Training in progress, step 500

.gitignore

@@ -0,0 +1 @@
+checkpoint-*/

added_tokens.json

@@ -0,0 +1 @@
+{"<s>": 32, "</s>": 33}

config.json

@@ -0,0 +1,107 @@
+{
+  "_name_or_path": "facebook/wav2vec2-xls-r-1b",
+  "activation_dropout": 0.055,
+  "adapter_kernel_size": 3,
+  "adapter_stride": 2,
+  "add_adapter": false,
+  "apply_spec_augment": true,
+  "architectures": [
+    "Wav2Vec2ForCTC"
+  ],
+  "attention_dropout": 0.094,
+  "bos_token_id": 1,
+  "classifier_proj_size": 256,
+  "codevector_dim": 1024,
+  "contrastive_logits_temperature": 0.1,
+  "conv_bias": true,
+  "conv_dim": [
+    512,
+    512,
+    512,
+    512,
+    512,
+    512,
+    512
+  ],
+  "conv_kernel": [
+    10,
+    3,
+    3,
+    3,
+    3,
+    2,
+    2
+  ],
+  "conv_stride": [
+    5,
+    2,
+    2,
+    2,
+    2,
+    2,
+    2
+  ],
+  "ctc_loss_reduction": "mean",
+  "ctc_zero_infinity": false,
+  "diversity_loss_weight": 0.1,
+  "do_stable_layer_norm": true,
+  "eos_token_id": 2,
+  "feat_extract_activation": "gelu",
+  "feat_extract_dropout": 0.0,
+  "feat_extract_norm": "layer",
+  "feat_proj_dropout": 0.04,
+  "feat_quantizer_dropout": 0.0,
+  "final_dropout": 0.0,
+  "hidden_act": "gelu",
+  "hidden_dropout": 0.047,
+  "hidden_size": 1280,
+  "initializer_range": 0.02,
+  "intermediate_size": 5120,
+  "layer_norm_eps": 1e-05,
+  "layerdrop": 0.041,
+  "mask_feature_length": 64,
+  "mask_feature_min_masks": 0,
+  "mask_feature_prob": 0.25,
+  "mask_time_length": 10,
+  "mask_time_min_masks": 2,
+  "mask_time_prob": 0.082,
+  "model_type": "wav2vec2",
+  "num_adapter_layers": 3,
+  "num_attention_heads": 16,
+  "num_codevector_groups": 2,
+  "num_codevectors_per_group": 320,
+  "num_conv_pos_embedding_groups": 16,
+  "num_conv_pos_embeddings": 128,
+  "num_feat_extract_layers": 7,
+  "num_hidden_layers": 48,
+  "num_negatives": 100,
+  "output_hidden_size": 1280,
+  "pad_token_id": 31,
+  "proj_codevector_dim": 1024,
+  "tdnn_dilation": [
+    1,
+    2,
+    3,
+    1,
+    1
+  ],
+  "tdnn_dim": [
+    512,
+    512,
+    512,
+    512,
+    1500
+  ],
+  "tdnn_kernel": [
+    5,
+    3,
+    3,
+    1,
+    1
+  ],
+  "torch_dtype": "float32",
+  "transformers_version": "4.17.0.dev0",
+  "use_weighted_layer_sum": false,
+  "vocab_size": 34,
+  "xvector_output_dim": 512
+}

preprocessor_config.json

@@ -0,0 +1,9 @@
+{
+  "do_normalize": true,
+  "feature_extractor_type": "Wav2Vec2FeatureExtractor",
+  "feature_size": 1,
+  "padding_side": "right",
+  "padding_value": 0,
+  "return_attention_mask": true,
+  "sampling_rate": 16000
+}

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b3a50cfef1996549c158e0a99768454dc69a4c7f4bdf1f625c1d1fee88064e7c
+size 3850486961

runs/Feb01_15-07-52_dante/1643724498.3105586/events.out.tfevents.1643724498.dante.3248238.1

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:86820d076458d9b054fbff45756b316d5b7eed7f41f8d4d96de226101043c1e4
+size 4753

runs/Feb01_15-07-52_dante/events.out.tfevents.1643724498.dante.3248238.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a1a282dd4c21da84f125f45c2113d11dabd21b9ce46c7658c6d2fadfd52890ca
+size 4699

runs/Feb01_15-29-26_dante/1643725823.3776064/events.out.tfevents.1643725823.dante.3265625.1

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:47e69b92273023789a04af1852f35ecddd9313ebeea762742a29c8bdd9a7a4aa
+size 4753

runs/Feb01_15-29-26_dante/events.out.tfevents.1643725823.dante.3265625.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:250ff7f76ef38c789f74a69c451d5fe3ac0c6318834efe8345cae5c33636f357
+size 5799

special_tokens_map.json

@@ -0,0 +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}, {"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

@@ -0,0 +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

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f7d438783868ddc79586702c9970b9881ef3452e2773d6c176fd67f57a76ec00
+size 3055

vocab.json

@@ -0,0 +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, "å": 27, "æ": 28, "ø": 29, "|": 0, "[UNK]": 30, "[PAD]": 31}

