0.12.0
76
+ - accelerate==0.5.0.dev0
77
+ - aiohttp==3.7.4.post0
78
+ - aiohttp-cors==0.7.0
79
+ - aioredis==1.3.1
80
+ - alabaster==0.7.12
81
+ - alembic==1.6.5
82
+ - antlr4-python3-runtime==4.8
83
+ - apache-beam==2.33.0
84
+ - apipkg==1.5
85
+ - appdirs==1.4.4
86
+ - apscheduler==3.7.0
87
+ - argparse==1.4.0
88
+ - arrow==1.1.0
89
+ - astunparse==1.6.3
90
+ - async-timeout==3.0.1
91
+ - attrs==21.2.0
92
+ - audioread==2.1.9
93
+ - avro-python3==1.9.2.1
94
+ - babel==2.9.1
95
+ - backcall==0.2.0
96
+ - beautifulsoup4==4.10.0
97
+ - binaryornot==0.4.4
98
+ - bitarray==2.3.4
99
+ - bitsandbytes-cuda110==0.26.0
100
+ - black==21.4b2
101
+ - blessings==1.7
102
+ - bottle==0.12.19
103
+ - brotli==1.0.9
104
+ - cachetools==4.2.2
105
+ - cffi==1.14.5
106
+ - chardet==4.0.0
107
+ - chex==0.0.7
108
+ - clang==5.0
109
+ - click==7.1.2
110
+ - cliff==3.8.0
111
+ - clldutils==3.10.0
112
+ - cloudpickle==2.0.0
113
+ - cmaes==0.8.2
114
+ - cmd2==2.1.2
115
+ - codecarbon==1.2.0
116
+ - colorama==0.4.4
117
+ - colorlog==5.0.1
118
+ - commonmark==0.9.1
119
+ - configparser==5.0.2
120
+ - cookiecutter==1.7.2
121
+ - crcmod==1.7
122
+ - csvw==1.11.0
123
+ - ctcdecode==1.0.3
124
+ - cycler==0.10.0
125
+ - cython==0.29.23
126
+ - dash==1.21.0
127
+ - dash-bootstrap-components==0.13.0
128
+ - dash-core-components==1.17.1
129
+ - dash-html-components==1.1.4
130
+ - dash-table==4.12.0
131
+ - datasets==1.17.0
132
+ - decorator==5.0.9
133
+ - deepspeed==0.4.5
134
+ - dill==0.3.4
135
+ - dm-tree==0.1.6
136
+ - doc-builder==0.0.1.dev0
137
+ - docker-pycreds==0.4.0
138
+ - docopt==0.6.2
139
+ - docutils==0.16
140
+ - editdistance==0.6.0
141
+ - execnet==1.8.1
142
+ - faiss-cpu==1.7.1
143
+ - faiss-gpu==1.7.1.post2
144
+ - fastavro==1.4.7
145
+ - filelock==3.0.12
146
+ - fire==0.4.0
147
+ - flake8==3.9.2
148
+ - flask==1.1.4
149
+ - flask-compress==1.10.1
150
+ - flatbuffers==1.12
151
+ - flax==0.3.4
152
+ - fsspec==2021.11.1
153
+ - fugashi==1.1.0
154
+ - future==0.18.2
155
+ - fvcore==0.1.5.post20220119
156
+ - gast==0.4.0
157
+ - gdown==4.2.0
158
+ - gitdb==4.0.7
159
+ - gitpython==3.1.18
160
+ - google-api-core==1.31.2
161
+ - google-auth==1.30.1
162
+ - google-auth-oauthlib==0.4.4
163
+ - google-pasta==0.2.0
164
+ - googleapis-common-protos==1.53.0
165
+ - gpustat==0.6.0
166
+ - greenlet==1.1.1
167
+ - grpcio==1.41.0
168
+ - h5py==3.1.0
169
+ - hdfs==2.6.0
170
+ - hiredis==2.0.0
171
+ - httplib2==0.19.1
172
+ - huggingface-hub==0.4.0
173
+ - hydra-core==1.1.1
174
+ - hypothesis==6.24.1
175
+ - idna==2.10
176
+ - imagesize==1.2.0
177
+ - importlib-resources==5.1.4
178
+ - iniconfig==1.1.1
179
+ - iopath==0.1.9
180
+ - ipadic==1.0.0
181
+ - ipdb==0.13.9
182
+ - ipython==7.26.0
183
+ - ipython-genutils==0.2.0
184
+ - isodate==0.6.0
185
+ - isort==5.8.0
186
+ - itsdangerous==1.1.0
187
+ - jams==0.3.4
188
+ - jax==0.2.19
189
+ - jaxlib==0.1.70
190
+ - jedi==0.18.0
191
+ - jinja2==2.11.3
192
+ - jinja2-time==0.2.0
193
+ - jiwer==2.2.0
194
+ - joblib==1.0.1
195
+ - jsonschema==3.2.0
196
+ - jupyter-core==4.9.1
197
+ - kenlm==0.0.0
198
+ - keras==2.7.0
199
+ - keras-nightly==2.5.0.dev2021032900
200
+ - keras-preprocessing==1.1.2
201
+ - keras2onnx==1.7.0
202
+ - kiwisolver==1.3.1
203
+ - libclang==12.0.0
204
+ - librosa==0.8.1
205
+ - llvmlite==0.36.0
206
+ - logging==0.4.9.6
207
+ - mako==1.1.4
208
+ - markdown==3.3.4
209
+ - markupsafe==1.1.1
210
+ - matplotlib==3.4.2
211
+ - matplotlib-inline==0.1.2
212
+ - mccabe==0.6.1
213
+ - mir-eval==0.6
214
+ - msgpack==1.0.2
215
+ - multidict==5.1.0
216
+ - multiprocess==0.70.11.1
217
+ - mypy-extensions==0.4.3
218
+ - nbformat==5.1.3
219
+ - ninja==1.10.2
220
+ - nltk==3.6.2
221
+ - numba==0.53.1
222
+ - numpy==1.19.5
223
+ - nvidia-ml-py3==7.352.0
224
+ - oauth2client==4.1.3
225
+ - oauthlib==3.1.1
226
+ - omegaconf==2.1.1
227
+ - onnx==1.9.0
228
+ - onnxconverter-common==1.8.1
229
+ - opencensus==0.7.13
230
+ - opencensus-context==0.1.2
231
+ - opt-einsum==3.3.0
232
+ - optax==0.0.9
233
+ - optuna==2.9.1
234
+ - orjson==3.6.4
235
+ - packaging==20.9
236
+ - pandas==1.2.4
237
+ - parameterized==0.8.1
238
+ - parso==0.8.2
239
+ - path==16.2.0
240
+ - pathspec==0.8.1
241
+ - pathtools==0.1.2
242
+ - pbr==5.6.0
243
+ - pexpect==4.8.0
244
+ - phonemizer==2.2.2
245
+ - phonetisaurus==0.3
246
+ - pickleshare==0.7.5
247
+ - pip==21.2.4
248
+ - plac==1.3.3
249
+ - plotly==5.2.1
250
+ - pluggy==0.13.1
251
+ - pooch==1.3.0
252
+ - portalocker==2.0.0
253
+ - poyo==0.5.0
254
+ - prettytable==2.1.0
255
+ - prometheus-client==0.11.0
256
+ - promise==2.3
257
+ - prompt-toolkit==3.0.18
258
+ - protobuf==3.17.2
259
+ - psutil==5.8.0
260
+ - ptyprocess==0.7.0
261
+ - py==1.10.0
262
+ - py-cpuinfo==8.0.0
263
+ - py-spy==0.3.8
264
+ - pyarrow==6.0.1
265
+ - pyasn1==0.4.8
266
+ - pyasn1-modules==0.2.8
267
+ - pybindgen==0.22.0
268
+ - pycocotools==2.0.4
269
+ - pycodestyle==2.7.0
270
+ - pycparser==2.20
271
+ - pyctcdecode==0.3.0
272
+ - pydantic==1.8.2
273
+ - pydot==1.4.2
274
+ - pyflakes==2.3.1
275
+ - pygments==2.9.0
276
+ - pygtrie==2.4.2
277
+ - pymongo==3.12.1
278
+ - pynvml==11.0.0
279
+ - pyparsing==2.4.7
280
+ - pyperclip==1.8.2
281
+ - pyrsistent==0.18.0
282
+ - pytest==6.2.4
283
+ - pytest-forked==1.3.0
284
+ - pytest-sugar==0.9.4
285
+ - pytest-xdist==2.2.1
286
+ - python-dateutil==2.8.1
287
+ - python-editor==1.0.4
288
+ - python-levenshtein==0.12.2
289
+ - python-slugify==5.0.2
290
+ - pytz==2021.1
291
+ - pyyaml==5.4.1
292
+ - ray==1.5.2
293
+ - recommonmark==0.7.1
294
+ - redis==3.5.3
295
+ - regex==2021.4.4
296
+ - requests==2.25.1
297
+ - requests-oauthlib==1.3.0
298
+ - resampy==0.2.2
299
+ - rfc3986==1.5.0
300
+ - rouge-score==0.0.4
301
+ - rsa==4.7.2
302
+ - ruamel-yaml==0.17.16
303
+ - ruamel-yaml-clib==0.2.6
304
+ - sacrebleu==1.5.1
305
+ - sacremoses==0.0.45
306
+ - scann==1.2.4
307
+ - scikit-learn==0.24.2
308
+ - scipy==1.6.3
309
+ - segments==2.2.0
310
+ - sentencepiece==0.1.94
311
+ - sentry-sdk==1.3.1
312
+ - seqio==0.0.6
313
+ - shortuuid==1.0.1
314
+ - smmap==4.0.0
315
+ - snowballstemmer==2.1.0
316
+ - sortedcontainers==2.4.0
317
+ - soundata==0.1.0
318
+ - soundfile==0.10.3.post1
319
+ - soupsieve==2.3.1
320
+ - sphinx==3.2.1
321
+ - sphinx-copybutton==0.3.1
322
+ - sphinx-markdown-tables==0.0.15
323
+ - sphinx-rtd-theme==0.4.3
324
+ - sphinxcontrib-applehelp==1.0.2
325
+ - sphinxcontrib-devhelp==1.0.2
326
+ - sphinxcontrib-htmlhelp==2.0.0
327
+ - sphinxcontrib-jsmath==1.0.1
328
+ - sphinxcontrib-qthelp==1.0.3
329
+ - sphinxcontrib-serializinghtml==1.1.5
330
+ - sphinxext-opengraph==0.4.1
331
+ - sqlalchemy==1.4.23
332
+ - stevedore==3.3.0
333
+ - subprocess32==3.5.4
334
+ - tabulate==0.8.9
335
+ - tenacity==8.0.1
336
+ - tensorboard==2.6.0
337
+ - tensorboard-data-server==0.6.1
338
+ - tensorboard-plugin-wit==1.8.0
339