wandb/debug-internal.log

@@ -0,0 +1 @@
+run-20220201_153024-1w85vsuu/logs/debug-internal.log

wandb/debug.log

@@ -0,0 +1 @@
+run-20220201_153024-1w85vsuu/logs/debug.log

wandb/latest-run

@@ -0,0 +1 @@
+run-20220201_153024-1w85vsuu

wandb/run-20220201_150818-33gv5m8t/files/code/run_speech_recognition_ctc.py

@@ -0,0 +1,792 @@
+#!/usr/bin/env python
+# coding=utf-8
+# Copyright 2021 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+
+""" Fine-tuning a 🤗 Transformers CTC model for automatic speech recognition"""
+
+import functools
+import json
+import logging
+import os
+import re
+import sys
+import warnings
+from dataclasses import dataclass, field
+from typing import Dict, List, Optional, Union
+
+import datasets
+import numpy as np
+import torch
+from datasets import DatasetDict, load_dataset, load_metric
+
+import transformers
+from transformers import (
+    AutoConfig,
+    AutoFeatureExtractor,
+    AutoModelForCTC,
+    AutoProcessor,
+    AutoTokenizer,
+    HfArgumentParser,
+    Trainer,
+    TrainingArguments,
+    Wav2Vec2Processor,
+    set_seed,
+)
+from transformers.trainer_utils import get_last_checkpoint, is_main_process
+from transformers.utils import check_min_version
+from transformers.utils.versions import require_version
+
+
+# Will error if the minimal version of Transformers is not installed. Remove at your own risks.
+check_min_version("4.16.0.dev0")
+
+require_version("datasets>=1.13.3", "To fix: pip install -r examples/pytorch/text-classification/requirements.txt")
+
+
+logger = logging.getLogger(__name__)
+
+
+def list_field(default=None, metadata=None):
+    return field(default_factory=lambda: default, metadata=metadata)
+
+
+@dataclass
+class ModelArguments:
+    """
+    Arguments pertaining to which model/config/tokenizer we are going to fine-tune from.
+    """
+
+    model_name_or_path: str = field(
+        metadata={"help": "Path to pretrained model or model identifier from huggingface.co/models"}
+    )
+    tokenizer_name_or_path: Optional[str] = field(
+        default=None,
+        metadata={"help": "Path to pretrained tokenizer or tokenizer identifier from huggingface.co/models"},
+    )
+    cache_dir: Optional[str] = field(
+        default=None,
+        metadata={"help": "Where do you want to store the pretrained models downloaded from huggingface.co"},
+    )
+    freeze_feature_encoder: bool = field(
+        default=True, metadata={"help": "Whether to freeze the feature encoder layers of the model."}
+    )
+    attention_dropout: float = field(
+        default=0.0, metadata={"help": "The dropout ratio for the attention probabilities."}
+    )
+    activation_dropout: float = field(
+        default=0.0, metadata={"help": "The dropout ratio for activations inside the fully connected layer."}
+    )
+    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: float = field(
+        default=0.0,
+        metadata={"help": "The dropout probability for the final projection layer."},
+    )
+    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: int = field(
+        default=10,
+        metadata={"help": "Length of vector span to mask along the time axis."},
+    )
+    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: int = field(
+        default=10,
+        metadata={"help": "Length of vector span to mask along the feature axis."},
+    )
+    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'."}
+    )
+    ctc_zero_infinity: Optional[bool] = field(
+        default=False, metadata={"help": "If True, will try yo aboud the CTC loss goinf to infinity."}
+    )
+
+@dataclass
+class DataTrainingArguments:
+    """
+    Arguments pertaining to what data we are going to input our model for training and eval.
+
+    Using `HfArgumentParser` we can turn this class
+    into argparse arguments to be able to specify them on
+    the command line.
+    """
+
+    dataset_name: str = field(
+        metadata={"help": "The configuration name of the dataset to use (via the datasets library)."}
+    )
+    dataset_config_name: str = field(
+        default=None, metadata={"help": "The configuration name of the dataset to use (via the datasets library)."}
+    )
+    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: 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: str = field(
+        default="audio",
+        metadata={"help": "The name of the dataset column containing the audio data. Defaults to 'audio'"},
+    )
+    text_column_name: str = field(
+        default="text",
+        metadata={"help": "The name of the dataset column containing the text data. Defaults to 'text'"},
+    )
+    overwrite_cache: bool = field(
+        default=False, metadata={"help": "Overwrite the cached preprocessed datasets or not."}
+    )
+    preprocessing_num_workers: Optional[int] = field(
+        default=None,
+        metadata={"help": "The number of processes to use for the preprocessing."},
+    )
+    max_train_samples: Optional[int] = field(
+        default=None,
+        metadata={
+            "help": "For debugging purposes or quicker training, truncate the number of training examples to this "
+            "value if set."
+        },
+    )
+    max_eval_samples: Optional[int] = field(
+        default=None,
+        metadata={
+            "help": "For debugging purposes or quicker training, truncate the number of validation examples to this "
+            "value if set."
+        },
+    )
+    chars_to_ignore: Optional[List[str]] = list_field(
+        default=None,
+        metadata={"help": "A list of characters to remove from the transcripts."},
+    )
+    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: 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: float = field(
+        default=0.0, metadata={"help": "Filter audio files that are shorter than `min_duration_in_seconds` seconds"}
+    )
+    preprocessing_only: bool = field(
+        default=False,
+        metadata={
+            "help": "Whether to only do data preprocessing and skip training. "
+            "This is especially useful when data preprocessing errors out in distributed training due to timeout. "
+            "In this case, one should run the preprocessing in a non-distributed setup with `preprocessing_only=True` "
+            "so that the cached datasets can consequently be loaded in distributed training"
+        },
+    )
+    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: str = field(
+        default="[UNK]",
+        metadata={"help": "The unk token for the tokenizer"},
+    )
+    pad_token: str = field(
+        default="[PAD]",
+        metadata={"help": "The padding token for the tokenizer"},
+    )
+    word_delimiter_token: str = field(
+        default="|",
+        metadata={"help": "The word delimiter token for the tokenizer"},
+    )
+    phoneme_language: Optional[str] = field(
+        default=None,
+        metadata={
+            "help": "The target language that should be used be"
+            " passed to the tokenizer for tokenization. Note that"
+            " this is only relevant if the model classifies the"
+            " input audio to a sequence of phoneme sequences."
+        },
+    )
+
+
+@dataclass
+class DataCollatorCTCWithPadding:
+    """
+    Data collator that will dynamically pad the inputs received.
+    Args:
+        processor (:class:`~transformers.AutoProcessor`)
+            The processor used for proccessing the data.
+        padding (:obj:`bool`, :obj:`str` or :class:`~transformers.tokenization_utils_base.PaddingStrategy`, `optional`, defaults to :obj:`True`):
+            Select a strategy to pad the returned sequences (according to the model's padding side and padding index)
+            among:
+            * :obj:`True` or :obj:`'longest'`: Pad to the longest sequence in the batch (or no padding if only a single
+              sequence if provided).
+            * :obj:`'max_length'`: Pad to a maximum length specified with the argument :obj:`max_length` or to the
+              maximum acceptable input length for the model if that argument is not provided.
+            * :obj:`False` or :obj:`'do_not_pad'` (default): No padding (i.e., can output a batch with sequences of
+              different lengths).
+        max_length (:obj:`int`, `optional`):
+            Maximum length of the ``input_values`` of the returned list and optionally padding length (see above).
+        max_length_labels (:obj:`int`, `optional`):
+            Maximum length of the ``labels`` returned list and optionally padding length (see above).
+        pad_to_multiple_of (:obj:`int`, `optional`):
+            If set will pad the sequence to a multiple of the provided value.
+            This is especially useful to enable the use of Tensor Cores on NVIDIA hardware with compute capability >=
+            7.5 (Volta).
+    """
+
+    processor: AutoProcessor
+    padding: Union[bool, str] = "longest"
+    pad_to_multiple_of: Optional[int] = None
+    pad_to_multiple_of_labels: Optional[int] = None
+
+    def __call__(self, features: List[Dict[str, Union[List[int], torch.Tensor]]]) -> Dict[str, torch.Tensor]:
+        # split inputs and labels since they have to be of different lenghts and need
+        # different padding methods
+        input_features = [{"input_values": feature["input_values"]} for feature in features]
+        label_features = [{"input_ids": feature["labels"]} for feature in features]
+
+        batch = self.processor.pad(
+            input_features,
+            padding=self.padding,
+            pad_to_multiple_of=self.pad_to_multiple_of,
+            return_tensors="pt",
+        )
+
+        with self.processor.as_target_processor():
+            labels_batch = self.processor.pad(
+                label_features,
+                padding=self.padding,
+                pad_to_multiple_of=self.pad_to_multiple_of_labels,
+                return_tensors="pt",
+            )
+
+        # replace padding with -100 to ignore loss correctly
+        labels = labels_batch["input_ids"].masked_fill(labels_batch.attention_mask.ne(1), -100)
+
+        batch["labels"] = labels
+
+        return batch
+
+
+def create_vocabulary_from_data(
+    datasets: DatasetDict,
+    word_delimiter_token: Optional[str] = None,
+    unk_token: Optional[str] = None,
+    pad_token: Optional[str] = None,
+):
+    # Given training and test labels create vocabulary
+    def extract_all_chars(batch):
+        all_text = " ".join(batch["target_text"])
+        vocab = list(set(all_text))
+        return {"vocab": [vocab], "all_text": [all_text]}
+
+    vocabs = datasets.map(
+        extract_all_chars,
+        batched=True,
+        batch_size=-1,
+        keep_in_memory=True,
+        remove_columns=datasets["train"].column_names,
+    )
+
+    # take union of all unique characters in each dataset
+    vocab_set = functools.reduce(
+        lambda vocab_1, vocab_2: set(vocab_1["vocab"][0]) | set(vocab_2["vocab"][0]), vocabs.values()
+    )
+
+    vocab_dict = {v: k for k, v in enumerate(sorted(list(vocab_set)))}
+
+    # replace white space with delimiter token
+    if word_delimiter_token is not None:
+        vocab_dict[word_delimiter_token] = vocab_dict[" "]
+        del vocab_dict[" "]
+
+    # add unk and pad token
+    if unk_token is not None:
+        vocab_dict[unk_token] = len(vocab_dict)
+
+    if pad_token is not None:
+        vocab_dict[pad_token] = len(vocab_dict)
+
+    return vocab_dict
+
+
+def main():
+    # See all possible arguments in src/transformers/training_args.py
+    # or by passing the --help flag to this script.
+    # We now keep distinct sets of args, for a cleaner separation of concerns.
+
+    parser = HfArgumentParser((ModelArguments, DataTrainingArguments, TrainingArguments))
+    if len(sys.argv) == 2 and sys.argv[1].endswith(".json"):
+        # If we pass only one argument to the script and it's the path to a json file,
+        # let's parse it to get our arguments.
+        model_args, data_args, training_args = parser.parse_json_file(json_file=os.path.abspath(sys.argv[1]))
+    else:
+        model_args, data_args, training_args = parser.parse_args_into_dataclasses()
+
+    # Detecting last checkpoint.
+    last_checkpoint = None
+    if os.path.isdir(training_args.output_dir) and training_args.do_train and not training_args.overwrite_output_dir:
+        last_checkpoint = get_last_checkpoint(training_args.output_dir)
+        if last_checkpoint is None and len(os.listdir(training_args.output_dir)) > 0:
+            raise ValueError(
+                f"Output directory ({training_args.output_dir}) already exists and is not empty. "
+                "Use --overwrite_output_dir to overcome."
+            )
+        elif last_checkpoint is not None:
+            logger.info(
+                f"Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change "
+                "the `--output_dir` or add `--overwrite_output_dir` to train from scratch."
+            )
+
+    # Setup logging
+    logging.basicConfig(
+        format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
+        datefmt="%m/%d/%Y %H:%M:%S",
+        handlers=[logging.StreamHandler(sys.stdout)],
+    )
+    logger.setLevel(logging.INFO if is_main_process(training_args.local_rank) else logging.WARN)
+
+    # Log on each process the small summary:
+    logger.warning(
+        f"Process rank: {training_args.local_rank}, device: {training_args.device}, n_gpu: {training_args.n_gpu}"
+        f"distributed training: {bool(training_args.local_rank != -1)}, 16-bits training: {training_args.fp16}"
+    )
+    # Set the verbosity to info of the Transformers logger (on main process only):
+    if is_main_process(training_args.local_rank):
+        transformers.utils.logging.set_verbosity_info()
+    logger.info("Training/evaluation parameters %s", training_args)
+
+    # Set seed before initializing model.
+    set_seed(training_args.seed)
+
+    # Pre-processing dataset
+    import re
+    def filter_numeric(entry):
+        return  (
+            "0" not in entry["text"]
+            and "1" not in entry["text"]
+            and "2" not in entry["text"]
+            and "3" not in entry["text"]
+            and "4" not in entry["text"]
+            and "5" not in entry["text"]
+            and "6" not in entry["text"]
+            and "7" not in entry["text"]
+            and "8" not in entry["text"]
+            and "9" not in entry["text"]
+        )
+
+    def filter_inaudible(entry):
+        return not re.search("\d|<inaudible>", entry["text"], flags=re.IGNORECASE)
+
+    def filter_nynorsk(entry):
+        return re.search("nb-no", entry["sentence_language_code"], flags=re.IGNORECASE)
+
+    def filter_tooshort(entry):
+        #print(f"The audio sample ({entry["audio"]["path"]}) is too small, and has been omitted. "
+        return (len(entry["text"]) <= len(entry["audio"]["array"]) // 320) and (len(entry["text"].strip()) >= 3)
+
+    def map_dataset(entry):
+        batch = {"text": entry["text"].lower()}
+        batch["text"] = re.sub('[áàâ]', 'a', batch["text"])
+        batch["text"] = re.sub('[ä]', 'æ', batch["text"])
+        batch["text"] = re.sub('[éèëê]', 'e', batch["text"])
+        batch["text"] = re.sub('[íìïî]', 'i', batch["text"])
+        batch["text"] = re.sub('[óòöô]', 'o', batch["text"])
+        batch["text"] = re.sub('[ö]', 'ø', batch["text"])
+        batch["text"] = re.sub('[ç]', 'c', batch["text"])
+        batch["text"] = re.sub('[úùüû]', 'u', batch["text"])
+        batch["text"] = re.sub('\s', ' ', batch["text"])
+        batch["text"] = re.sub('<ee>', 'eee', batch["text"])
+        batch["text"] = re.sub('<qq>', 'qqq', batch["text"])
+        batch["text"] = re.sub('<mm>', 'mmm', batch["text"])
+        # batch["text"] = re.sub('<inaudible>', '?', batch["text"])
+        if "<" in batch["text"]:
+            raise ValueError(batch["text"])
+        return batch
+
+    # 1. First, let's load the dataset
+    raw_datasets = DatasetDict()
+
+    if training_args.do_train:
+        raw_datasets["train"] = load_dataset(
+            data_args.dataset_name,
+            data_args.dataset_config_name,
+            split=data_args.train_split_name,
+            use_auth_token=data_args.use_auth_token,
+        )
+        raw_datasets["train"] = raw_datasets["train"].filter(filter_numeric).filter(filter_inaudible).filter(filter_nynorsk).filter(filter_tooshort)
+        raw_datasets["train"] = raw_datasets["train"].map(map_dataset)
+
+        if data_args.audio_column_name not in raw_datasets["train"].column_names:
+            raise ValueError(
+                f"--audio_column_name '{data_args.audio_column_name}' not found in dataset '{data_args.dataset_name}'. "
+                "Make sure to set `--audio_column_name` to the correct audio column - one of "
+                f"{', '.join(raw_datasets['train'].column_names)}."
+            )
+
+        if data_args.text_column_name not in raw_datasets["train"].column_names:
+            raise ValueError(
+                f"--text_column_name {data_args.text_column_name} not found in dataset '{data_args.dataset_name}'. "
+                "Make sure to set `--text_column_name` to the correct text column - one of "
+                f"{', '.join(raw_datasets['train'].column_names)}."
+            )
+
+        if data_args.max_train_samples is not None:
+            raw_datasets["train"] = raw_datasets["train"].select(range(data_args.max_train_samples))
+
+    if training_args.do_eval:
+        raw_datasets["eval"] = load_dataset(
+            data_args.dataset_name,
+            data_args.dataset_config_name,
+            split=data_args.eval_split_name,
+            use_auth_token=data_args.use_auth_token,
+        )
+        raw_datasets["eval"] = raw_datasets["eval"].filter(filter_numeric).filter(filter_inaudible).filter(filter_nynorsk).filter(filter_tooshort)
+        raw_datasets["eval"] = raw_datasets["eval"].map(map_dataset)
+
+        if data_args.max_eval_samples is not None:
+            raw_datasets["eval"] = raw_datasets["eval"].select(range(data_args.max_eval_samples))
+
+
+    # 2. We remove some special characters from the datasets
+    # that make training complicated and do not help in transcribing the speech
+    # E.g. characters, such as `,` and `.` do not really have an acoustic characteristic
+    # that could be easily picked up by the model
+    #chars_to_ignore_regex = (
+    #    f'[{"".join(data_args.chars_to_ignore)}]' if data_args.chars_to_ignore is not None else None
+    #)
+    chars_to_ignore_regex = '[\,\?\.\!\-\;\:\"\“\%\‘\”\�\'\–\_\\\+\#\/]'
+    
+    text_column_name = data_args.text_column_name
+
+    def remove_special_characters(batch):
+        if chars_to_ignore_regex is not None:
+            batch["target_text"] = re.sub(chars_to_ignore_regex, "", batch[text_column_name]).lower() + " "
+        else:
+            batch["target_text"] = batch[text_column_name].lower() + " "
+        return batch
+
+    with training_args.main_process_first(desc="dataset map special characters removal"):
+        raw_datasets = raw_datasets.map(
+            remove_special_characters,
+            remove_columns=[text_column_name],
+            desc="remove special characters from datasets",
+        )
+
+    # save special tokens for tokenizer
+    word_delimiter_token = data_args.word_delimiter_token
+    unk_token = data_args.unk_token
+    pad_token = data_args.pad_token
+
+    # 3. Next, let's load the config as we might need it to create
+    # the tokenizer
+    # load config
+    config = AutoConfig.from_pretrained(
+        model_args.model_name_or_path, cache_dir=model_args.cache_dir, use_auth_token=data_args.use_auth_token
+    )
+
+    # 4. Next, if no tokenizer file is defined,
+    # we create the vocabulary of the model by extracting all unique characters from
+    # the training and evaluation datasets
+    # We need to make sure that only first rank saves vocabulary
+    # make sure all processes wait until vocab is created
+    tokenizer_name_or_path = model_args.tokenizer_name_or_path
+    tokenizer_kwargs = {}
+    if tokenizer_name_or_path is None:
+        # save vocab in training output dir
+        tokenizer_name_or_path = training_args.output_dir
+
+        vocab_file = os.path.join(tokenizer_name_or_path, "vocab.json")
+
+        with training_args.main_process_first():
+            if training_args.overwrite_output_dir and os.path.isfile(vocab_file):
+                os.remove(vocab_file)
+
+        with training_args.main_process_first(desc="dataset map vocabulary creation"):
+            if not os.path.isfile(vocab_file):
+                os.makedirs(tokenizer_name_or_path, exist_ok=True)
+                vocab_dict = create_vocabulary_from_data(
+                    raw_datasets,
+                    word_delimiter_token=word_delimiter_token,
+                    unk_token=unk_token,
+                    pad_token=pad_token,
+                )
+
+                # save vocab dict to be loaded into tokenizer
+                with open(vocab_file, "w") as file:
+                    json.dump(vocab_dict, file)
+
+        # if tokenizer has just been created
+        # it is defined by `tokenizer_class` if present in config else by `model_type`
+        tokenizer_kwargs = {
+            "config": config if config.tokenizer_class is not None else None,
+            "tokenizer_type": config.model_type if config.tokenizer_class is None else None,
+            "unk_token": unk_token,
+            "pad_token": pad_token,
+            "word_delimiter_token": word_delimiter_token,
+        }
+
+    # 5. Now we can instantiate the feature extractor, tokenizer and model
+    # Note for distributed training, the .from_pretrained methods guarantee that only
+    # one local process can concurrently download model & vocab.
+
+    # load feature_extractor and tokenizer
+    tokenizer = AutoTokenizer.from_pretrained(
+        tokenizer_name_or_path,
+        use_auth_token=data_args.use_auth_token,
+        **tokenizer_kwargs,
+    )
+    feature_extractor = AutoFeatureExtractor.from_pretrained(
+        model_args.model_name_or_path, cache_dir=model_args.cache_dir, use_auth_token=data_args.use_auth_token
+    )
+
+    # adapt config
+    config.update(
+        {
+            "feat_proj_dropout": model_args.feat_proj_dropout,
+            "attention_dropout": model_args.attention_dropout,
+            "hidden_dropout": model_args.hidden_dropout,
+            "final_dropout": model_args.final_dropout,
+            "mask_time_prob": model_args.mask_time_prob,
+            "mask_time_length": model_args.mask_time_length,
+            "mask_feature_prob": model_args.mask_feature_prob,
+            "mask_feature_length": model_args.mask_feature_length,
+            "gradient_checkpointing": training_args.gradient_checkpointing,
+            "layerdrop": model_args.layerdrop,
+            "ctc_loss_reduction": model_args.ctc_loss_reduction,
+            "ctc_zero_infinity": model_args.ctc_zero_infinity,
+            "pad_token_id": tokenizer.pad_token_id,
+            "vocab_size": len(tokenizer),
+            "activation_dropout": model_args.activation_dropout,
+        }
+    )
+
+    # create model
+    model = AutoModelForCTC.from_pretrained(
+        model_args.model_name_or_path,
+        cache_dir=model_args.cache_dir,
+        config=config,
+        use_auth_token=data_args.use_auth_token,
+    )
+
+    # freeze encoder
+    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,
+    # so that we just need to set the correct target sampling rate and normalize the input
+    # via the `feature_extractor`
+
+    # make sure that dataset decodes audio with correct sampling rate
+    dataset_sampling_rate = next(iter(raw_datasets.values())).features[data_args.audio_column_name].sampling_rate
+    if dataset_sampling_rate != feature_extractor.sampling_rate:
+        raw_datasets = raw_datasets.cast_column(
+            data_args.audio_column_name, datasets.features.Audio(sampling_rate=feature_extractor.sampling_rate)
+        )
+
+    # derive max & min input length for sample rate & max duration
+    max_input_length = data_args.max_duration_in_seconds * feature_extractor.sampling_rate
+    min_input_length = data_args.min_duration_in_seconds * feature_extractor.sampling_rate
+    audio_column_name = data_args.audio_column_name
+    num_workers = data_args.preprocessing_num_workers
+
+    # `phoneme_language` is only relevant if the model is fine-tuned on phoneme classification
+    phoneme_language = data_args.phoneme_language
+
+    # Preprocessing the datasets.
+    # We need to read the audio files as arrays and tokenize the targets.
+    def prepare_dataset(batch):
+        # load audio
+        sample = batch[audio_column_name]
+
+        inputs = feature_extractor(sample["array"], sampling_rate=sample["sampling_rate"])
+        batch["input_values"] = inputs.input_values[0]
+        batch["input_length"] = len(batch["input_values"])
+
+        # encode targets
+        additional_kwargs = {}
+        if phoneme_language is not None:
+            additional_kwargs["phonemizer_lang"] = phoneme_language
+
+        batch["labels"] = tokenizer(batch["target_text"], **additional_kwargs).input_ids
+        return batch
+
+    with training_args.main_process_first(desc="dataset map preprocessing"):
+        vectorized_datasets = raw_datasets.map(
+            prepare_dataset,
+            remove_columns=next(iter(raw_datasets.values())).column_names,
+            num_proc=num_workers,
+            desc="preprocess datasets",
+        )
+
+        def is_audio_in_length_range(length):
+            return length > min_input_length and length < max_input_length
+
+        # filter data that is shorter than min_input_length
+        vectorized_datasets = vectorized_datasets.filter(
+            is_audio_in_length_range,
+            num_proc=num_workers,
+            input_columns=["input_length"],
+        )
+
+    # 7. Next, we can prepare the training.
+    # Let's use word error rate (WER) as our evaluation metric,
+    # instantiate a data collator and the trainer
+
+    # Define evaluation metrics during training, *i.e.* word error rate, character error rate
+    eval_metrics = {metric: load_metric(metric) for metric in data_args.eval_metrics}
+
+    # for large datasets it is advised to run the preprocessing on a
+    # single machine first with ``args.preprocessing_only`` since there will mostly likely
+    # be a timeout when running the script in distributed mode.
+    # In a second step ``args.preprocessing_only`` can then be set to `False` to load the
+    # cached dataset
+    if data_args.preprocessing_only:
+        logger.info(f"Data preprocessing finished. Files cached at {vectorized_datasets.cache_files}")
+        return
+
+    def compute_metrics(pred):
+        pred_logits = pred.predictions
+        pred_ids = np.argmax(pred_logits, axis=-1)
+
+        pred.label_ids[pred.label_ids == -100] = tokenizer.pad_token_id
+
+        pred_str = tokenizer.batch_decode(pred_ids)
+        # we do not want to group tokens when computing the metrics
+        label_str = tokenizer.batch_decode(pred.label_ids, group_tokens=False)
+
+        metrics = {k: v.compute(predictions=pred_str, references=label_str) for k, v in eval_metrics.items()}
+
+        return metrics
+
+    # Now save everything to be able to create a single processor later
+    if is_main_process(training_args.local_rank):
+        # save feature extractor, tokenizer and config
+        feature_extractor.save_pretrained(training_args.output_dir)
+        tokenizer.save_pretrained(training_args.output_dir)
+        config.save_pretrained(training_args.output_dir)
+
+    try:
+        processor = AutoProcessor.from_pretrained(training_args.output_dir)
+    except (OSError, KeyError):
+        warnings.warn(
+            "Loading a processor from a feature extractor config that does not"
+            " include a `processor_class` attribute is deprecated and will be removed in v5. Please add the following "
+            " attribute to your `preprocessor_config.json` file to suppress this warning: "
+            " `'processor_class': 'Wav2Vec2Processor'`",
+            FutureWarning,
+        )
+        processor = Wav2Vec2Processor.from_pretrained(training_args.output_dir)
+
+    # Instantiate custom data collator
+    data_collator = DataCollatorCTCWithPadding(processor=processor)
+
+    # Initialize Trainer
+    trainer = Trainer(
+        model=model,
+        data_collator=data_collator,
+        args=training_args,
+        compute_metrics=compute_metrics,
+        train_dataset=vectorized_datasets["train"] if training_args.do_train else None,
+        eval_dataset=vectorized_datasets["eval"] if training_args.do_eval else None,
+        tokenizer=feature_extractor,
+    )
+
+    # 8. Finally, we can start training
+
+    # Training
+    if training_args.do_train:
+
+        # use last checkpoint if exist
+        if last_checkpoint is not None:
+            checkpoint = last_checkpoint
+        elif os.path.isdir(model_args.model_name_or_path):
+            checkpoint = model_args.model_name_or_path
+        else:
+            checkpoint = None
+
+        train_result = trainer.train(resume_from_checkpoint=checkpoint)
+        trainer.save_model()
+
+        metrics = train_result.metrics
+        max_train_samples = (
+            data_args.max_train_samples
+            if data_args.max_train_samples is not None
+            else len(vectorized_datasets["train"])
+        )
+        metrics["train_samples"] = min(max_train_samples, len(vectorized_datasets["train"]))
+
+        trainer.log_metrics("train", metrics)
+        trainer.save_metrics("train", metrics)
+        trainer.save_state()
+
+    # Evaluation
+    results = {}
+    if training_args.do_eval:
+        logger.info("*** Evaluate ***")
+        metrics = trainer.evaluate()
+        max_eval_samples = (
+            data_args.max_eval_samples if data_args.max_eval_samples is not None else len(vectorized_datasets["eval"])
+        )
+        metrics["eval_samples"] = min(max_eval_samples, len(vectorized_datasets["eval"]))
+
+        trainer.log_metrics("eval", metrics)
+        trainer.save_metrics("eval", metrics)
+
+    # Write model card and (optionally) push to hub
+    config_name = data_args.dataset_config_name if data_args.dataset_config_name is not None else "na"
+    kwargs = {
+        "finetuned_from": model_args.model_name_or_path,
+        "tasks": "speech-recognition",
+        "tags": ["automatic-speech-recognition", data_args.dataset_name],
+        "dataset_args": f"Config: {config_name}, Training split: {data_args.train_split_name}, Eval split: {data_args.eval_split_name}",
+        "dataset": f"{data_args.dataset_name.upper()} - {config_name.upper()}",
+    }
+    if "common_voice" in data_args.dataset_name:
+        kwargs["language"] = config_name
+
+    if training_args.push_to_hub:
+        trainer.push_to_hub(**kwargs)
+    else:
+        trainer.create_model_card(**kwargs)
+
+    return results
+
+
+if __name__ == "__main__":
+    main()