+ - tensorboardx==2.4
340
+ - tensorflow==2.7.0
341
+ - tensorflow-estimator==2.7.0
342
+ - tensorflow-hub==0.12.0
343
+ - tensorflow-io-gcs-filesystem==0.23.1
344
+ - tensorflow-metadata==1.2.0
345
+ - tensorflow-text==2.6.0
346
+ - termcolor==1.1.0
347
+ - text-unidecode==1.3
348
+ - tfds-nightly==4.4.0.dev202110120106
349
+ - threadpoolctl==2.1.0
350
+ - timeout-decorator==0.5.0
351
+ - timm==0.4.12
352
+ - tokenizers==0.10.3
353
+ - toml==0.10.2
354
+ - toolz==0.11.1
355
+ - torch==1.10.1+cu113
356
+ - torchaudio==0.10.1+cu113
357
+ - tqdm==4.62.3
358
+ - traitlets==5.0.5
359
+ - triton==1.0.0
360
+ - tzlocal==2.1
361
+ - unidic==1.0.3
362
+ - unidic-lite==1.0.8
363
+ - uritemplate==4.1.1
364
+ - urllib3==1.26.5
365
+ - wandb==0.12.1
366
+ - wasabi==0.8.2
367
+ - wcwidth==0.2.5
368
+ - werkzeug==1.0.1
369
+ - wrapt==1.12.1
370
+ - xxhash==2.0.2
371
+ - yacs==0.1.8
372
+ - yarl==1.6.3
373
+ - zipp==3.4.1
374
+ prefix: /home/patrick/anaconda3/envs/hugging_face
wandb/run-20220202_223026-1j0459xm/files/config.yaml ADDED
The diff for this file is too large to render. See raw diff
 
wandb/run-20220202_223026-1j0459xm/files/output.log ADDED
@@ -0,0 +1,405 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+
2
+
3
+
4
+
5
+
6
+
7
+
8
+
9
+
10
+
11
+
12
+
13
+
14
+
15
+
16
+
17
+
18
+
19
+
20
+
21
+
22
+
23
+
24
+
25
+
26
+
27
+
28
+
29
+
30
+
31
+
32
+
33
+
34
+
35
+
36
+
37
+
38
+
39
+
40
+
41
+
42
+
43
+
44
+
45
+
46
+
47
+
48
+
49
+
50
+
51
+
52
+
53
+
54
+
55
+
56
+
57
+
58
+
59
+
60
+
61
+
62
+ 1%|▎ | 100/18250 [02:02<4:08:51, 1.22it/s]
63
+
64
+
65
+
66
+
67
+
68
+
69
+
70
+
71
+
72
+
73
+
74
+
75
+
76
+
77
+
78
+
79
+
80
+
81
+
82
+
83
+
84
+
85
+
86
+
87
+
88
+
89
+
90
+
91
+
92
+
93
+
94
+
95
+
96
+
97
+
98
+
99
+
100
+
101
+
102
+
103
+
104
+
105
+
106
+
107
+
108
+
109
+
110
+
111
+
112
+
113
+
114
+
115
+
116
+
117
+
118
+
119
+
120
+
121
+
122
+
123
+ 1%|▌ | 200/18250 [04:03<4:10:12, 1.20it/s]
124
+
125
+
126
+
127
+
128
+
129
+
130
+
131
+
132
+
133
+
134
+
135
+
136
+
137
+
138
+
139
+
140
+
141
+
142
+
143
+
144
+
145
+
146
+
147
+
148
+
149
+
150
+
151
+
152
+
153
+
154
+
155
+
156
+
157
+
158
+
159
+
160
+
161
+
162
+
163
+
164
+
165
+
166
+
167
+
168
+
169
+
170
+
171
+
172
+
173
+
174
+
175
+
176
+
177
+
178
+
179
+
180
+
181
+
182
+
183
+ 2%|▊ | 300/18250 [06:04<4:06:59, 1.21it/s]
184
+
185
+
186
+
187
+
188
+
189
+
190
+
191
+
192
+
193
+
194
+
195
+
196
+
197
+
198
+
199
+
200
+
201
+
202
+
203
+
204
+
205
+
206
+
207
+
208
+
209
+
210
+
211
+
212
+
213
+
214
+
215
+
216
+
217
+
218
+
219
+
220
+
221
+
222
+
223
+
224
+
225
+
226
+
227
+
228
+
229
+
230
+
231
+
232
+
233
+
234
+
235
+
236
+
237
+
238
+
239
+
240
+
241
+
242
+
243
+
244
+
245
+ 2%|█▏ | 399/18250 [08:07<5:07:58, 1.04s/it]
246
+
247
+
248
+
249
+
250
+
251
+
252
+
253
+
254
+
255
+
256
+
257
+
258
+
259
+
260
+
261
+
262
+
263
+
264
+
265
+
266
+
267
+
268
+
269
+
270
+
271
+
272
+
273
+
274
+
275
+
276
+
277
+
278
+
279
+
280
+
281
+
282
+
283
+
284
+
285
+
286
+
287
+
288
+
289
+
290
+
291
+
292
+
293
+
294
+
295
+
296
+
297
+
298
+
299
+
300
+
301
+
302
+
303
+
304
+ 3%|█▍ | 500/18250 [10:09<5:23:38, 1.09s/it]The following columns in the evaluation set don't have a corresponding argument in `Wav2Vec2ForCTC.forward` and have been ignored: input_length. If input_length are not expected by `Wav2Vec2ForCTC.forward`, you can safely ignore this message.