wandb/run-20220201_150818-33gv5m8t/files/config.yaml

@@ -0,0 +1,659 @@
+wandb_version: 1
+
+_n_gpu:
+  desc: null
+  value: 1
+_name_or_path:
+  desc: null
+  value: facebook/wav2vec2-xls-r-1b
+_wandb:
+  desc: null
+  value:
+    cli_version: 0.12.9
+    code_path: code/run_speech_recognition_ctc.py
+    framework: huggingface
+    huggingface_version: 4.17.0.dev0
+    is_jupyter_run: false
+    is_kaggle_kernel: false
+    m:
+    - 1: train/global_step
+      6:
+      - 3
+    python_version: 3.9.7
+    start_time: 1643724498
+    t:
+      1:
+      - 1
+      - 2
+      - 3
+      - 5
+      - 11
+      2:
+      - 1
+      - 2
+      - 3
+      - 5
+      - 11
+      3:
+      - 1
+      - 7
+      - 13
+      4: 3.9.7
+      5: 0.12.9
+      6: 4.17.0.dev0
+      8:
+      - 5
+activation_dropout:
+  desc: null
+  value: 0.055
+adafactor:
+  desc: null
+  value: false
+adam_beta1:
+  desc: null
+  value: 0.9
+adam_beta2:
+  desc: null
+  value: 0.999
+adam_epsilon:
+  desc: null
+  value: 1.0e-08
+adapter_kernel_size:
+  desc: null
+  value: 3
+adapter_stride:
+  desc: null
+  value: 2
+add_adapter:
+  desc: null
+  value: false
+add_cross_attention:
+  desc: null
+  value: false
+apply_spec_augment:
+  desc: null
+  value: true
+architectures:
+  desc: null
+  value:
+  - Wav2Vec2ForPreTraining
+attention_dropout:
+  desc: null
+  value: 0.094
+bad_words_ids:
+  desc: null
+  value: null
+bf16:
+  desc: null
+  value: false
+bf16_full_eval:
+  desc: null
+  value: false
+bos_token_id:
+  desc: null
+  value: 1
+chunk_size_feed_forward:
+  desc: null
+  value: 0
+classifier_proj_size:
+  desc: null
+  value: 256
+codevector_dim:
+  desc: null
+  value: 1024
+contrastive_logits_temperature:
+  desc: null
+  value: 0.1
+conv_bias:
+  desc: null
+  value: true
+conv_dim:
+  desc: null
+  value:
+  - 512
+  - 512
+  - 512
+  - 512
+  - 512
+  - 512
+  - 512
+conv_kernel:
+  desc: null
+  value:
+  - 10
+  - 3
+  - 3
+  - 3
+  - 3
+  - 2
+  - 2
+conv_stride:
+  desc: null
+  value:
+  - 5
+  - 2
+  - 2
+  - 2
+  - 2
+  - 2
+  - 2
+cross_attention_hidden_size:
+  desc: null
+  value: null
+ctc_loss_reduction:
+  desc: null
+  value: mean
+ctc_zero_infinity:
+  desc: null
+  value: false
+dataloader_drop_last:
+  desc: null
+  value: false
+dataloader_num_workers:
+  desc: null
+  value: 0
+dataloader_pin_memory:
+  desc: null
+  value: true
+ddp_bucket_cap_mb:
+  desc: null
+  value: None
+ddp_find_unused_parameters:
+  desc: null
+  value: None
+debug:
+  desc: null
+  value: '[]'
+decoder_start_token_id:
+  desc: null
+  value: null
+deepspeed:
+  desc: null
+  value: None
+disable_tqdm:
+  desc: null
+  value: false
+diversity_loss_weight:
+  desc: null
+  value: 0.1
+diversity_penalty:
+  desc: null
+  value: 0.0
+do_eval:
+  desc: null
+  value: true
+do_predict:
+  desc: null
+  value: false
+do_sample:
+  desc: null
+  value: false
+do_stable_layer_norm:
+  desc: null
+  value: true
+do_train:
+  desc: null
+  value: true
+early_stopping:
+  desc: null
+  value: false
+encoder_no_repeat_ngram_size:
+  desc: null
+  value: 0
+eos_token_id:
+  desc: null
+  value: 2
+eval_accumulation_steps:
+  desc: null
+  value: None
+eval_batch_size:
+  desc: null
+  value: 16
+eval_steps:
+  desc: null
+  value: 500
+evaluation_strategy:
+  desc: null
+  value: steps
+feat_extract_activation:
+  desc: null
+  value: gelu
+feat_extract_dropout:
+  desc: null
+  value: 0.0
+feat_extract_norm:
+  desc: null
+  value: layer
+feat_proj_dropout:
+  desc: null
+  value: 0.04
+feat_quantizer_dropout:
+  desc: null
+  value: 0.0
+final_dropout:
+  desc: null
+  value: 0.0
+finetuning_task:
+  desc: null
+  value: null
+forced_bos_token_id:
+  desc: null
+  value: null
+forced_eos_token_id:
+  desc: null
+  value: null
+fp16:
+  desc: null
+  value: true
+fp16_backend:
+  desc: null
+  value: auto
+fp16_full_eval:
+  desc: null
+  value: false
+fp16_opt_level:
+  desc: null
+  value: O1
+gradient_accumulation_steps:
+  desc: null
+  value: 2
+gradient_checkpointing:
+  desc: null
+  value: true
+greater_is_better:
+  desc: null
+  value: None
+group_by_length:
+  desc: null
+  value: true
+half_precision_backend:
+  desc: null
+  value: amp
+hidden_act:
+  desc: null
+  value: gelu
+hidden_dropout:
+  desc: null
+  value: 0.047
+hidden_size:
+  desc: null
+  value: 1280
+hub_model_id:
+  desc: null
+  value: NbAiLab/wav2vec2-xls-r-1b-npsc
+hub_strategy:
+  desc: null
+  value: every_save
+hub_token:
+  desc: null
+  value: <HUB_TOKEN>
+id2label:
+  desc: null
+  value:
+    '0': LABEL_0
+    '1': LABEL_1
+ignore_data_skip:
+  desc: null
+  value: false
+initializer_range:
+  desc: null
+  value: 0.02
+intermediate_size:
+  desc: null
+  value: 5120
+is_decoder:
+  desc: null
+  value: false
+is_encoder_decoder:
+  desc: null
+  value: false
+label2id:
+  desc: null
+  value:
+    LABEL_0: 0
+    LABEL_1: 1
+label_names:
+  desc: null
+  value: None
+label_smoothing_factor:
+  desc: null
+  value: 0.0
+layer_norm_eps:
+  desc: null
+  value: 1.0e-05
+layerdrop:
+  desc: null
+  value: 0.041
+learning_rate:
+  desc: null
+  value: 0.0001
+length_column_name:
+  desc: null
+  value: input_length
+length_penalty:
+  desc: null
+  value: 1.0
+load_best_model_at_end:
+  desc: null
+  value: false
+local_rank:
+  desc: null
+  value: -1
+log_level:
+  desc: null
+  value: -1
+log_level_replica:
+  desc: null
+  value: -1
+log_on_each_node:
+  desc: null
+  value: true
+logging_dir:
+  desc: null
+  value: ./runs/Feb01_15-07-52_dante
+logging_first_step:
+  desc: null
+  value: false
+logging_nan_inf_filter:
+  desc: null
+  value: true
+logging_steps:
+  desc: null
+  value: 100
+logging_strategy:
+  desc: null
+  value: steps
+lr_scheduler_type:
+  desc: null
+  value: linear
+mask_feature_length:
+  desc: null
+  value: 64
+mask_feature_min_masks:
+  desc: null
+  value: 0
+mask_feature_prob:
+  desc: null
+  value: 0.25
+mask_time_length:
+  desc: null
+  value: 10
+mask_time_min_masks:
+  desc: null
+  value: 2
+mask_time_prob:
+  desc: null
+  value: 0.082
+max_grad_norm:
+  desc: null
+  value: 1.0
+max_length:
+  desc: null
+  value: 20
+max_steps:
+  desc: null
+  value: -1
+metric_for_best_model:
+  desc: null
+  value: None
+min_length:
+  desc: null
+  value: 0
+model_type:
+  desc: null
+  value: wav2vec2
+mp_parameters:
+  desc: null
+  value: ''
+no_cuda:
+  desc: null
+  value: false
+no_repeat_ngram_size:
+  desc: null
+  value: 0
+num_adapter_layers:
+  desc: null
+  value: 3
+num_attention_heads:
+  desc: null
+  value: 16
+num_beam_groups:
+  desc: null
+  value: 1
+num_beams:
+  desc: null
+  value: 1
+num_codevector_groups:
+  desc: null
+  value: 2
+num_codevectors_per_group:
+  desc: null
+  value: 320
+num_conv_pos_embedding_groups:
+  desc: null
+  value: 16
+num_conv_pos_embeddings:
+  desc: null
+  value: 128
+num_feat_extract_layers:
+  desc: null
+  value: 7
+num_hidden_layers:
+  desc: null
+  value: 48
+num_negatives:
+  desc: null
+  value: 100
+num_return_sequences:
+  desc: null
+  value: 1
+num_train_epochs:
+  desc: null
+  value: 15.0
+optim:
+  desc: null
+  value: adamw_hf
+output_attentions:
+  desc: null
+  value: false
+output_dir:
+  desc: null
+  value: ./
+output_hidden_size:
+  desc: null
+  value: 1280
+output_hidden_states:
+  desc: null
+  value: false
+output_scores:
+  desc: null
+  value: false
+overwrite_output_dir:
+  desc: null
+  value: true
+pad_token_id:
+  desc: null
+  value: 31
+past_index:
+  desc: null
+  value: -1
+per_device_eval_batch_size:
+  desc: null
+  value: 16
+per_device_train_batch_size:
+  desc: null
+  value: 16
+per_gpu_eval_batch_size:
+  desc: null
+  value: None
+per_gpu_train_batch_size:
+  desc: null
+  value: None
+prediction_loss_only:
+  desc: null
+  value: false
+prefix:
+  desc: null
+  value: null
+problem_type:
+  desc: null
+  value: null
+proj_codevector_dim:
+  desc: null
+  value: 1024
+pruned_heads:
+  desc: null
+  value: {}
+push_to_hub:
+  desc: null
+  value: true
+push_to_hub_model_id:
+  desc: null
+  value: None
+push_to_hub_organization:
+  desc: null
+  value: None
+push_to_hub_token:
+  desc: null
+  value: <PUSH_TO_HUB_TOKEN>
+remove_invalid_values:
+  desc: null
+  value: false
+remove_unused_columns:
+  desc: null
+  value: true
+repetition_penalty:
+  desc: null
+  value: 1.0
+report_to:
+  desc: null
+  value: '[''tensorboard'', ''wandb'']'
+resume_from_checkpoint:
+  desc: null
+  value: None
+return_dict:
+  desc: null
+  value: true
+return_dict_in_generate:
+  desc: null
+  value: false
+run_name:
+  desc: null
+  value: ./
+save_on_each_node:
+  desc: null
+  value: false
+save_steps:
+  desc: null
+  value: 500
+save_strategy:
+  desc: null
+  value: steps
+save_total_limit:
+  desc: null
+  value: 3
+seed:
+  desc: null
+  value: 42
+sep_token_id:
+  desc: null
+  value: null
+sharded_ddp:
+  desc: null
+  value: '[]'
+skip_memory_metrics:
+  desc: null
+  value: true
+task_specific_params:
+  desc: null
+  value: null
+tdnn_dilation:
+  desc: null
+  value:
+  - 1
+  - 2
+  - 3
+  - 1
+  - 1
+tdnn_dim:
+  desc: null
+  value:
+  - 512
+  - 512
+  - 512
+  - 512
+  - 1500
+tdnn_kernel:
+  desc: null
+  value:
+  - 5
+  - 3
+  - 3
+  - 1
+  - 1
+temperature:
+  desc: null
+  value: 1.0
+tf32:
+  desc: null
+  value: None
+tie_encoder_decoder:
+  desc: null
+  value: false
+tie_word_embeddings:
+  desc: null
+  value: true
+tokenizer_class:
+  desc: null
+  value: null
+top_k:
+  desc: null
+  value: 50
+top_p:
+  desc: null
+  value: 1.0
+torch_dtype:
+  desc: null
+  value: float32
+torchscript:
+  desc: null
+  value: false
+tpu_metrics_debug:
+  desc: null
+  value: false
+tpu_num_cores:
+  desc: null
+  value: None
+train_batch_size:
+  desc: null
+  value: 16
+transformers_version:
+  desc: null
+  value: 4.17.0.dev0
+use_bfloat16:
+  desc: null
+  value: false
+use_legacy_prediction_loop:
+  desc: null
+  value: false
+use_weighted_layer_sum:
+  desc: null
+  value: false
+vocab_size:
+  desc: null
+  value: 34
+warmup_ratio:
+  desc: null
+  value: 0.0
+warmup_steps:
+  desc: null
+  value: 2000
+weight_decay:
+  desc: null
+  value: 0.0
+xpu_backend:
+  desc: null
+  value: None
+xvector_output_dim:
+  desc: null
+  value: 512

wandb/run-20220201_150818-33gv5m8t/files/output.log

@@ -0,0 +1,37 @@
+
+  0%|                                                                                                                                                                                                                                                           | 0/23265 [00:00<?, ?it/s]Traceback (most recent call last):
+  File "/mnt/lv_ai_1_dante/javierr/wav2vec2-xls-r-1b-npsc/run_speech_recognition_ctc.py", line 792, in <module>
+    main()
+  File "/mnt/lv_ai_1_dante/javierr/wav2vec2-xls-r-1b-npsc/run_speech_recognition_ctc.py", line 743, in main
+    train_result = trainer.train(resume_from_checkpoint=checkpoint)
+  File "/mnt/lv_ai_1_dante/javierr/audio/lib/python3.9/site-packages/transformers/trainer.py", line 1384, in train
+    tr_loss_step = self.training_step(model, inputs)
+  File "/mnt/lv_ai_1_dante/javierr/audio/lib/python3.9/site-packages/transformers/trainer.py", line 1959, in training_step
+    loss = self.compute_loss(model, inputs)
+  File "/mnt/lv_ai_1_dante/javierr/audio/lib/python3.9/site-packages/transformers/trainer.py", line 1991, in compute_loss
+    outputs = model(**inputs)
+  File "/mnt/lv_ai_1_dante/javierr/audio/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1102, in _call_impl
+    return forward_call(*input, **kwargs)
+  File "/mnt/lv_ai_1_dante/javierr/audio/lib/python3.9/site-packages/transformers/models/wav2vec2/modeling_wav2vec2.py", line 1756, in forward
+    outputs = self.wav2vec2(
+  File "/mnt/lv_ai_1_dante/javierr/audio/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1102, in _call_impl
+    return forward_call(*input, **kwargs)
+  File "/mnt/lv_ai_1_dante/javierr/audio/lib/python3.9/site-packages/transformers/models/wav2vec2/modeling_wav2vec2.py", line 1346, in forward
+    extract_features = self.feature_extractor(input_values)
+  File "/mnt/lv_ai_1_dante/javierr/audio/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1102, in _call_impl
+    return forward_call(*input, **kwargs)
+  File "/mnt/lv_ai_1_dante/javierr/audio/lib/python3.9/site-packages/transformers/models/wav2vec2/modeling_wav2vec2.py", line 514, in forward
+    hidden_states = conv_layer(hidden_states)
+  File "/mnt/lv_ai_1_dante/javierr/audio/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1102, in _call_impl
+    return forward_call(*input, **kwargs)
+  File "/mnt/lv_ai_1_dante/javierr/audio/lib/python3.9/site-packages/transformers/models/wav2vec2/modeling_wav2vec2.py", line 386, in forward
+    hidden_states = self.conv(hidden_states)
+  File "/mnt/lv_ai_1_dante/javierr/audio/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1102, in _call_impl
+    return forward_call(*input, **kwargs)
+  File "/mnt/lv_ai_1_dante/javierr/audio/lib/python3.9/site-packages/torch/nn/modules/conv.py", line 301, in forward
+    return self._conv_forward(input, self.weight, self.bias)
+  File "/mnt/lv_ai_1_dante/javierr/audio/lib/python3.9/site-packages/torch/nn/modules/conv.py", line 297, in _conv_forward
+    return F.conv1d(input, weight, bias, self.stride,
+RuntimeError: CUDA error: no kernel image is available for execution on the device
+CUDA kernel errors might be asynchronously reported at some other API call,so the stacktrace below might be incorrect.
+For debugging consider passing CUDA_LAUNCH_BLOCKING=1.