305
+ ***** Running Evaluation *****
306
+ Num examples = 4843
307
+ Batch size = 8
308
+ {'loss': 3.3224, 'learning_rate': 1.8712499999999997e-05, 'epoch': 1.37}
309
+
310
+
311
+
312
+
313
+
314
+
315
+
316
+
317
+
318
+
319
+
320
+
321
+
322
+
323
+
324
+
325
+
326
+
327
+
328
+
329
+
330
+
331
+
332
+
333
+
334
+
335
+
336
+
337
+
338
+
339
+
340
+
341
+
342
+
343
+
344
+
345
+
346
+
347
+
348
+
349
+
350
+
351
+
352
+
353
+
354
+
355
+
356
+
357
+
358
+
359
+
360
+
361
+
362
+
363
+
364
+
365
+
366
+
367
+
368
+
369
+
370
+
371
+
372
+
373
+
374
+
375
+
376
+
377
+
378
+
379
+
380
+
381
+
382
+
383
+
384
+
385
+
386
+
387
+
388
+
389
+
390
+
391
+
392
+
393
+
394
+
395
+
396
+
397
+
398
+
399
+
400
+
401
+
402
+ Configuration saved in ./checkpoint-500/config.json
403
+ {'eval_loss': 3.335383415222168, 'eval_wer': 1.0, 'eval_runtime': 191.0801, 'eval_samples_per_second': 25.345, 'eval_steps_per_second': 3.171, 'epoch': 1.37}
404
+ Model weights saved in ./checkpoint-500/pytorch_model.bin
405
+ Configuration saved in ./checkpoint-500/preprocessor_config.json
wandb/run-20220202_223026-1j0459xm/files/requirements.txt ADDED
@@ -0,0 +1,318 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ absl-py==0.12.0
2
+ accelerate==0.5.0.dev0
3
+ aiohttp-cors==0.7.0
4
+ aiohttp==3.7.4.post0
5
+ aioredis==1.3.1
6
+ alabaster==0.7.12
7
+ alembic==1.6.5
8
+ antlr4-python3-runtime==4.8
9
+ apache-beam==2.33.0
10
+ apipkg==1.5
11
+ appdirs==1.4.4
12
+ apscheduler==3.7.0
13
+ argparse==1.4.0
14
+ arrow==1.1.0
15
+ astunparse==1.6.3
16
+ async-timeout==3.0.1
17
+ attrs==21.2.0
18
+ audioread==2.1.9
19
+ avro-python3==1.9.2.1
20
+ babel==2.9.1
21
+ backcall==0.2.0
22
+ beautifulsoup4==4.10.0
23
+ binaryornot==0.4.4
24
+ bitarray==2.3.4
25
+ bitsandbytes-cuda110==0.26.0
26
+ black==21.4b2
27
+ blessings==1.7
28
+ bottle==0.12.19
29
+ brotli==1.0.9
30
+ brotlipy==0.7.0
31
+ cachetools==4.2.2
32
+ certifi==2021.10.8
33
+ cffi==1.15.0
34
+ chardet==4.0.0
35
+ charset-normalizer==2.0.4
36
+ chex==0.0.7
37
+ clang==5.0
38
+ click==7.1.2
39
+ cliff==3.8.0
40
+ clldutils==3.10.0
41
+ cloudpickle==2.0.0
42
+ cmaes==0.8.2
43
+ cmd2==2.1.2
44
+ codecarbon==1.2.0
45
+ colorama==0.4.4
46
+ colorlog==5.0.1
47
+ commonmark==0.9.1
48
+ configparser==5.0.2
49
+ cookiecutter==1.7.2
50
+ crcmod==1.7
51
+ cryptography==36.0.0
52
+ csvw==1.11.0
53
+ ctcdecode==1.0.3
54
+ cycler==0.10.0
55
+ cython==0.29.23
56
+ dash-bootstrap-components==0.13.0
57
+ dash-core-components==1.17.1
58
+ dash-html-components==1.1.4
59
+ dash-table==4.12.0
60
+ dash==1.21.0
61
+ datasets==1.17.0
62
+ decorator==5.0.9
63
+ deepspeed==0.4.5
64
+ dill==0.3.4
65
+ dm-tree==0.1.6
66
+ doc-builder==0.0.1.dev0
67
+ docker-pycreds==0.4.0
68
+ docopt==0.6.2
69
+ docutils==0.16
70
+ editdistance==0.6.0
71
+ execnet==1.8.1
72
+ fairseq==1.0.0a0+ade9bec
73
+ faiss-cpu==1.7.1
74
+ faiss-gpu==1.7.1.post2
75
+ fastavro==1.4.7
76
+ filelock==3.0.12
77
+ fire==0.4.0
78
+ flake8==3.9.2
79
+ flask-compress==1.10.1
80
+ flask==1.1.4
81
+ flatbuffers==1.12
82
+ flax==0.3.4
83
+ fsspec==2021.11.1
84
+ fugashi==1.1.0
85
+ future==0.18.2
86
+ fvcore==0.1.5.post20220119
87
+ gast==0.4.0
88
+ gdown==4.2.0
89
+ gitdb==4.0.7
90
+ gitpython==3.1.18
91
+ google-api-core==1.31.2
92
+ google-auth-oauthlib==0.4.4
93
+ google-auth==1.30.1
94
+ google-pasta==0.2.0
95
+ googleapis-common-protos==1.53.0
96
+ gpustat==0.6.0
97
+ greenlet==1.1.1
98
+ grpcio==1.41.0
99
+ h5py==3.1.0
100
+ hdfs==2.6.0
101
+ hiredis==2.0.0
102
+ httplib2==0.19.1
103
+ huggingface-hub==0.4.0
104
+ hydra-core==1.1.1
105
+ hypothesis==6.24.1
106
+ idna==3.3
107
+ imagesize==1.2.0
108
+ importlib-resources==5.1.4
109
+ iniconfig==1.1.1
110
+ iopath==0.1.9
111
+ ipadic==1.0.0
112
+ ipdb==0.13.9
113
+ ipython-genutils==0.2.0
114
+ ipython==7.26.0
115
+ isodate==0.6.0
116
+ isort==5.8.0
117
+ itsdangerous==1.1.0
118
+ jams==0.3.4
119
+ jax==0.2.19