wandb/run-20220201_150818-33gv5m8t/files/requirements.txt

@@ -0,0 +1,109 @@
+absl-py==1.0.0
+aiohttp==3.8.1
+aiosignal==1.2.0
+appdirs==1.4.4
+astunparse==1.6.3
+async-timeout==4.0.2
+attrs==21.4.0
+audioread==2.1.9
+cachetools==5.0.0
+certifi==2021.10.8
+cffi==1.15.0
+charset-normalizer==2.0.11
+click==8.0.3
+configparser==5.2.0
+datasets==1.18.3.dev0
+decorator==5.1.1
+deepspeed==0.5.10
+dill==0.3.4
+docker-pycreds==0.4.0
+fairscale==0.4.5
+filelock==3.4.2
+flatbuffers==2.0
+frozenlist==1.3.0
+fsspec==2022.1.0
+gast==0.4.0
+gitdb==4.0.9
+gitpython==3.1.26
+google-auth-oauthlib==0.4.6
+google-auth==2.6.0
+google-pasta==0.2.0
+grpcio==1.43.0
+h5py==3.6.0
+hjson==3.0.2
+huggingface-hub==0.4.0
+idna==3.3
+importlib-metadata==4.10.1
+jiwer==2.3.0
+joblib==1.1.0
+keras-preprocessing==1.1.2
+keras==2.7.0
+libclang==13.0.0
+librosa==0.8.1
+llvmlite==0.38.0
+markdown==3.3.6
+multidict==6.0.2
+multiprocess==0.70.12.2
+ninja==1.10.2.3
+numba==0.55.1
+numpy==1.21.5
+oauthlib==3.2.0
+opt-einsum==3.3.0
+packaging==21.3
+pandas==1.4.0
+pathtools==0.1.2
+pip==20.3.4
+pkg-resources==0.0.0
+pooch==1.6.0
+promise==2.3
+protobuf==3.19.4
+psutil==5.9.0
+py-cpuinfo==8.0.0
+pyarrow==6.0.1
+pyasn1-modules==0.2.8
+pyasn1==0.4.8
+pycparser==2.21
+pyparsing==3.0.7
+python-dateutil==2.8.2
+python-levenshtein==0.12.2
+pytz==2021.3
+pyyaml==6.0
+regex==2022.1.18
+requests-oauthlib==1.3.1
+requests==2.27.1
+resampy==0.2.2
+rsa==4.8
+sacremoses==0.0.47
+scikit-learn==1.0.2
+scipy==1.7.3
+sentry-sdk==1.5.4
+setuptools==44.1.1
+shortuuid==1.0.8
+six==1.16.0
+smmap==5.0.0
+soundfile==0.10.3.post1
+subprocess32==3.5.4
+tensorboard-data-server==0.6.1
+tensorboard-plugin-wit==1.8.1
+tensorboard==2.8.0
+tensorflow-estimator==2.7.0
+tensorflow-io-gcs-filesystem==0.23.1
+tensorflow==2.7.0
+termcolor==1.1.0
+threadpoolctl==3.1.0
+tokenizers==0.11.4
+torch==1.10.2
+torchaudio==0.10.2
+tqdm==4.62.3
+transformers==4.17.0.dev0
+triton==1.0.0
+typing-extensions==4.0.1
+urllib3==1.26.8
+wandb==0.12.9
+werkzeug==2.0.2
+wheel==0.37.1
+wrapt==1.13.3
+xxhash==2.0.2
+yarl==1.7.2
+yaspin==2.1.0
+zipp==3.7.0

wandb/run-20220201_150818-33gv5m8t/files/wandb-metadata.json

@@ -0,0 +1,65 @@
+{
+    "os": "Linux-5.13.0-27-generic-x86_64-with-glibc2.34",
+    "python": "3.9.7",
+    "heartbeatAt": "2022-02-01T14:08:21.090129",
+    "startedAt": "2022-02-01T14:08:18.929956",
+    "docker": null,
+    "gpu": "NVIDIA RTX A6000",
+    "gpu_count": 2,
+    "cpu_count": 96,
+    "cuda": null,
+    "args": [
+        "--dataset_name=NbAiLab/NPSC",
+        "--model_name_or_path=facebook/wav2vec2-xls-r-1b",
+        "--hub_model_id=NbAiLab/wav2vec2-xls-r-1b-npsc",
+        "--dataset_config_name=16K_mp3",
+        "--output_dir=./",
+        "--overwrite_output_dir",
+        "--num_train_epochs=15",
+        "--per_device_train_batch_size=16",
+        "--per_device_eval_batch_size=16",
+        "--gradient_accumulation_steps=2",
+        "--learning_rate=1e-4",
+        "--warmup_steps=2000",
+        "--length_column_name=input_length",
+        "--evaluation_strategy=steps",
+        "--text_column_name=text",
+        "--save_steps=500",
+        "--eval_steps=500",
+        "--logging_steps=100",
+        "--layerdrop=0.041",
+        "--attention_dropout=0.094",
+        "--activation_dropout=0.055",
+        "--hidden_dropout=0.047",
+        "--save_total_limit=3",
+        "--freeze_feature_encoder",
+        "--feat_proj_dropout=0.04",
+        "--mask_time_prob=0.082",
+        "--mask_time_length=10",
+        "--mask_feature_prob=0.25",
+        "--mask_feature_length=64",
+        "--gradient_checkpointing",
+        "--min_duration_in_seconds=0.5",
+        "--max_duration_in_seconds=30.0",
+        "--use_auth_token",
+        "--seed=42",
+        "--fp16",
+        "--group_by_length",
+        "--do_train",
+        "--do_eval",
+        "--push_to_hub",
+        "--preprocessing_num_workers=32"
+    ],
+    "state": "running",
+    "program": "/mnt/lv_ai_1_dante/javierr/wav2vec2-xls-r-1b-npsc/run_speech_recognition_ctc.py",
+    "codePath": "run_speech_recognition_ctc.py",
+    "git": {
+        "remote": "https://huggingface.co/NbAiLab/wav2vec2-xls-r-1b-npsc",
+        "commit": "49ef0066d0c9d470f4582bb5d9d905606961208d"
+    },
+    "email": "versae@gmail.com",
+    "root": "/mnt/lv_ai_1_dante/javierr/wav2vec2-xls-r-1b-npsc",
+    "host": "dante",
+    "username": "javierr",
+    "executable": "/mnt/lv_ai_1_dante/javierr/audio/bin/python"
+}

wandb/run-20220201_150818-33gv5m8t/files/wandb-summary.json

@@ -0,0 +1 @@
+{"_wandb": {"runtime": 6}}

wandb/run-20220201_150818-33gv5m8t/logs/debug-internal.log

@@ -0,0 +1,154 @@
+2022-02-01 15:08:19,400 INFO    MainThread:3248864 [internal.py:wandb_internal():87] W&B internal server running at pid: 3248864, started at: 2022-02-01 15:08:19.399884
+2022-02-01 15:08:19,402 DEBUG   HandlerThread:3248864 [handler.py:handle_request():130] handle_request: check_version
+2022-02-01 15:08:19,402 INFO    WriterThread:3248864 [datastore.py:open_for_write():77] open: /mnt/lv_ai_1_dante/javierr/wav2vec2-xls-r-1b-npsc/wandb/run-20220201_150818-33gv5m8t/run-33gv5m8t.wandb
+2022-02-01 15:08:19,405 DEBUG   SenderThread:3248864 [sender.py:send():234] send: header
+2022-02-01 15:08:19,405 DEBUG   SenderThread:3248864 [sender.py:send_request():248] send_request: check_version
+2022-02-01 15:08:19,562 DEBUG   SenderThread:3248864 [sender.py:send():234] send: run
+2022-02-01 15:08:19,847 INFO    SenderThread:3248864 [dir_watcher.py:__init__():169] watching files in: /mnt/lv_ai_1_dante/javierr/wav2vec2-xls-r-1b-npsc/wandb/run-20220201_150818-33gv5m8t/files
+2022-02-01 15:08:19,848 INFO    SenderThread:3248864 [sender.py:_start_run_threads():804] run started: 33gv5m8t with start time 1643724498
+2022-02-01 15:08:19,848 DEBUG   SenderThread:3248864 [sender.py:send():234] send: summary
+2022-02-01 15:08:19,848 INFO    SenderThread:3248864 [sender.py:_save_file():939] saving file wandb-summary.json with policy end
+2022-02-01 15:08:19,850 DEBUG   HandlerThread:3248864 [handler.py:handle_request():130] handle_request: run_start
+2022-02-01 15:08:20,854 INFO    Thread-8  :3248864 [dir_watcher.py:_on_file_created():217] file/dir created: /mnt/lv_ai_1_dante/javierr/wav2vec2-xls-r-1b-npsc/wandb/run-20220201_150818-33gv5m8t/files/wandb-summary.json
+2022-02-01 15:08:21,089 DEBUG   HandlerThread:3248864 [meta.py:__init__():40] meta init
+2022-02-01 15:08:21,090 DEBUG   HandlerThread:3248864 [meta.py:__init__():54] meta init done
+2022-02-01 15:08:21,090 DEBUG   HandlerThread:3248864 [meta.py:probe():214] probe
+2022-02-01 15:08:21,096 DEBUG   HandlerThread:3248864 [meta.py:_setup_git():204] setup git
+2022-02-01 15:08:21,129 DEBUG   HandlerThread:3248864 [meta.py:_setup_git():211] setup git done
+2022-02-01 15:08:21,130 DEBUG   HandlerThread:3248864 [meta.py:_save_code():92] save code
+2022-02-01 15:08:21,144 DEBUG   HandlerThread:3248864 [meta.py:_save_code():113] save code done
+2022-02-01 15:08:21,144 DEBUG   HandlerThread:3248864 [meta.py:_save_patches():130] save patches
+2022-02-01 15:08:21,223 DEBUG   HandlerThread:3248864 [meta.py:_save_patches():172] save patches done
+2022-02-01 15:08:21,223 DEBUG   HandlerThread:3248864 [meta.py:_save_pip():58] save pip
+2022-02-01 15:08:21,224 DEBUG   HandlerThread:3248864 [meta.py:_save_pip():72] save pip done
+2022-02-01 15:08:21,224 DEBUG   HandlerThread:3248864 [meta.py:probe():252] probe done
+2022-02-01 15:08:21,230 DEBUG   SenderThread:3248864 [sender.py:send():234] send: files
+2022-02-01 15:08:21,231 INFO    SenderThread:3248864 [sender.py:_save_file():939] saving file wandb-metadata.json with policy now
+2022-02-01 15:08:21,232 INFO    SenderThread:3248864 [sender.py:_save_file():939] saving file code/run_speech_recognition_ctc.py with policy now
+2022-02-01 15:08:21,240 DEBUG   HandlerThread:3248864 [handler.py:handle_request():130] handle_request: stop_status
+2022-02-01 15:08:21,245 DEBUG   SenderThread:3248864 [sender.py:send_request():248] send_request: stop_status
+2022-02-01 15:08:21,493 DEBUG   SenderThread:3248864 [sender.py:send():234] send: config
+2022-02-01 15:08:21,495 DEBUG   SenderThread:3248864 [sender.py:send():234] send: metric
+2022-02-01 15:08:21,496 DEBUG   SenderThread:3248864 [sender.py:send():234] send: metric
+2022-02-01 15:08:21,496 WARNING SenderThread:3248864 [sender.py:send_metric():897] Seen metric with glob (shouldnt happen)
+2022-02-01 15:08:21,852 INFO    Thread-8  :3248864 [dir_watcher.py:_on_file_created():217] file/dir created: /mnt/lv_ai_1_dante/javierr/wav2vec2-xls-r-1b-npsc/wandb/run-20220201_150818-33gv5m8t/files/wandb-metadata.json
+2022-02-01 15:08:21,852 INFO    Thread-8  :3248864 [dir_watcher.py:_on_file_created():217] file/dir created: /mnt/lv_ai_1_dante/javierr/wav2vec2-xls-r-1b-npsc/wandb/run-20220201_150818-33gv5m8t/files/code/run_speech_recognition_ctc.py
+2022-02-01 15:08:21,853 INFO    Thread-8  :3248864 [dir_watcher.py:_on_file_created():217] file/dir created: /mnt/lv_ai_1_dante/javierr/wav2vec2-xls-r-1b-npsc/wandb/run-20220201_150818-33gv5m8t/files/requirements.txt
+2022-02-01 15:08:21,853 INFO    Thread-8  :3248864 [dir_watcher.py:_on_file_created():217] file/dir created: /mnt/lv_ai_1_dante/javierr/wav2vec2-xls-r-1b-npsc/wandb/run-20220201_150818-33gv5m8t/files/output.log
+2022-02-01 15:08:21,853 INFO    Thread-8  :3248864 [dir_watcher.py:_on_file_created():217] file/dir created: /mnt/lv_ai_1_dante/javierr/wav2vec2-xls-r-1b-npsc/wandb/run-20220201_150818-33gv5m8t/files/code
+2022-02-01 15:08:21,960 INFO    Thread-11 :3248864 [upload_job.py:push():137] Uploaded file /tmp/tmpxu9hw9m_wandb/bkne7cxj-wandb-metadata.json
+2022-02-01 15:08:21,969 INFO    Thread-12 :3248864 [upload_job.py:push():137] Uploaded file /tmp/tmpxu9hw9m_wandb/1l3kkiz5-code/run_speech_recognition_ctc.py
+2022-02-01 15:08:23,853 INFO    Thread-8  :3248864 [dir_watcher.py:_on_file_modified():230] file/dir modified: /mnt/lv_ai_1_dante/javierr/wav2vec2-xls-r-1b-npsc/wandb/run-20220201_150818-33gv5m8t/files/output.log
+2022-02-01 15:08:25,854 INFO    Thread-8  :3248864 [dir_watcher.py:_on_file_modified():230] file/dir modified: /mnt/lv_ai_1_dante/javierr/wav2vec2-xls-r-1b-npsc/wandb/run-20220201_150818-33gv5m8t/files/output.log
+2022-02-01 15:08:25,932 DEBUG   HandlerThread:3248864 [handler.py:handle_request():130] handle_request: poll_exit
+2022-02-01 15:08:25,933 DEBUG   SenderThread:3248864 [sender.py:send():234] send: telemetry
+2022-02-01 15:08:25,933 DEBUG   SenderThread:3248864 [sender.py:send():234] send: exit
+2022-02-01 15:08:25,933 INFO    SenderThread:3248864 [sender.py:send_exit():366] handling exit code: 1
+2022-02-01 15:08:25,934 INFO    SenderThread:3248864 [sender.py:send_exit():368] handling runtime: 6
+2022-02-01 15:08:25,934 INFO    SenderThread:3248864 [sender.py:_save_file():939] saving file wandb-summary.json with policy end
+2022-02-01 15:08:25,934 INFO    SenderThread:3248864 [sender.py:send_exit():374] send defer
+2022-02-01 15:08:25,935 DEBUG   SenderThread:3248864 [sender.py:send_request():248] send_request: poll_exit
+2022-02-01 15:08:25,936 DEBUG   HandlerThread:3248864 [handler.py:handle_request():130] handle_request: defer
+2022-02-01 15:08:25,936 INFO    HandlerThread:3248864 [handler.py:handle_request_defer():147] handle defer: 0
+2022-02-01 15:08:25,936 DEBUG   SenderThread:3248864 [sender.py:send_request():248] send_request: defer
+2022-02-01 15:08:25,936 INFO    SenderThread:3248864 [sender.py:send_request_defer():383] handle sender defer: 0
+2022-02-01 15:08:25,936 INFO    SenderThread:3248864 [sender.py:transition_state():387] send defer: 1
+2022-02-01 15:08:25,937 DEBUG   HandlerThread:3248864 [handler.py:handle_request():130] handle_request: defer
+2022-02-01 15:08:25,937 INFO    HandlerThread:3248864 [handler.py:handle_request_defer():147] handle defer: 1
+2022-02-01 15:08:26,020 DEBUG   SenderThread:3248864 [sender.py:send_request():248] send_request: defer
+2022-02-01 15:08:26,020 INFO    SenderThread:3248864 [sender.py:send_request_defer():383] handle sender defer: 1
+2022-02-01 15:08:26,020 INFO    SenderThread:3248864 [sender.py:transition_state():387] send defer: 2
+2022-02-01 15:08:26,021 DEBUG   SenderThread:3248864 [sender.py:send():234] send: stats
+2022-02-01 15:08:26,022 DEBUG   HandlerThread:3248864 [handler.py:handle_request():130] handle_request: defer
+2022-02-01 15:08:26,022 INFO    HandlerThread:3248864 [handler.py:handle_request_defer():147] handle defer: 2
+2022-02-01 15:08:26,022 DEBUG   SenderThread:3248864 [sender.py:send_request():248] send_request: defer
+2022-02-01 15:08:26,022 INFO    SenderThread:3248864 [sender.py:send_request_defer():383] handle sender defer: 2
+2022-02-01 15:08:26,022 INFO    SenderThread:3248864 [sender.py:transition_state():387] send defer: 3
+2022-02-01 15:08:26,023 DEBUG   HandlerThread:3248864 [handler.py:handle_request():130] handle_request: defer
+2022-02-01 15:08:26,023 INFO    HandlerThread:3248864 [handler.py:handle_request_defer():147] handle defer: 3
+2022-02-01 15:08:26,023 DEBUG   SenderThread:3248864 [sender.py:send():234] send: summary
+2022-02-01 15:08:26,024 INFO    SenderThread:3248864 [sender.py:_save_file():939] saving file wandb-summary.json with policy end
+2022-02-01 15:08:26,024 DEBUG   SenderThread:3248864 [sender.py:send_request():248] send_request: defer
+2022-02-01 15:08:26,024 INFO    SenderThread:3248864 [sender.py:send_request_defer():383] handle sender defer: 3
+2022-02-01 15:08:26,024 INFO    SenderThread:3248864 [sender.py:transition_state():387] send defer: 4
+2022-02-01 15:08:26,025 DEBUG   HandlerThread:3248864 [handler.py:handle_request():130] handle_request: defer
+2022-02-01 15:08:26,025 INFO    HandlerThread:3248864 [handler.py:handle_request_defer():147] handle defer: 4
+2022-02-01 15:08:26,025 DEBUG   SenderThread:3248864 [sender.py:send_request():248] send_request: defer
+2022-02-01 15:08:26,026 INFO    SenderThread:3248864 [sender.py:send_request_defer():383] handle sender defer: 4
+2022-02-01 15:08:26,038 DEBUG   HandlerThread:3248864 [handler.py:handle_request():130] handle_request: poll_exit
+2022-02-01 15:08:26,367 INFO    SenderThread:3248864 [sender.py:transition_state():387] send defer: 5
+2022-02-01 15:08:26,367 DEBUG   SenderThread:3248864 [sender.py:send_request():248] send_request: poll_exit
+2022-02-01 15:08:26,368 DEBUG   HandlerThread:3248864 [handler.py:handle_request():130] handle_request: defer
+2022-02-01 15:08:26,368 INFO    HandlerThread:3248864 [handler.py:handle_request_defer():147] handle defer: 5
+2022-02-01 15:08:26,368 DEBUG   SenderThread:3248864 [sender.py:send_request():248] send_request: defer
+2022-02-01 15:08:26,368 INFO    SenderThread:3248864 [sender.py:send_request_defer():383] handle sender defer: 5
+2022-02-01 15:08:26,368 INFO    SenderThread:3248864 [dir_watcher.py:finish():283] shutting down directory watcher
+2022-02-01 15:08:26,469 DEBUG   HandlerThread:3248864 [handler.py:handle_request():130] handle_request: poll_exit
+2022-02-01 15:08:26,854 INFO    Thread-8  :3248864 [dir_watcher.py:_on_file_modified():230] file/dir modified: /mnt/lv_ai_1_dante/javierr/wav2vec2-xls-r-1b-npsc/wandb/run-20220201_150818-33gv5m8t/files/wandb-summary.json
+2022-02-01 15:08:26,855 INFO    SenderThread:3248864 [dir_watcher.py:_on_file_modified():230] file/dir modified: /mnt/lv_ai_1_dante/javierr/wav2vec2-xls-r-1b-npsc/wandb/run-20220201_150818-33gv5m8t/files/config.yaml
+2022-02-01 15:08:26,856 INFO    SenderThread:3248864 [dir_watcher.py:_on_file_modified():230] file/dir modified: /mnt/lv_ai_1_dante/javierr/wav2vec2-xls-r-1b-npsc/wandb/run-20220201_150818-33gv5m8t/files/output.log
+2022-02-01 15:08:26,856 INFO    SenderThread:3248864 [dir_watcher.py:finish():313] scan: /mnt/lv_ai_1_dante/javierr/wav2vec2-xls-r-1b-npsc/wandb/run-20220201_150818-33gv5m8t/files
+2022-02-01 15:08:26,856 INFO    SenderThread:3248864 [dir_watcher.py:finish():327] scan save: /mnt/lv_ai_1_dante/javierr/wav2vec2-xls-r-1b-npsc/wandb/run-20220201_150818-33gv5m8t/files/config.yaml config.yaml
+2022-02-01 15:08:26,856 INFO    SenderThread:3248864 [dir_watcher.py:finish():327] scan save: /mnt/lv_ai_1_dante/javierr/wav2vec2-xls-r-1b-npsc/wandb/run-20220201_150818-33gv5m8t/files/requirements.txt requirements.txt
+2022-02-01 15:08:26,857 INFO    SenderThread:3248864 [dir_watcher.py:finish():327] scan save: /mnt/lv_ai_1_dante/javierr/wav2vec2-xls-r-1b-npsc/wandb/run-20220201_150818-33gv5m8t/files/wandb-summary.json wandb-summary.json
+2022-02-01 15:08:26,857 INFO    SenderThread:3248864 [dir_watcher.py:finish():327] scan save: /mnt/lv_ai_1_dante/javierr/wav2vec2-xls-r-1b-npsc/wandb/run-20220201_150818-33gv5m8t/files/output.log output.log
+2022-02-01 15:08:26,857 INFO    SenderThread:3248864 [dir_watcher.py:finish():327] scan save: /mnt/lv_ai_1_dante/javierr/wav2vec2-xls-r-1b-npsc/wandb/run-20220201_150818-33gv5m8t/files/wandb-metadata.json wandb-metadata.json
+2022-02-01 15:08:26,869 INFO    SenderThread:3248864 [dir_watcher.py:finish():327] scan save: /mnt/lv_ai_1_dante/javierr/wav2vec2-xls-r-1b-npsc/wandb/run-20220201_150818-33gv5m8t/files/code/run_speech_recognition_ctc.py code/run_speech_recognition_ctc.py
+2022-02-01 15:08:26,869 INFO    SenderThread:3248864 [sender.py:transition_state():387] send defer: 6
+2022-02-01 15:08:26,869 DEBUG   SenderThread:3248864 [sender.py:send_request():248] send_request: poll_exit
+2022-02-01 15:08:26,881 DEBUG   HandlerThread:3248864 [handler.py:handle_request():130] handle_request: defer
+2022-02-01 15:08:26,881 INFO    HandlerThread:3248864 [handler.py:handle_request_defer():147] handle defer: 6
+2022-02-01 15:08:26,882 DEBUG   SenderThread:3248864 [sender.py:send_request():248] send_request: defer
+2022-02-01 15:08:26,882 INFO    SenderThread:3248864 [sender.py:send_request_defer():383] handle sender defer: 6
+2022-02-01 15:08:26,882 INFO    SenderThread:3248864 [file_pusher.py:finish():177] shutting down file pusher
+2022-02-01 15:08:26,984 DEBUG   HandlerThread:3248864 [handler.py:handle_request():130] handle_request: poll_exit
+2022-02-01 15:08:26,984 DEBUG   SenderThread:3248864 [sender.py:send_request():248] send_request: poll_exit
+2022-02-01 15:08:27,086 DEBUG   HandlerThread:3248864 [handler.py:handle_request():130] handle_request: poll_exit
+2022-02-01 15:08:27,087 DEBUG   SenderThread:3248864 [sender.py:send_request():248] send_request: poll_exit
+2022-02-01 15:08:27,189 DEBUG   HandlerThread:3248864 [handler.py:handle_request():130] handle_request: poll_exit
+2022-02-01 15:08:27,189 DEBUG   SenderThread:3248864 [sender.py:send_request():248] send_request: poll_exit
+2022-02-01 15:08:27,291 DEBUG   HandlerThread:3248864 [handler.py:handle_request():130] handle_request: poll_exit
+2022-02-01 15:08:27,292 DEBUG   SenderThread:3248864 [sender.py:send_request():248] send_request: poll_exit
+2022-02-01 15:08:27,394 DEBUG   HandlerThread:3248864 [handler.py:handle_request():130] handle_request: poll_exit
+2022-02-01 15:08:27,394 DEBUG   SenderThread:3248864 [sender.py:send_request():248] send_request: poll_exit
+2022-02-01 15:08:27,497 DEBUG   HandlerThread:3248864 [handler.py:handle_request():130] handle_request: poll_exit
+2022-02-01 15:08:27,497 DEBUG   SenderThread:3248864 [sender.py:send_request():248] send_request: poll_exit
+2022-02-01 15:08:27,545 INFO    Thread-13 :3248864 [upload_job.py:push():137] Uploaded file /mnt/lv_ai_1_dante/javierr/wav2vec2-xls-r-1b-npsc/wandb/run-20220201_150818-33gv5m8t/files/config.yaml
+2022-02-01 15:08:27,551 INFO    Thread-14 :3248864 [upload_job.py:push():137] Uploaded file /mnt/lv_ai_1_dante/javierr/wav2vec2-xls-r-1b-npsc/wandb/run-20220201_150818-33gv5m8t/files/requirements.txt
+2022-02-01 15:08:27,553 INFO    Thread-16 :3248864 [upload_job.py:push():137] Uploaded file /mnt/lv_ai_1_dante/javierr/wav2vec2-xls-r-1b-npsc/wandb/run-20220201_150818-33gv5m8t/files/output.log
+2022-02-01 15:08:27,565 INFO    Thread-15 :3248864 [upload_job.py:push():137] Uploaded file /mnt/lv_ai_1_dante/javierr/wav2vec2-xls-r-1b-npsc/wandb/run-20220201_150818-33gv5m8t/files/wandb-summary.json
+2022-02-01 15:08:27,599 DEBUG   HandlerThread:3248864 [handler.py:handle_request():130] handle_request: poll_exit
+2022-02-01 15:08:27,600 DEBUG   SenderThread:3248864 [sender.py:send_request():248] send_request: poll_exit
+2022-02-01 15:08:27,702 DEBUG   HandlerThread:3248864 [handler.py:handle_request():130] handle_request: poll_exit
+2022-02-01 15:08:27,702 DEBUG   SenderThread:3248864 [sender.py:send_request():248] send_request: poll_exit
+2022-02-01 15:08:27,766 INFO    Thread-7  :3248864 [sender.py:transition_state():387] send defer: 7
+2022-02-01 15:08:27,767 DEBUG   HandlerThread:3248864 [handler.py:handle_request():130] handle_request: defer
+2022-02-01 15:08:27,767 INFO    HandlerThread:3248864 [handler.py:handle_request_defer():147] handle defer: 7
+2022-02-01 15:08:27,767 DEBUG   SenderThread:3248864 [sender.py:send_request():248] send_request: defer
+2022-02-01 15:08:27,767 INFO    SenderThread:3248864 [sender.py:send_request_defer():383] handle sender defer: 7
+2022-02-01 15:08:27,805 DEBUG   HandlerThread:3248864 [handler.py:handle_request():130] handle_request: poll_exit
+2022-02-01 15:08:28,272 INFO    SenderThread:3248864 [sender.py:transition_state():387] send defer: 8
+2022-02-01 15:08:28,272 DEBUG   SenderThread:3248864 [sender.py:send_request():248] send_request: poll_exit
+2022-02-01 15:08:28,273 DEBUG   HandlerThread:3248864 [handler.py:handle_request():130] handle_request: defer
+2022-02-01 15:08:28,273 INFO    HandlerThread:3248864 [handler.py:handle_request_defer():147] handle defer: 8
+2022-02-01 15:08:28,274 DEBUG   SenderThread:3248864 [sender.py:send_request():248] send_request: defer
+2022-02-01 15:08:28,274 INFO    SenderThread:3248864 [sender.py:send_request_defer():383] handle sender defer: 8
+2022-02-01 15:08:28,274 INFO    SenderThread:3248864 [sender.py:transition_state():387] send defer: 9
+2022-02-01 15:08:28,275 DEBUG   HandlerThread:3248864 [handler.py:handle_request():130] handle_request: defer
+2022-02-01 15:08:28,275 DEBUG   SenderThread:3248864 [sender.py:send():234] send: final
+2022-02-01 15:08:28,275 INFO    HandlerThread:3248864 [handler.py:handle_request_defer():147] handle defer: 9
+2022-02-01 15:08:28,276 DEBUG   SenderThread:3248864 [sender.py:send():234] send: footer
+2022-02-01 15:08:28,276 DEBUG   SenderThread:3248864 [sender.py:send_request():248] send_request: defer
+2022-02-01 15:08:28,276 INFO    SenderThread:3248864 [sender.py:send_request_defer():383] handle sender defer: 9
+2022-02-01 15:08:28,375 DEBUG   HandlerThread:3248864 [handler.py:handle_request():130] handle_request: poll_exit
+2022-02-01 15:08:28,375 DEBUG   SenderThread:3248864 [sender.py:send_request():248] send_request: poll_exit
+2022-02-01 15:08:28,375 INFO    SenderThread:3248864 [file_pusher.py:join():182] waiting for file pusher
+2022-02-01 15:08:28,858 DEBUG   HandlerThread:3248864 [handler.py:handle_request():130] handle_request: get_summary
+2022-02-01 15:08:28,859 DEBUG   HandlerThread:3248864 [handler.py:handle_request():130] handle_request: sampled_history
+2022-02-01 15:08:28,861 DEBUG   HandlerThread:3248864 [handler.py:handle_request():130] handle_request: shutdown
+2022-02-01 15:08:28,861 INFO    HandlerThread:3248864 [handler.py:finish():731] shutting down handler
+2022-02-01 15:08:29,275 INFO    WriterThread:3248864 [datastore.py:close():281] close: /mnt/lv_ai_1_dante/javierr/wav2vec2-xls-r-1b-npsc/wandb/run-20220201_150818-33gv5m8t/run-33gv5m8t.wandb
+2022-02-01 15:08:29,856 INFO    SenderThread:3248864 [sender.py:finish():1070] shutting down sender
+2022-02-01 15:08:29,856 INFO    SenderThread:3248864 [file_pusher.py:finish():177] shutting down file pusher
+2022-02-01 15:08:29,856 INFO    SenderThread:3248864 [file_pusher.py:join():182] waiting for file pusher
+2022-02-01 15:08:29,859 INFO    MainThread:3248864 [internal.py:handle_exit():77] Internal process exited