120
+ jaxlib==0.1.70
121
+ jedi==0.18.0
122
+ jinja2-time==0.2.0
123
+ jinja2==2.11.3
124
+ jiwer==2.2.0
125
+ joblib==1.0.1
126
+ jsonschema==3.2.0
127
+ jupyter-core==4.9.1
128
+ kenlm==0.0.0
129
+ keras-nightly==2.5.0.dev2021032900
130
+ keras-preprocessing==1.1.2
131
+ keras2onnx==1.7.0
132
+ keras==2.7.0
133
+ kiwisolver==1.3.1
134
+ libclang==12.0.0
135
+ librosa==0.8.1
136
+ llvmlite==0.36.0
137
+ logging==0.4.9.6
138
+ mako==1.1.4
139
+ markdown==3.3.4
140
+ markupsafe==1.1.1
141
+ matplotlib-inline==0.1.2
142
+ matplotlib==3.4.2
143
+ mccabe==0.6.1
144
+ mir-eval==0.6
145
+ mkl-fft==1.3.0
146
+ mkl-random==1.2.1
147
+ mkl-service==2.4.0
148
+ mpi4py==3.0.3
149
+ msgpack==1.0.2
150
+ multidict==5.1.0
151
+ multiprocess==0.70.11.1
152
+ mypy-extensions==0.4.3
153
+ nbformat==5.1.3
154
+ ninja==1.10.2
155
+ nltk==3.6.2
156
+ numba==0.53.1
157
+ numpy==1.20.2
158
+ nvidia-ml-py3==7.352.0
159
+ oauth2client==4.1.3
160
+ oauthlib==3.1.1
161
+ olefile==0.46
162
+ omegaconf==2.1.1
163
+ onnx==1.9.0
164
+ onnxconverter-common==1.8.1
165
+ opencensus-context==0.1.2
166
+ opencensus==0.7.13
167
+ opt-einsum==3.3.0
168
+ optax==0.0.9
169
+ optuna==2.9.1
170
+ orjson==3.6.4
171
+ packaging==20.9
172
+ pandas==1.2.4
173
+ parameterized==0.8.1
174
+ parso==0.8.2
175
+ path==16.2.0
176
+ pathspec==0.8.1
177
+ pathtools==0.1.2
178
+ pbr==5.6.0
179
+ pexpect==4.8.0
180
+ phonemizer==2.2.2
181
+ phonetisaurus==0.3
182
+ pickleshare==0.7.5
183
+ pillow==8.4.0
184
+ pip==21.2.4
185
+ plac==1.3.3
186
+ plotly==5.2.1
187
+ pluggy==0.13.1
188
+ pooch==1.3.0
189
+ portalocker==2.0.0
190
+ poyo==0.5.0
191
+ prettytable==2.1.0
192
+ prometheus-client==0.11.0
193
+ promise==2.3
194
+ prompt-toolkit==3.0.18
195
+ protobuf==3.17.2
196
+ psutil==5.8.0
197
+ ptyprocess==0.7.0
198
+ py-cpuinfo==8.0.0
199
+ py-spy==0.3.8
200
+ py==1.10.0
201
+ pyarrow==6.0.1
202
+ pyasn1-modules==0.2.8
203
+ pyasn1==0.4.8
204
+ pybindgen==0.22.0
205
+ pycocotools==2.0.4
206
+ pycodestyle==2.7.0
207
+ pycparser==2.21
208
+ pyctcdecode==0.3.0
209
+ pydantic==1.8.2
210
+ pydot==1.4.2
211
+ pyflakes==2.3.1
212
+ pygments==2.9.0
213
+ pygtrie==2.4.2
214
+ pymongo==3.12.1
215
+ pynvml==11.0.0
216
+ pyopenssl==21.0.0
217
+ pyparsing==2.4.7
218
+ pyperclip==1.8.2
219
+ pyrsistent==0.18.0
220
+ pysocks==1.7.1
221
+ pytest-forked==1.3.0
222
+ pytest-sugar==0.9.4
223
+ pytest-xdist==2.2.1
224
+ pytest==6.2.4
225
+ python-dateutil==2.8.1
226
+ python-editor==1.0.4
227
+ python-levenshtein==0.12.2
228
+ python-slugify==5.0.2
229
+ pytz==2021.1
230
+ pyyaml==5.4.1
231
+ ray==1.5.2
232
+ recommonmark==0.7.1
233
+ redis==3.5.3
234
+ regex==2021.4.4
235
+ requests-oauthlib==1.3.0
236
+ requests==2.27.1
237
+ resampy==0.2.2
238
+ rfc3986==1.5.0
239
+ rouge-score==0.0.4
240
+ rsa==4.7.2
241
+ ruamel.yaml.clib==0.2.6
242
+ ruamel.yaml==0.17.16
243
+ sacrebleu==1.5.1
244
+ sacremoses==0.0.45
245
+ scann==1.2.4
246
+ scikit-learn==0.24.2
247
+ scipy==1.6.3
248
+ segments==2.2.0
249
+ sentencepiece==0.1.94
250
+ sentry-sdk==1.3.1
251
+ seqio==0.0.6
252
+ setuptools==58.0.4
253
+ shortuuid==1.0.1
254
+ six==1.16.0
255
+ smmap==4.0.0
256
+ snowballstemmer==2.1.0
257
+ sortedcontainers==2.4.0
258
+ soundata==0.1.0
259
+ soundfile==0.10.3.post1
260
+ soupsieve==2.3.1
261
+ sphinx-copybutton==0.3.1
262
+ sphinx-markdown-tables==0.0.15
263
+ sphinx-rtd-theme==0.4.3
264
+ sphinx==3.2.1
265
+ sphinxcontrib-applehelp==1.0.2
266
+ sphinxcontrib-devhelp==1.0.2
267
+ sphinxcontrib-htmlhelp==2.0.0
268
+ sphinxcontrib-jsmath==1.0.1
269
+ sphinxcontrib-qthelp==1.0.3
270
+ sphinxcontrib-serializinghtml==1.1.5
271
+ sphinxext-opengraph==0.4.1
272
+ sqlalchemy==1.4.23
273
+ stevedore==3.3.0
274
+ subprocess32==3.5.4
275
+ tabulate==0.8.9
276
+ tenacity==8.0.1
277
+ tensorboard-data-server==0.6.1
278
+ tensorboard-plugin-wit==1.8.0
279
+ tensorboard==2.6.0
280
+ tensorboardx==2.4
281