wandb/run-20220201_150818-33gv5m8t/logs/debug.log

@@ -0,0 +1,149 @@
+2022-02-01 15:08:18,932 INFO    MainThread:3248238 [wandb_setup.py:_flush():71] setting env: {'project': 'wav2vec2', 'entity': 'NbAiLab'}
+2022-02-01 15:08:18,933 INFO    MainThread:3248238 [wandb_setup.py:_flush():71] setting login settings: {}
+2022-02-01 15:08:18,933 INFO    MainThread:3248238 [wandb_init.py:_log_setup():371] Logging user logs to /mnt/lv_ai_1_dante/javierr/wav2vec2-xls-r-1b-npsc/wandb/run-20220201_150818-33gv5m8t/logs/debug.log
+2022-02-01 15:08:18,933 INFO    MainThread:3248238 [wandb_init.py:_log_setup():372] Logging internal logs to /mnt/lv_ai_1_dante/javierr/wav2vec2-xls-r-1b-npsc/wandb/run-20220201_150818-33gv5m8t/logs/debug-internal.log
+2022-02-01 15:08:18,934 INFO    MainThread:3248238 [wandb_init.py:init():404] calling init triggers
+2022-02-01 15:08:18,934 INFO    MainThread:3248238 [wandb_init.py:init():409] wandb.init called with sweep_config: {}
+config: {}
+2022-02-01 15:08:18,934 INFO    MainThread:3248238 [wandb_init.py:init():460] starting backend
+2022-02-01 15:08:18,934 INFO    MainThread:3248238 [backend.py:_multiprocessing_setup():99] multiprocessing start_methods=fork,spawn,forkserver, using: spawn
+2022-02-01 15:08:18,971 INFO    MainThread:3248238 [backend.py:ensure_launched():216] starting backend process...
+2022-02-01 15:08:18,993 INFO    MainThread:3248238 [backend.py:ensure_launched():221] started backend process with pid: 3248864
+2022-02-01 15:08:18,994 INFO    MainThread:3248238 [wandb_init.py:init():469] backend started and connected
+2022-02-01 15:08:19,003 INFO    MainThread:3248238 [wandb_init.py:init():533] updated telemetry
+2022-02-01 15:08:19,058 INFO    MainThread:3248238 [wandb_init.py:init():563] communicating current version
+2022-02-01 15:08:19,560 INFO    MainThread:3248238 [wandb_init.py:init():568] got version response 
+2022-02-01 15:08:19,560 INFO    MainThread:3248238 [wandb_init.py:init():578] communicating run to backend with 30 second timeout
+2022-02-01 15:08:19,849 INFO    MainThread:3248238 [wandb_init.py:init():606] starting run threads in backend
+2022-02-01 15:08:21,239 INFO    MainThread:3248238 [wandb_run.py:_console_start():1810] atexit reg
+2022-02-01 15:08:21,239 INFO    MainThread:3248238 [wandb_run.py:_redirect():1684] redirect: SettingsConsole.REDIRECT
+2022-02-01 15:08:21,241 INFO    MainThread:3248238 [wandb_run.py:_redirect():1689] Redirecting console.
+2022-02-01 15:08:21,243 INFO    MainThread:3248238 [wandb_run.py:_redirect():1745] Redirects installed.
+2022-02-01 15:08:21,243 INFO    MainThread:3248238 [wandb_init.py:init():633] run started, returning control to user process
+2022-02-01 15:08:21,265 INFO    MainThread:3248238 [wandb_run.py:_config_callback():956] 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': 31, '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-1b', 'transformers_version': '4.17.0.dev0', 'feat_extract_dropout': 0.0, 'model_type': 'wav2vec2', 'num_feat_extract_layers': 7, 'hidden_size': 1280, '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': 48, 'intermediate_size': 5120, 'hidden_act': 'gelu', 'num_attention_heads': 16, 'hidden_dropout': 0.047, 'attention_dropout': 0.094, 'activation_dropout': 0.055, 'feat_proj_dropout': 0.04, 'final_dropout': 0.0, 'layerdrop': 0.041, 'layer_norm_eps': 1e-05, 'initializer_range': 0.02, 'vocab_size': 34, 'do_stable_layer_norm': True, 'use_weighted_layer_sum': False, 'apply_spec_augment': True, 'mask_time_prob': 0.082, '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': 1024, 'proj_codevector_dim': 1024, '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': 1280, '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': 16, 'per_device_eval_batch_size': 16, 'per_gpu_train_batch_size': 'None', 'per_gpu_eval_batch_size': 'None', 'gradient_accumulation_steps': 2, 'eval_accumulation_steps': 'None', 'learning_rate': 0.0001, 'weight_decay': 0.0, 'adam_beta1': 0.9, 'adam_beta2': 0.999, 'adam_epsilon': 1e-08, 'max_grad_norm': 1.0, 'num_train_epochs': 15.0, '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/Feb01_15-07-52_dante', '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']", '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': 'NbAiLab/wav2vec2-xls-r-1b-npsc', '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': 16, 'eval_batch_size': 16}
+2022-02-01 15:08:21,269 INFO    MainThread:3248238 [wandb_watch.py:watch():43] Watching
+2022-02-01 15:08:23,566 INFO    MainThread:3248238 [wandb_run.py:_atexit_cleanup():1780] got exitcode: 1
+2022-02-01 15:08:23,569 INFO    MainThread:3248238 [wandb_run.py:_restore():1752] restore
+2022-02-01 15:08:25,936 INFO    MainThread:3248238 [wandb_run.py:_wait_for_finish():1912] got exit ret: file_counts {
+  wandb_count: 1
+  other_count: 1
+}
+pusher_stats {
+  uploaded_bytes: 35314
+  total_bytes: 35314
+}
+
+2022-02-01 15:08:26,368 INFO    MainThread:3248238 [wandb_run.py:_wait_for_finish():1912] got exit ret: file_counts {
+  wandb_count: 1
+  other_count: 1
+}
+pusher_stats {
+  uploaded_bytes: 35314
+  total_bytes: 35314
+}
+
+2022-02-01 15:08:26,882 INFO    MainThread:3248238 [wandb_run.py:_wait_for_finish():1912] got exit ret: file_counts {
+  wandb_count: 5
+  other_count: 1
+}
+pusher_stats {
+  uploaded_bytes: 35314
+  total_bytes: 50031
+}
+
+2022-02-01 15:08:26,985 INFO    MainThread:3248238 [wandb_run.py:_wait_for_finish():1912] got exit ret: file_counts {
+  wandb_count: 5
+  other_count: 1
+}
+pusher_stats {
+  uploaded_bytes: 35314
+  total_bytes: 50031
+}
+
+2022-02-01 15:08:27,087 INFO    MainThread:3248238 [wandb_run.py:_wait_for_finish():1912] got exit ret: file_counts {
+  wandb_count: 5
+  other_count: 1
+}
+pusher_stats {
+  uploaded_bytes: 35314
+  total_bytes: 50031
+}
+
+2022-02-01 15:08:27,190 INFO    MainThread:3248238 [wandb_run.py:_wait_for_finish():1912] got exit ret: file_counts {
+  wandb_count: 5
+  other_count: 1
+}
+pusher_stats {
+  uploaded_bytes: 35314
+  total_bytes: 50031
+}
+
+2022-02-01 15:08:27,293 INFO    MainThread:3248238 [wandb_run.py:_wait_for_finish():1912] got exit ret: file_counts {
+  wandb_count: 5
+  other_count: 1
+}
+pusher_stats {
+  uploaded_bytes: 50031
+  total_bytes: 50031
+}
+
+2022-02-01 15:08:27,395 INFO    MainThread:3248238 [wandb_run.py:_wait_for_finish():1912] got exit ret: file_counts {
+  wandb_count: 5
+  other_count: 1
+}
+pusher_stats {
+  uploaded_bytes: 50031
+  total_bytes: 50031
+}
+
+2022-02-01 15:08:27,498 INFO    MainThread:3248238 [wandb_run.py:_wait_for_finish():1912] got exit ret: file_counts {
+  wandb_count: 5
+  other_count: 1
+}
+pusher_stats {
+  uploaded_bytes: 50031
+  total_bytes: 50031
+}
+
+2022-02-01 15:08:27,601 INFO    MainThread:3248238 [wandb_run.py:_wait_for_finish():1912] got exit ret: file_counts {
+  wandb_count: 5
+  other_count: 1
+}
+pusher_stats {
+  uploaded_bytes: 50031
+  total_bytes: 50031
+}
+
+2022-02-01 15:08:27,703 INFO    MainThread:3248238 [wandb_run.py:_wait_for_finish():1912] got exit ret: file_counts {
+  wandb_count: 5
+  other_count: 1
+}
+pusher_stats {
+  uploaded_bytes: 50031
+  total_bytes: 50031
+}
+
+2022-02-01 15:08:28,273 INFO    MainThread:3248238 [wandb_run.py:_wait_for_finish():1912] got exit ret: file_counts {
+  wandb_count: 5
+  other_count: 1
+}
+pusher_stats {
+  uploaded_bytes: 50031
+  total_bytes: 50031
+}
+
+2022-02-01 15:08:28,856 INFO    MainThread:3248238 [wandb_run.py:_wait_for_finish():1912] got exit ret: done: true
+exit_result {
+}
+file_counts {
+  wandb_count: 5
+  other_count: 1
+}
+pusher_stats {
+  uploaded_bytes: 50031
+  total_bytes: 50031
+}
+local_info {
+}
+
+2022-02-01 15:08:30,178 INFO    MainThread:3248238 [wandb_run.py:_append_files():2180] logging synced files