+ tensorflow-estimator==2.7.0
282
+ tensorflow-hub==0.12.0
283
+ tensorflow-io-gcs-filesystem==0.23.1
284
+ tensorflow-metadata==1.2.0
285
+ tensorflow-text==2.6.0
286
+ tensorflow==2.7.0
287
+ termcolor==1.1.0
288
+ text-unidecode==1.3
289
+ tfds-nightly==4.4.0.dev202110120106
290
+ threadpoolctl==2.1.0
291
+ timeout-decorator==0.5.0
292
+ timm==0.4.12
293
+ tokenizers==0.10.3
294
+ toml==0.10.2
295
+ toolz==0.11.1
296
+ torch==1.10.1+cu113
297
+ torchaudio==0.10.1+cu113
298
+ torchvision==0.12.0.dev20220120
299
+ tqdm==4.62.3
300
+ traitlets==5.0.5
301
+ transformers==4.12.0.dev0
302
+ triton==1.0.0
303
+ typing-extensions==3.10.0.2
304
+ tzlocal==2.1
305
+ unidic-lite==1.0.8
306
+ unidic==1.0.3
307
+ uritemplate==4.1.1
308
+ urllib3==1.26.7
309
+ wandb==0.12.1
310
+ wasabi==0.8.2
311
+ wcwidth==0.2.5
312
+ werkzeug==1.0.1
313
+ wheel==0.37.0
314
+ wrapt==1.12.1
315
+ xxhash==2.0.2
316
+ yacs==0.1.8
317
+ yarl==1.6.3
318
+ zipp==3.4.1
wandb/run-20220202_223026-1j0459xm/files/wandb-metadata.json ADDED
@@ -0,0 +1,76 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "os": "Linux-5.3.0-64-generic-x86_64-with-glibc2.17",
3
+ "python": "3.8.12",
4
+ "heartbeatAt": "2022-02-02T22:30:27.893012",
5
+ "startedAt": "2022-02-02T22:30:26.118460",
6
+ "docker": null,
7
+ "gpu": "TITAN RTX",
8
+ "gpu_count": 2,
9
+ "cpu_count": 24,
10
+ "cuda": null,
11
+ "args": [
12
+ "--dataset_name=mozilla-foundation/common_voice_8_0",
13
+ "--model_name_or_path=facebook/wav2vec2-xls-r-300m",
14
+ "--dataset_config_name=sv-SE",
15
+ "--output_dir=./",
16
+ "--overwrite_output_dir",
17
+ "--num_train_epochs=50",
18
+ "--per_device_train_batch_size=8",
19
+ "--per_device_eval_batch_size=8",
20
+ "--gradient_accumulation_steps=4",
21
+ "--learning_rate=7.5e-5",
22
+ "--warmup_steps=2000",
23
+ "--length_column_name=input_length",
24
+ "--evaluation_strategy=steps",
25
+ "--text_column_name=sentence",
26
+ "--save_steps=500",
27
+ "--eval_steps=500",
28
+ "--logging_steps=100",
29
+ "--layerdrop=0.0",
30
+ "--activation_dropout=0.1",
31
+ "--save_total_limit=3",
32
+ "--freeze_feature_encoder",
33
+ "--feat_proj_dropout=0.0",
34
+ "--mask_time_prob=0.75",
35
+ "--mask_time_length=10",
36
+ "--mask_feature_prob=0.25",
37
+ "--mask_feature_length=64",
38
+ "--chars_to_ignore",
39
+ ",",
40
+ "?",
41
+ ".",
42
+ "!",
43
+ "-",
44
+ ";",
45
+ ":",
46
+ "\"",
47
+ "\u201c",
48
+ "%",
49
+ "\u2018",
50
+ "\u201d",
51
+ "\ufffd",
52
+ "\u2014",
53
+ "\u2019",
54
+ "\u2026",
55
+ "\u2013",
56
+ "--gradient_checkpointing",
57
+ "--use_auth_token",
58
+ "--fp16",
59
+ "--group_by_length",
60
+ "--do_train",
61
+ "--do_eval",
62
+ "--push_to_hub"
63
+ ],
64
+ "state": "running",
65
+ "program": "run_speech_recognition_ctc.py",
66
+ "codePath": "run_speech_recognition_ctc.py",
67
+ "git": {
68
+ "remote": "https://huggingface.co/patrickvonplaten/xls-r-300m-sv-cv8",
69
+ "commit": "fcede83eb365576c0ac730cd5765d4e6d1c7b3d9"
70
+ },
71
+ "email": "patrick.v.platen@gmail.com",
72
+ "root": "/home/patrick/experiments/xls-r-300m-sv-cv8",
73
+ "host": "brutasse",
74
+ "username": "patrick",
75
+ "executable": "/home/patrick/anaconda3/envs/hugging_face/bin/python"
76
+ }
wandb/run-20220202_223026-1j0459xm/files/wandb-summary.json ADDED
The diff for this file is too large to render. See raw diff
 
wandb/run-20220202_223026-1j0459xm/logs/debug-internal.log ADDED
The diff for this file is too large to render. See raw diff
 
wandb/run-20220202_223026-1j0459xm/logs/debug.log ADDED
@@ -0,0 +1,25 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2022-02-02 22:30:26,119 INFO MainThread:21084 [wandb_setup.py:_flush():69] setting env: {}
2
+ 2022-02-02 22:30:26,119 INFO MainThread:21084 [wandb_setup.py:_flush():69] setting login settings: {}
3
+ 2022-02-02 22:30:26,120 INFO MainThread:21084 [wandb_init.py:_log_setup():342] Logging user logs to /home/patrick/experiments/xls-r-300m-sv-cv8/wandb/run-20220202_223026-1j0459xm/logs/debug.log