wandb/run-20220201_153024-1w85vsuu/files/code/run_speech_recognition_ctc.py

@@ -0,0 +1,792 @@
+#!/usr/bin/env python
+# coding=utf-8
+# Copyright 2021 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+
+""" Fine-tuning a 🤗 Transformers CTC model for automatic speech recognition"""
+
+import functools
+import json
+import logging
+import os
+import re
+import sys
+import warnings
+from dataclasses import dataclass, field
+from typing import Dict, List, Optional, Union
+
+import datasets
+import numpy as np
+import torch
+from datasets import DatasetDict, load_dataset, load_metric
+
+import transformers
+from transformers import (
+    AutoConfig,
+    AutoFeatureExtractor,
+    AutoModelForCTC,
+    AutoProcessor,
+    AutoTokenizer,
+    HfArgumentParser,
+    Trainer,
+    TrainingArguments,
+    Wav2Vec2Processor,
+    set_seed,
+)
+from transformers.trainer_utils import get_last_checkpoint, is_main_process
+from transformers.utils import check_min_version
+from transformers.utils.versions import require_version
+
+
+# Will error if the minimal version of Transformers is not installed. Remove at your own risks.
+check_min_version("4.16.0.dev0")
+
+require_version("datasets>=1.13.3", "To fix: pip install -r examples/pytorch/text-classification/requirements.txt")
+
+
+logger = logging.getLogger(__name__)
+
+
+def list_field(default=None, metadata=None):
+    return field(default_factory=lambda: default, metadata=metadata)
+
+
+@dataclass
+class ModelArguments:
+    """
+    Arguments pertaining to which model/config/tokenizer we are going to fine-tune from.
+    """
+
+    model_name_or_path: str = field(
+        metadata={"help": "Path to pretrained model or model identifier from huggingface.co/models"}
+    )
+    tokenizer_name_or_path: Optional[str] = field(
+        default=None,
+        metadata={"help": "Path to pretrained tokenizer or tokenizer identifier from huggingface.co/models"},
+    )
+    cache_dir: Optional[str] = field(
+        default=None,
+        metadata={"help": "Where do you want to store the pretrained models downloaded from huggingface.co"},
+    )
+    freeze_feature_encoder: bool = field(
+        default=True, metadata={"help": "Whether to freeze the feature encoder layers of the model."}
+    )
+    attention_dropout: float = field(
+        default=0.0, metadata={"help": "The dropout ratio for the attention probabilities."}
+    )
+    activation_dropout: float = field(
+        default=0.0, metadata={"help": "The dropout ratio for activations inside the fully connected layer."}
+    )
+    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: float = field(
+        default=0.0,
+        metadata={"help": "The dropout probability for the final projection layer."},
+    )
+    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: int = field(
+        default=10,
+        metadata={"help": "Length of vector span to mask along the time axis."},
+    )
+    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: int = field(
+        default=10,
+        metadata={"help": "Length of vector span to mask along the feature axis."},
+    )
+    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'."}
+    )
+    ctc_zero_infinity: Optional[bool] = field(
+        default=False, metadata={"help": "If True, will try yo aboud the CTC loss goinf to infinity."}
+    )
+
+@dataclass
+class DataTrainingArguments:
+    """
+    Arguments pertaining to what data we are going to input our model for training and eval.
+
+    Using `HfArgumentParser` we can turn this class
+    into argparse arguments to be able to specify them on
+    the command line.
+    """
+
+    dataset_name: str = field(
+        metadata={"help": "The configuration name of the dataset to use (via the datasets library)."}
+    )
+    dataset_config_name: str = field(
+        default=None, metadata={"help": "The configuration name of the dataset to use (via the datasets library)."}
+    )
+    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: 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: str = field(
+        default="audio",
+        metadata={"help": "The name of the dataset column containing the audio data. Defaults to 'audio'"},
+    )
+    text_column_name: str = field(
+        default="text",
+        metadata={"help": "The name of the dataset column containing the text data. Defaults to 'text'"},
+    )
+    overwrite_cache: bool = field(
+        default=False, metadata={"help": "Overwrite the cached preprocessed datasets or not."}
+    )
+    preprocessing_num_workers: Optional[int] = field(
+        default=None,
+        metadata={"help": "The number of processes to use for the preprocessing."},
+    )
+    max_train_samples: Optional[int] = field(
+        default=None,
+        metadata={
+            "help": "For debugging purposes or quicker training, truncate the number of training examples to this "
+            "value if set."
+        },
+    )
+    max_eval_samples: Optional[int] = field(
+        default=None,
+        metadata={
+            "help": "For debugging purposes or quicker training, truncate the number of validation examples to this "
+            "value if set."
+        },
+    )
+    chars_to_ignore: Optional[List[str]] = list_field(
+        default=None,
+        metadata={"help": "A list of characters to remove from the transcripts."},
+    )
+    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: 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: float = field(
+        default=0.0, metadata={"help": "Filter audio files that are shorter than `min_duration_in_seconds` seconds"}
+    )
+    preprocessing_only: bool = field(
+        default=False,
+        metadata={
+            "help": "Whether to only do data preprocessing and skip training. "
+            "This is especially useful when data preprocessing errors out in distributed training due to timeout. "
+            "In this case, one should run the preprocessing in a non-distributed setup with `preprocessing_only=True` "
+            "so that the cached datasets can consequently be loaded in distributed training"
+        },
+    )
+    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: str = field(
+        default="[UNK]",
+        metadata={"help": "The unk token for the tokenizer"},
+    )
+    pad_token: str = field(
+        default="[PAD]",
+        metadata={"help": "The padding token for the tokenizer"},
+    )
+    word_delimiter_token: str = field(
+        default="|",
+        metadata={"help": "The word delimiter token for the tokenizer"},
+    )
+    phoneme_language: Optional[str] = field(
+        default=None,
+        metadata={
+            "help": "The target language that should be used be"
+            " passed to the tokenizer for tokenization. Note that"
+            " this is only relevant if the model classifies the"
+            " input audio to a sequence of phoneme sequences."
+        },
+    )
+
+
+@dataclass
+class DataCollatorCTCWithPadding:
+    """
+    Data collator that will dynamically pad the inputs received.
+    Args:
+        processor (:class:`~transformers.AutoProcessor`)
+            The processor used for proccessing the data.
+        padding (:obj:`bool`, :obj:`str` or :class:`~transformers.tokenization_utils_base.PaddingStrategy`, `optional`, defaults to :obj:`True`):
+            Select a strategy to pad the returned sequences (according to the model's padding side and padding index)
+            among:
+            * :obj:`True` or :obj:`'longest'`: Pad to the longest sequence in the batch (or no padding if only a single
+              sequence if provided).
+            * :obj:`'max_length'`: Pad to a maximum length specified with the argument :obj:`max_length` or to the
+              maximum acceptable input length for the model if that argument is not provided.
+            * :obj:`False` or :obj:`'do_not_pad'` (default): No padding (i.e., can output a batch with sequences of
+              different lengths).
+        max_length (:obj:`int`, `optional`):
+            Maximum length of the ``input_values`` of the returned list and optionally padding length (see above).
+        max_length_labels (:obj:`int`, `optional`):
+            Maximum length of the ``labels`` returned list and optionally padding length (see above).
+        pad_to_multiple_of (:obj:`int`, `optional`):
+            If set will pad the sequence to a multiple of the provided value.
+            This is especially useful to enable the use of Tensor Cores on NVIDIA hardware with compute capability >=
+            7.5 (Volta).
+    """
+
+    processor: AutoProcessor
+    padding: Union[bool, str] = "longest"
+    pad_to_multiple_of: Optional[int] = None
+    pad_to_multiple_of_labels: Optional[int] = None
+
+    def __call__(self, features: List[Dict[str, Union[List[int], torch.Tensor]]]) -> Dict[str, torch.Tensor]:
+        # split inputs and labels since they have to be of different lenghts and need
+        # different padding methods
+        input_features = [{"input_values": feature["input_values"]} for feature in features]
+        label_features = [{"input_ids": feature["labels"]} for feature in features]
+
+        batch = self.processor.pad(
+            input_features,
+            padding=self.padding,
+            pad_to_multiple_of=self.pad_to_multiple_of,
+            return_tensors="pt",
+        )
+
+        with self.processor.as_target_processor():
+            labels_batch = self.processor.pad(
+                label_features,
+                padding=self.padding,
+                pad_to_multiple_of=self.pad_to_multiple_of_labels,
+                return_tensors="pt",
+            )
+
+        # replace padding with -100 to ignore loss correctly
+        labels = labels_batch["input_ids"].masked_fill(labels_batch.attention_mask.ne(1), -100)
+
+        batch["labels"] = labels
+
+        return batch
+
+
+def create_vocabulary_from_data(
+    datasets: DatasetDict,
+    word_delimiter_token: Optional[str] = None,
+    unk_token: Optional[str] = None,
+    pad_token: Optional[str] = None,
+):
+    # Given training and test labels create vocabulary
+    def extract_all_chars(batch):
+        all_text = " ".join(batch["target_text"])
+        vocab = list(set(all_text))
+        return {"vocab": [vocab], "all_text": [all_text]}
+
+    vocabs = datasets.map(
+        extract_all_chars,
+        batched=True,
+        batch_size=-1,
+        keep_in_memory=True,
+        remove_columns=datasets["train"].column_names,
+    )
+
+    # take union of all unique characters in each dataset
+    vocab_set = functools.reduce(
+        lambda vocab_1, vocab_2: set(vocab_1["vocab"][0]) | set(vocab_2["vocab"][0]), vocabs.values()
+    )
+
+    vocab_dict = {v: k for k, v in enumerate(sorted(list(vocab_set)))}
+
+    # replace white space with delimiter token
+    if word_delimiter_token is not None:
+        vocab_dict[word_delimiter_token] = vocab_dict[" "]
+        del vocab_dict[" "]
+
+    # add unk and pad token
+    if unk_token is not None:
+        vocab_dict[unk_token] = len(vocab_dict)
+
+    if pad_token is not None:
+        vocab_dict[pad_token] = len(vocab_dict)
+
+    return vocab_dict
+
+
+def main():
+    # See all possible arguments in src/transformers/training_args.py
+    # or by passing the --help flag to this script.
+    # We now keep distinct sets of args, for a cleaner separation of concerns.
+
+    parser = HfArgumentParser((ModelArguments, DataTrainingArguments, TrainingArguments))
+    if len(sys.argv) == 2 and sys.argv[1].endswith(".json"):
+        # If we pass only one argument to the script and it's the path to a json file,
+        # let's parse it to get our arguments.
+        model_args, data_args, training_args = parser.parse_json_file(json_file=os.path.abspath(sys.argv[1]))
+    else:
+        model_args, data_args, training_args = parser.parse_args_into_dataclasses()
+
+    # Detecting last checkpoint.
+    last_checkpoint = None
+    if os.path.isdir(training_args.output_dir) and training_args.do_train and not training_args.overwrite_output_dir:
+        last_checkpoint = get_last_checkpoint(training_args.output_dir)
+        if last_checkpoint is None and len(os.listdir(training_args.output_dir)) > 0:
+            raise ValueError(
+                f"Output directory ({training_args.output_dir}) already exists and is not empty. "
+                "Use --overwrite_output_dir to overcome."
+            )
+        elif last_checkpoint is not None:
+            logger.info(
+                f"Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change "
+                "the `--output_dir` or add `--overwrite_output_dir` to train from scratch."
+            )
+
+    # Setup logging
+    logging.basicConfig(
+        format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
+        datefmt="%m/%d/%Y %H:%M:%S",
+        handlers=[logging.StreamHandler(sys.stdout)],
+    )
+    logger.setLevel(logging.INFO if is_main_process(training_args.local_rank) else logging.WARN)
+
+    # Log on each process the small summary:
+    logger.warning(
+        f"Process rank: {training_args.local_rank}, device: {training_args.device}, n_gpu: {training_args.n_gpu}"
+        f"distributed training: {bool(training_args.local_rank != -1)}, 16-bits training: {training_args.fp16}"
+    )
+    # Set the verbosity to info of the Transformers logger (on main process only):
+    if is_main_process(training_args.local_rank):
+        transformers.utils.logging.set_verbosity_info()
+    logger.info("Training/evaluation parameters %s", training_args)
+
+    # Set seed before initializing model.
+    set_seed(training_args.seed)
+
+    # Pre-processing dataset
+    import re
+    def filter_numeric(entry):
+        return  (
+            "0" not in entry["text"]
+            and "1" not in entry["text"]
+            and "2" not in entry["text"]
+            and "3" not in entry["text"]
+            and "4" not in entry["text"]
+            and "5" not in entry["text"]
+            and "6" not in entry["text"]
+            and "7" not in entry["text"]
+            and "8" not in entry["text"]
+            and "9" not in entry["text"]
+        )
+
+    def filter_inaudible(entry):
+        return not re.search("\d|<inaudible>", entry["text"], flags=re.IGNORECASE)
+
+    def filter_nynorsk(entry):
+        return re.search("nb-no", entry["sentence_language_code"], flags=re.IGNORECASE)
+
+    def filter_tooshort(entry):
+        #print(f"The audio sample ({entry["audio"]["path"]}) is too small, and has been omitted. "
+        return (len(entry["text"]) <= len(entry["audio"]["array"]) // 320) and (len(entry["text"].strip()) >= 3)
+
+    def map_dataset(entry):
+        batch = {"text": entry["text"].lower()}
+        batch["text"] = re.sub('[áàâ]', 'a', batch["text"])
+        batch["text"] = re.sub('[ä]', 'æ', batch["text"])
+        batch["text"] = re.sub('[éèëê]', 'e', batch["text"])
+        batch["text"] = re.sub('[íìïî]', 'i', batch["text"])
+        batch["text"] = re.sub('[óòöô]', 'o', batch["text"])
+        batch["text"] = re.sub('[ö]', 'ø', batch["text"])
+        batch["text"] = re.sub('[ç]', 'c', batch["text"])
+        batch["text"] = re.sub('[úùüû]', 'u', batch["text"])
+        batch["text"] = re.sub('\s', ' ', batch["text"])
+        batch["text"] = re.sub('<ee>', 'eee', batch["text"])
+        batch["text"] = re.sub('<qq>', 'qqq', batch["text"])
+        batch["text"] = re.sub('<mm>', 'mmm', batch["text"])
+        # batch["text"] = re.sub('<inaudible>', '?', batch["text"])
+        if "<" in batch["text"]:
+            raise ValueError(batch["text"])
+        return batch
+
+    # 1. First, let's load the dataset
+    raw_datasets = DatasetDict()
+
+    if training_args.do_train:
+        raw_datasets["train"] = load_dataset(
+            data_args.dataset_name,
+            data_args.dataset_config_name,
+            split=data_args.train_split_name,
+            use_auth_token=data_args.use_auth_token,
+        )
+        raw_datasets["train"] = raw_datasets["train"].filter(filter_numeric).filter(filter_inaudible).filter(filter_nynorsk).filter(filter_tooshort)
+        raw_datasets["train"] = raw_datasets["train"].map(map_dataset)
+
+        if data_args.audio_column_name not in raw_datasets["train"].column_names:
+            raise ValueError(
+                f"--audio_column_name '{data_args.audio_column_name}' not found in dataset '{data_args.dataset_name}'. "
+                "Make sure to set `--audio_column_name` to the correct audio column - one of "
+                f"{', '.join(raw_datasets['train'].column_names)}."
+            )
+
+        if data_args.text_column_name not in raw_datasets["train"].column_names:
+            raise ValueError(
+                f"--text_column_name {data_args.text_column_name} not found in dataset '{data_args.dataset_name}'. "
+                "Make sure to set `--text_column_name` to the correct text column - one of "
+                f"{', '.join(raw_datasets['train'].column_names)}."
+            )
+
+        if data_args.max_train_samples is not None:
+            raw_datasets["train"] = raw_datasets["train"].select(range(data_args.max_train_samples))
+
+    if training_args.do_eval:
+        raw_datasets["eval"] = load_dataset(
+            data_args.dataset_name,
+            data_args.dataset_config_name,
+            split=data_args.eval_split_name,
+            use_auth_token=data_args.use_auth_token,
+        )
+        raw_datasets["eval"] = raw_datasets["eval"].filter(filter_numeric).filter(filter_inaudible).filter(filter_nynorsk).filter(filter_tooshort)
+        raw_datasets["eval"] = raw_datasets["eval"].map(map_dataset)
+
+        if data_args.max_eval_samples is not None:
+            raw_datasets["eval"] = raw_datasets["eval"].select(range(data_args.max_eval_samples))
+
+
+    # 2. We remove some special characters from the datasets
+    # that make training complicated and do not help in transcribing the speech
+    # E.g. characters, such as `,` and `.` do not really have an acoustic characteristic
+    # that could be easily picked up by the model
+    #chars_to_ignore_regex = (
+    #    f'[{"".join(data_args.chars_to_ignore)}]' if data_args.chars_to_ignore is not None else None
+    #)
+    chars_to_ignore_regex = '[\,\?\.\!\-\;\:\"\“\%\‘\”\�\'\–\_\\\+\#\/]'
+    
+    text_column_name = data_args.text_column_name
+
+    def remove_special_characters(batch):
+        if chars_to_ignore_regex is not None:
+            batch["target_text"] = re.sub(chars_to_ignore_regex, "", batch[text_column_name]).lower() + " "
+        else:
+            batch["target_text"] = batch[text_column_name].lower() + " "
+        return batch
+
+    with training_args.main_process_first(desc="dataset map special characters removal"):
+        raw_datasets = raw_datasets.map(
+            remove_special_characters,
+            remove_columns=[text_column_name],
+            desc="remove special characters from datasets",
+        )
+
+    # save special tokens for tokenizer
+    word_delimiter_token = data_args.word_delimiter_token
+    unk_token = data_args.unk_token
+    pad_token = data_args.pad_token
+
+    # 3. Next, let's load the config as we might need it to create
+    # the tokenizer
+    # load config
+    config = AutoConfig.from_pretrained(
+        model_args.model_name_or_path, cache_dir=model_args.cache_dir, use_auth_token=data_args.use_auth_token
+    )
+
+    # 4. Next, if no tokenizer file is defined,
+    # we create the vocabulary of the model by extracting all unique characters from
+    # the training and evaluation datasets
+    # We need to make sure that only first rank saves vocabulary
+    # make sure all processes wait until vocab is created
+    tokenizer_name_or_path = model_args.tokenizer_name_or_path
+    tokenizer_kwargs = {}
+    if tokenizer_name_or_path is None:
+        # save vocab in training output dir
+        tokenizer_name_or_path = training_args.output_dir
+
+        vocab_file = os.path.join(tokenizer_name_or_path, "vocab.json")
+
+        with training_args.main_process_first():
+            if training_args.overwrite_output_dir and os.path.isfile(vocab_file):
+                os.remove(vocab_file)
+
+        with training_args.main_process_first(desc="dataset map vocabulary creation"):
+            if not os.path.isfile(vocab_file):
+                os.makedirs(tokenizer_name_or_path, exist_ok=True)
+                vocab_dict = create_vocabulary_from_data(
+                    raw_datasets,
+                    word_delimiter_token=word_delimiter_token,
+                    unk_token=unk_token,
+                    pad_token=pad_token,
+                )
+
+                # save vocab dict to be loaded into tokenizer
+                with open(vocab_file, "w") as file:
+                    json.dump(vocab_dict, file)
+
+        # if tokenizer has just been created
+        # it is defined by `tokenizer_class` if present in config else by `model_type`
+        tokenizer_kwargs = {
+            "config": config if config.tokenizer_class is not None else None,
+            "tokenizer_type": config.model_type if config.tokenizer_class is None else None,
+            "unk_token": unk_token,
+            "pad_token": pad_token,
+            "word_delimiter_token": word_delimiter_token,
+        }
+
+    # 5. Now we can instantiate the feature extractor, tokenizer and model
+    # Note for distributed training, the .from_pretrained methods guarantee that only
+    # one local process can concurrently download model & vocab.
+
+    # load feature_extractor and tokenizer
+    tokenizer = AutoTokenizer.from_pretrained(
+        tokenizer_name_or_path,
+        use_auth_token=data_args.use_auth_token,
+        **tokenizer_kwargs,
+    )
+    feature_extractor = AutoFeatureExtractor.from_pretrained(
+        model_args.model_name_or_path, cache_dir=model_args.cache_dir, use_auth_token=data_args.use_auth_token
+    )
+
+    # adapt config
+    config.update(
+        {
+            "feat_proj_dropout": model_args.feat_proj_dropout,
+            "attention_dropout": model_args.attention_dropout,
+            "hidden_dropout": model_args.hidden_dropout,
+            "final_dropout": model_args.final_dropout,
+            "mask_time_prob": model_args.mask_time_prob,
+            "mask_time_length": model_args.mask_time_length,
+            "mask_feature_prob": model_args.mask_feature_prob,
+            "mask_feature_length": model_args.mask_feature_length,
+            "gradient_checkpointing": training_args.gradient_checkpointing,
+            "layerdrop": model_args.layerdrop,
+            "ctc_loss_reduction": model_args.ctc_loss_reduction,
+            "ctc_zero_infinity": model_args.ctc_zero_infinity,
+            "pad_token_id": tokenizer.pad_token_id,
+            "vocab_size": len(tokenizer),
+            "activation_dropout": model_args.activation_dropout,
+        }
+    )
+
+    # create model
+    model = AutoModelForCTC.from_pretrained(
+        model_args.model_name_or_path,
+        cache_dir=model_args.cache_dir,
+        config=config,
+        use_auth_token=data_args.use_auth_token,
+    )
+
+    # freeze encoder
+    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,
+    # so that we just need to set the correct target sampling rate and normalize the input
+    # via the `feature_extractor`
+
+    # make sure that dataset decodes audio with correct sampling rate
+    dataset_sampling_rate = next(iter(raw_datasets.values())).features[data_args.audio_column_name].sampling_rate
+    if dataset_sampling_rate != feature_extractor.sampling_rate:
+        raw_datasets = raw_datasets.cast_column(
+            data_args.audio_column_name, datasets.features.Audio(sampling_rate=feature_extractor.sampling_rate)
+        )
+
+    # derive max & min input length for sample rate & max duration
+    max_input_length = data_args.max_duration_in_seconds * feature_extractor.sampling_rate
+    min_input_length = data_args.min_duration_in_seconds * feature_extractor.sampling_rate
+    audio_column_name = data_args.audio_column_name
+    num_workers = data_args.preprocessing_num_workers
+
+    # `phoneme_language` is only relevant if the model is fine-tuned on phoneme classification
+    phoneme_language = data_args.phoneme_language
+
+    # Preprocessing the datasets.
+    # We need to read the audio files as arrays and tokenize the targets.
+    def prepare_dataset(batch):
+        # load audio
+        sample = batch[audio_column_name]
+
+        inputs = feature_extractor(sample["array"], sampling_rate=sample["sampling_rate"])
+        batch["input_values"] = inputs.input_values[0]
+        batch["input_length"] = len(batch["input_values"])
+
+        # encode targets
+        additional_kwargs = {}
+        if phoneme_language is not None:
+            additional_kwargs["phonemizer_lang"] = phoneme_language
+
+        batch["labels"] = tokenizer(batch["target_text"], **additional_kwargs).input_ids
+        return batch
+
+    with training_args.main_process_first(desc="dataset map preprocessing"):
+        vectorized_datasets = raw_datasets.map(
+            prepare_dataset,
+            remove_columns=next(iter(raw_datasets.values())).column_names,
+            num_proc=num_workers,
+            desc="preprocess datasets",
+        )
+
+        def is_audio_in_length_range(length):
+            return length > min_input_length and length < max_input_length
+
+        # filter data that is shorter than min_input_length
+        vectorized_datasets = vectorized_datasets.filter(
+            is_audio_in_length_range,
+            num_proc=num_workers,
+            input_columns=["input_length"],
+        )
+
+    # 7. Next, we can prepare the training.
+    # Let's use word error rate (WER) as our evaluation metric,
+    # instantiate a data collator and the trainer
+
+    # Define evaluation metrics during training, *i.e.* word error rate, character error rate
+    eval_metrics = {metric: load_metric(metric) for metric in data_args.eval_metrics}
+
+    # for large datasets it is advised to run the preprocessing on a
+    # single machine first with ``args.preprocessing_only`` since there will mostly likely
+    # be a timeout when running the script in distributed mode.
+    # In a second step ``args.preprocessing_only`` can then be set to `False` to load the
+    # cached dataset
+    if data_args.preprocessing_only:
+        logger.info(f"Data preprocessing finished. Files cached at {vectorized_datasets.cache_files}")
+        return
+
+    def compute_metrics(pred):
+        pred_logits = pred.predictions
+        pred_ids = np.argmax(pred_logits, axis=-1)
+
+        pred.label_ids[pred.label_ids == -100] = tokenizer.pad_token_id
+
+        pred_str = tokenizer.batch_decode(pred_ids)
+        # we do not want to group tokens when computing the metrics
+        label_str = tokenizer.batch_decode(pred.label_ids, group_tokens=False)
+
+        metrics = {k: v.compute(predictions=pred_str, references=label_str) for k, v in eval_metrics.items()}
+
+        return metrics
+
+    # Now save everything to be able to create a single processor later
+    if is_main_process(training_args.local_rank):
+        # save feature extractor, tokenizer and config
+        feature_extractor.save_pretrained(training_args.output_dir)
+        tokenizer.save_pretrained(training_args.output_dir)
+        config.save_pretrained(training_args.output_dir)
+
+    try:
+        processor = AutoProcessor.from_pretrained(training_args.output_dir)
+    except (OSError, KeyError):
+        warnings.warn(
+            "Loading a processor from a feature extractor config that does not"
+            " include a `processor_class` attribute is deprecated and will be removed in v5. Please add the following "
+            " attribute to your `preprocessor_config.json` file to suppress this warning: "
+            " `'processor_class': 'Wav2Vec2Processor'`",
+            FutureWarning,
+        )
+        processor = Wav2Vec2Processor.from_pretrained(training_args.output_dir)
+
+    # Instantiate custom data collator
+    data_collator = DataCollatorCTCWithPadding(processor=processor)
+
+    # Initialize Trainer
+    trainer = Trainer(
+        model=model,
+        data_collator=data_collator,
+        args=training_args,
+        compute_metrics=compute_metrics,
+        train_dataset=vectorized_datasets["train"] if training_args.do_train else None,
+        eval_dataset=vectorized_datasets["eval"] if training_args.do_eval else None,
+        tokenizer=feature_extractor,
+    )
+
+    # 8. Finally, we can start training
+
+    # Training
+    if training_args.do_train:
+
+        # use last checkpoint if exist
+        if last_checkpoint is not None:
+            checkpoint = last_checkpoint
+        elif os.path.isdir(model_args.model_name_or_path):
+            checkpoint = model_args.model_name_or_path
+        else:
+            checkpoint = None
+
+        train_result = trainer.train(resume_from_checkpoint=checkpoint)
+        trainer.save_model()
+
+        metrics = train_result.metrics
+        max_train_samples = (
+            data_args.max_train_samples
+            if data_args.max_train_samples is not None
+            else len(vectorized_datasets["train"])
+        )
+        metrics["train_samples"] = min(max_train_samples, len(vectorized_datasets["train"]))
+
+        trainer.log_metrics("train", metrics)
+        trainer.save_metrics("train", metrics)
+        trainer.save_state()
+
+    # Evaluation
+    results = {}
+    if training_args.do_eval:
+        logger.info("*** Evaluate ***")
+        metrics = trainer.evaluate()
+        max_eval_samples = (
+            data_args.max_eval_samples if data_args.max_eval_samples is not None else len(vectorized_datasets["eval"])
+        )
+        metrics["eval_samples"] = min(max_eval_samples, len(vectorized_datasets["eval"]))
+
+        trainer.log_metrics("eval", metrics)
+        trainer.save_metrics("eval", metrics)
+
+    # Write model card and (optionally) push to hub
+    config_name = data_args.dataset_config_name if data_args.dataset_config_name is not None else "na"
+    kwargs = {
+        "finetuned_from": model_args.model_name_or_path,
+        "tasks": "speech-recognition",
+        "tags": ["automatic-speech-recognition", data_args.dataset_name],
+        "dataset_args": f"Config: {config_name}, Training split: {data_args.train_split_name}, Eval split: {data_args.eval_split_name}",
+        "dataset": f"{data_args.dataset_name.upper()} - {config_name.upper()}",
+    }
+    if "common_voice" in data_args.dataset_name:
+        kwargs["language"] = config_name
+
+    if training_args.push_to_hub:
+        trainer.push_to_hub(**kwargs)
+    else:
+        trainer.create_model_card(**kwargs)
+
+    return results
+
+
+if __name__ == "__main__":
+    main()