4
+ 2022-02-02 22:30:26,120 INFO MainThread:21084 [wandb_init.py:_log_setup():343] Logging internal logs to /home/patrick/experiments/xls-r-300m-sv-cv8/wandb/run-20220202_223026-1j0459xm/logs/debug-internal.log
5
+ 2022-02-02 22:30:26,120 INFO MainThread:21084 [wandb_init.py:init():375] calling init triggers
6
+ 2022-02-02 22:30:26,120 INFO MainThread:21084 [wandb_init.py:init():380] wandb.init called with sweep_config: {}
7
+ config: {}
8
+ 2022-02-02 22:30:26,120 INFO MainThread:21084 [wandb_init.py:init():424] starting backend
9
+ 2022-02-02 22:30:26,120 INFO MainThread:21084 [backend.py:_multiprocessing_setup():70] multiprocessing start_methods=fork,spawn,forkserver, using: spawn
10
+ 2022-02-02 22:30:26,142 INFO MainThread:21084 [backend.py:ensure_launched():135] starting backend process...
11
+ 2022-02-02 22:30:26,161 INFO MainThread:21084 [backend.py:ensure_launched():139] started backend process with pid: 21197
12
+ 2022-02-02 22:30:26,162 INFO MainThread:21084 [wandb_init.py:init():429] backend started and connected
13
+ 2022-02-02 22:30:26,163 INFO MainThread:21084 [wandb_init.py:init():477] updated telemetry
14
+ 2022-02-02 22:30:26,163 INFO MainThread:21084 [wandb_init.py:init():500] communicating current version
15
+ 2022-02-02 22:30:26,620 INFO MainThread:21084 [wandb_init.py:init():505] got version response upgrade_message: "wandb version 0.12.10 is available! To upgrade, please run:\n $ pip install wandb --upgrade"
16
+
17
+ 2022-02-02 22:30:26,620 INFO MainThread:21084 [wandb_init.py:init():513] communicating run to backend with 30 second timeout
18
+ 2022-02-02 22:30:26,733 INFO MainThread:21084 [wandb_init.py:init():540] starting run threads in backend
19
+ 2022-02-02 22:30:31,737 INFO MainThread:21084 [wandb_run.py:_console_start():1601] atexit reg
20
+ 2022-02-02 22:30:31,737 INFO MainThread:21084 [wandb_run.py:_redirect():1475] redirect: SettingsConsole.REDIRECT
21
+ 2022-02-02 22:30:31,737 INFO MainThread:21084 [wandb_run.py:_redirect():1480] Redirecting console.
22
+ 2022-02-02 22:30:31,739 INFO MainThread:21084 [wandb_run.py:_redirect():1536] Redirects installed.
23
+ 2022-02-02 22:30:31,739 INFO MainThread:21084 [wandb_init.py:init():565] run started, returning control to user process
24
+ 2022-02-02 22:30:31,741 INFO MainThread:21084 [wandb_run.py:_config_callback():843] config_cb None None {'return_dict': True, 'output_hidden_states': False, 'output_attentions': False, 'torchscript': False, 'torch_dtype': 'float32', 'use_bfloat16': False, 'pruned_heads': {}, 'tie_word_embeddings': True, 'is_encoder_decoder': False, 'is_decoder': False, 'cross_attention_hidden_size': None, 'add_cross_attention': False, 'tie_encoder_decoder': False, 'max_length': 20, 'min_length': 0, 'do_sample': False, 'early_stopping': False, 'num_beams': 1, 'num_beam_groups': 1, 'diversity_penalty': 0.0, 'temperature': 1.0, 'top_k': 50, 'top_p': 1.0, 'repetition_penalty': 1.0, 'length_penalty': 1.0, 'no_repeat_ngram_size': 0, 'encoder_no_repeat_ngram_size': 0, 'bad_words_ids': None, 'num_return_sequences': 1, 'chunk_size_feed_forward': 0, 'output_scores': False, 'return_dict_in_generate': False, 'forced_bos_token_id': None, 'forced_eos_token_id': None, 'remove_invalid_values': False, 'architectures': ['Wav2Vec2ForPreTraining'], 'finetuning_task': None, 'id2label': {0: 'LABEL_0', 1: 'LABEL_1'}, 'label2id': {'LABEL_0': 0, 'LABEL_1': 1}, 'tokenizer_class': None, 'prefix': None, 'bos_token_id': 1, 'pad_token_id': 34, 'eos_token_id': 2, 'sep_token_id': None, 'decoder_start_token_id': None, 'task_specific_params': None, 'problem_type': None, '_name_or_path': 'facebook/wav2vec2-xls-r-300m', 'transformers_version': '4.17.0.dev0', 'feat_extract_dropout': 0.0, 'model_type': 'wav2vec2', 'num_feat_extract_layers': 7, 'hidden_size': 1024, 'feat_extract_norm': 'layer', 'feat_extract_activation': 'gelu', 'conv_dim': [512, 512, 512, 512, 512, 512, 512], 'conv_stride': [5, 2, 2, 2, 2, 2, 2], 'conv_kernel': [10, 3, 3, 3, 3, 2, 2], 'conv_bias': True, 'num_conv_pos_embeddings': 128, 'num_conv_pos_embedding_groups': 16, 'num_hidden_layers': 24, 'intermediate_size': 4096, 'hidden_act': 'gelu', 'num_attention_heads': 16, 'hidden_dropout': 0.0, 'attention_dropout': 0.0, 'activation_dropout': 0.1, 'feat_proj_dropout': 0.0, 'final_dropout': 0.0, 'layerdrop': 0.0, 'layer_norm_eps': 1e-05, 'initializer_range': 0.02, 'vocab_size': 37, 'do_stable_layer_norm': True, 'use_weighted_layer_sum': False, 'apply_spec_augment': True, 'mask_time_prob': 0.75, 'mask_time_length': 10, 'mask_time_min_masks': 2, 'mask_feature_prob': 0.25, 'mask_feature_length': 64, 'mask_feature_min_masks': 0, 'num_codevectors_per_group': 320, 'num_codevector_groups': 2, 'contrastive_logits_temperature': 0.1, 'feat_quantizer_dropout': 0.0, 'num_negatives': 100, 'codevector_dim': 768, 'proj_codevector_dim': 768, 'diversity_loss_weight': 0.1, 'ctc_loss_reduction': 'mean', 'ctc_zero_infinity': False, 'add_adapter': False, 'adapter_kernel_size': 3, 'adapter_stride': 2, 'num_adapter_layers': 3, 'output_hidden_size': 1024, 'classifier_proj_size': 256, 'tdnn_dim': [512, 512, 512, 512, 1500], 'tdnn_kernel': [5, 3, 3, 1, 1], 'tdnn_dilation': [1, 2, 3, 1, 1], 'xvector_output_dim': 512, 'output_dir': './', 'overwrite_output_dir': True, 'do_train': True, 'do_eval': True, 'do_predict': False, 'evaluation_strategy': 'steps', 'prediction_loss_only': False, 'per_device_train_batch_size': 8, 'per_device_eval_batch_size': 8, 'per_gpu_train_batch_size': 'None', 'per_gpu_eval_batch_size': 'None', 'gradient_accumulation_steps': 4, 'eval_accumulation_steps': 'None', 'learning_rate': 7.5e-05, 'weight_decay': 0.0, 'adam_beta1': 0.9, 'adam_beta2': 0.999, 'adam_epsilon': 1e-08, 'max_grad_norm': 1.0, 'num_train_epochs': 50.0, 'max_steps': -1, 'lr_scheduler_type': 'linear', 'warmup_ratio': 0.0, 'warmup_steps': 2000, 'log_level': -1, 'log_level_replica': -1, 'log_on_each_node': True, 'logging_dir': './runs/Feb02_22-30-13_brutasse', 'logging_strategy': 'steps', 'logging_first_step': False, 'logging_steps': 100, 'logging_nan_inf_filter': True, 'save_strategy': 'steps', 'save_steps': 500, 'save_total_limit': 3, 'save_on_each_node': False, 'no_cuda': False, 'seed': 42, 'bf16': False, 'fp16': True, 'fp16_opt_level': 'O1', 'half_precision_backend': 'amp', 'bf16_full_eval': False, 'fp16_full_eval': False, 'tf32': 'None', 'local_rank': -1, 'xpu_backend': 'None', 'tpu_num_cores': 'None', 'tpu_metrics_debug': False, 'debug': '[]', 'dataloader_drop_last': False, 'eval_steps': 500, 'dataloader_num_workers': 0, 'past_index': -1, 'run_name': './', 'disable_tqdm': False, 'remove_unused_columns': True, 'label_names': 'None', 'load_best_model_at_end': False, 'metric_for_best_model': 'None', 'greater_is_better': 'None', 'ignore_data_skip': False, 'sharded_ddp': '[]', 'deepspeed': 'None', 'label_smoothing_factor': 0.0, 'optim': 'adamw_hf', 'adafactor': False, 'group_by_length': True, 'length_column_name': 'input_length', 'report_to': "['tensorboard', 'wandb', 'codecarbon']", 'ddp_find_unused_parameters': 'None', 'ddp_bucket_cap_mb': 'None', 'dataloader_pin_memory': True, 'skip_memory_metrics': True, 'use_legacy_prediction_loop': False, 'push_to_hub': True, 'resume_from_checkpoint': 'None', 'hub_model_id': 'None', 'hub_strategy': 'every_save', 'hub_token': '<HUB_TOKEN>', 'gradient_checkpointing': True, 'fp16_backend': 'auto', 'push_to_hub_model_id': 'None', 'push_to_hub_organization': 'None', 'push_to_hub_token': '<PUSH_TO_HUB_TOKEN>', '_n_gpu': 1, 'mp_parameters': '', 'train_batch_size': 8, 'eval_batch_size': 8}
25
+ 2022-02-02 22:30:31,745 INFO MainThread:21084 [wandb_watch.py:watch():43] Watching
wandb/run-20220202_223026-1j0459xm/run-1j0459xm.wandb ADDED
Binary file (3.39 MB). View file