wandb/run-20220201_153024-1w85vsuu/files/output.log

@@ -0,0 +1,719 @@
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+  0%|█                                                                                                                                                                                                                                              | 100/23265 [06:27<5:40:44,  1.13it/s]
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+  1%|██                                                                                                                                                                                                                                             | 199/23265 [12:55<6:10:42,  1.04it/s]
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+  1%|███                                                                                                                                                                                                                                            | 300/23265 [19:23<5:54:38,  1.08it/s]
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+  2%|████                                                                                                                                                                                                                                           | 399/23265 [25:54<6:11:33,  1.03it/s]
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+  2%|█████▏                                                                                                                                                                                                                                         | 500/23265 [32:26<5:34:06,  1.14it/s]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.
+***** Running Evaluation *****
+  Num examples = 5437
+  Batch size = 16
+{'loss': 0.8361, 'learning_rate': 2.48e-05, 'epoch': 0.32}
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+Configuration saved in ./checkpoint-500/config.json
+{'eval_loss': 0.6304140686988831, 'eval_wer': 0.4970241305264396, 'eval_runtime': 553.4275, 'eval_samples_per_second': 9.824, 'eval_steps_per_second': 0.614, 'epoch': 0.32}
+Model weights saved in ./checkpoint-500/pytorch_model.bin
+Configuration saved in ./checkpoint-500/preprocessor_config.json

wandb/run-20220201_153024-1w85vsuu/files/requirements.txt

@@ -0,0 +1,111 @@
+absl-py==1.0.0
+aiohttp==3.8.1
+aiosignal==1.2.0
+appdirs==1.4.4
+astunparse==1.6.3
+async-timeout==4.0.2
+attrs==21.4.0
+audioread==2.1.9
+cachetools==5.0.0
+certifi==2021.10.8
+cffi==1.15.0
+charset-normalizer==2.0.11
+click==8.0.3
+configparser==5.2.0
+datasets==1.18.3.dev0
+decorator==5.1.1
+deepspeed==0.5.10
+dill==0.3.4
+docker-pycreds==0.4.0
+fairscale==0.4.5
+filelock==3.4.2
+flatbuffers==2.0
+frozenlist==1.3.0
+fsspec==2022.1.0
+gast==0.4.0
+gitdb==4.0.9
+gitpython==3.1.26
+google-auth-oauthlib==0.4.6
+google-auth==2.6.0
+google-pasta==0.2.0
+grpcio==1.43.0
+h5py==3.6.0
+hjson==3.0.2
+huggingface-hub==0.4.0
+idna==3.3
+importlib-metadata==4.10.1
+jiwer==2.3.0
+joblib==1.1.0
+keras-preprocessing==1.1.2
+keras==2.7.0
+libclang==13.0.0
+librosa==0.8.1
+llvmlite==0.38.0
+markdown==3.3.6
+multidict==6.0.2
+multiprocess==0.70.12.2
+ninja==1.10.2.3
+numba==0.55.1
+numpy==1.21.5
+oauthlib==3.2.0
+opt-einsum==3.3.0
+packaging==21.3
+pandas==1.4.0
+pathtools==0.1.2
+pillow==9.0.0
+pip==20.3.4
+pkg-resources==0.0.0
+pooch==1.6.0
+promise==2.3
+protobuf==3.19.4
+psutil==5.9.0
+py-cpuinfo==8.0.0
+pyarrow==6.0.1
+pyasn1-modules==0.2.8
+pyasn1==0.4.8
+pycparser==2.21
+pyparsing==3.0.7
+python-dateutil==2.8.2
+python-levenshtein==0.12.2
+pytz==2021.3
+pyyaml==6.0
+regex==2022.1.18
+requests-oauthlib==1.3.1
+requests==2.27.1
+resampy==0.2.2
+rsa==4.8
+sacremoses==0.0.47
+scikit-learn==1.0.2
+scipy==1.7.3
+sentry-sdk==1.5.4
+setuptools==44.1.1
+shortuuid==1.0.8
+six==1.16.0
+smmap==5.0.0
+soundfile==0.10.3.post1
+subprocess32==3.5.4
+tensorboard-data-server==0.6.1
+tensorboard-plugin-wit==1.8.1
+tensorboard==2.8.0
+tensorflow-estimator==2.7.0
+tensorflow-io-gcs-filesystem==0.23.1
+tensorflow==2.7.0
+termcolor==1.1.0
+threadpoolctl==3.1.0
+tokenizers==0.11.4
+torch==1.10.2+cu113
+torchaudio==0.10.2+cu113
+torchvision==0.11.3+cu113
+tqdm==4.62.3
+transformers==4.17.0.dev0
+triton==1.0.0
+typing-extensions==4.0.1
+urllib3==1.26.8
+wandb==0.12.9
+werkzeug==2.0.2
+wheel==0.37.1
+wrapt==1.13.3
+xxhash==2.0.2
+yarl==1.7.2
+yaspin==2.1.0
+zipp==3.7.0

wandb/run-20220201_153024-1w85vsuu/files/wandb-metadata.json

@@ -0,0 +1,65 @@
+{
+    "os": "Linux-5.13.0-27-generic-x86_64-with-glibc2.34",
+    "python": "3.9.7",
+    "heartbeatAt": "2022-02-01T14:30:26.487845",
+    "startedAt": "2022-02-01T14:30:24.198879",
+    "docker": null,
+    "gpu": "NVIDIA RTX A6000",
+    "gpu_count": 2,
+    "cpu_count": 96,
+    "cuda": null,
+    "args": [
+        "--dataset_name=NbAiLab/NPSC",
+        "--model_name_or_path=facebook/wav2vec2-xls-r-1b",
+        "--hub_model_id=NbAiLab/wav2vec2-xls-r-1b-npsc",
+        "--dataset_config_name=16K_mp3",
+        "--output_dir=./",
+        "--overwrite_output_dir",
+        "--num_train_epochs=15",
+        "--per_device_train_batch_size=16",
+        "--per_device_eval_batch_size=16",
+        "--gradient_accumulation_steps=2",
+        "--learning_rate=1e-4",
+        "--warmup_steps=2000",
+        "--length_column_name=input_length",
+        "--evaluation_strategy=steps",
+        "--text_column_name=text",
+        "--save_steps=500",
+        "--eval_steps=500",
+        "--logging_steps=100",
+        "--layerdrop=0.041",
+        "--attention_dropout=0.094",
+        "--activation_dropout=0.055",
+        "--hidden_dropout=0.047",
+        "--save_total_limit=3",
+        "--freeze_feature_encoder",
+        "--feat_proj_dropout=0.04",
+        "--mask_time_prob=0.082",
+        "--mask_time_length=10",
+        "--mask_feature_prob=0.25",
+        "--mask_feature_length=64",
+        "--gradient_checkpointing",
+        "--min_duration_in_seconds=0.5",
+        "--max_duration_in_seconds=30.0",
+        "--use_auth_token",
+        "--seed=42",
+        "--fp16",
+        "--group_by_length",
+        "--do_train",
+        "--do_eval",
+        "--push_to_hub",
+        "--preprocessing_num_workers=32"
+    ],
+    "state": "running",
+    "program": "/mnt/lv_ai_1_dante/javierr/wav2vec2-xls-r-1b-npsc/run_speech_recognition_ctc.py",
+    "codePath": "run_speech_recognition_ctc.py",
+    "git": {
+        "remote": "https://huggingface.co/NbAiLab/wav2vec2-xls-r-1b-npsc",
+        "commit": "49ef0066d0c9d470f4582bb5d9d905606961208d"
+    },
+    "email": "versae@gmail.com",
+    "root": "/mnt/lv_ai_1_dante/javierr/wav2vec2-xls-r-1b-npsc",
+    "host": "dante",
+    "username": "javierr",
+    "executable": "/mnt/lv_ai_1_dante/javierr/audio/bin/python"
+}

wandb/run-20220201_153024-1w85vsuu/logs/debug.log

@@ -0,0 +1,24 @@
+2022-02-01 15:30:24,201 INFO    MainThread:3265625 [wandb_setup.py:_flush():71] setting env: {'project': 'wav2vec2', 'entity': 'NbAiLab'}
+2022-02-01 15:30:24,202 INFO    MainThread:3265625 [wandb_setup.py:_flush():71] setting login settings: {}
+2022-02-01 15:30:24,202 INFO    MainThread:3265625 [wandb_init.py:_log_setup():371] Logging user logs to /mnt/lv_ai_1_dante/javierr/wav2vec2-xls-r-1b-npsc/wandb/run-20220201_153024-1w85vsuu/logs/debug.log
+2022-02-01 15:30:24,202 INFO    MainThread:3265625 [wandb_init.py:_log_setup():372] Logging internal logs to /mnt/lv_ai_1_dante/javierr/wav2vec2-xls-r-1b-npsc/wandb/run-20220201_153024-1w85vsuu/logs/debug-internal.log
+2022-02-01 15:30:24,203 INFO    MainThread:3265625 [wandb_init.py:init():404] calling init triggers
+2022-02-01 15:30:24,203 INFO    MainThread:3265625 [wandb_init.py:init():409] wandb.init called with sweep_config: {}
+config: {}
+2022-02-01 15:30:24,203 INFO    MainThread:3265625 [wandb_init.py:init():460] starting backend
+2022-02-01 15:30:24,203 INFO    MainThread:3265625 [backend.py:_multiprocessing_setup():99] multiprocessing start_methods=fork,spawn,forkserver, using: spawn
+2022-02-01 15:30:24,277 INFO    MainThread:3265625 [backend.py:ensure_launched():216] starting backend process...
+2022-02-01 15:30:24,340 INFO    MainThread:3265625 [backend.py:ensure_launched():221] started backend process with pid: 3266867
+2022-02-01 15:30:24,342 INFO    MainThread:3265625 [wandb_init.py:init():469] backend started and connected
+2022-02-01 15:30:24,351 INFO    MainThread:3265625 [wandb_init.py:init():533] updated telemetry
+2022-02-01 15:30:24,502 INFO    MainThread:3265625 [wandb_init.py:init():563] communicating current version
+2022-02-01 15:30:24,961 INFO    MainThread:3265625 [wandb_init.py:init():568] got version response 
+2022-02-01 15:30:24,961 INFO    MainThread:3265625 [wandb_init.py:init():578] communicating run to backend with 30 second timeout
+2022-02-01 15:30:25,252 INFO    MainThread:3265625 [wandb_init.py:init():606] starting run threads in backend
+2022-02-01 15:30:26,642 INFO    MainThread:3265625 [wandb_run.py:_console_start():1810] atexit reg
+2022-02-01 15:30:26,643 INFO    MainThread:3265625 [wandb_run.py:_redirect():1684] redirect: SettingsConsole.REDIRECT
+2022-02-01 15:30:26,645 INFO    MainThread:3265625 [wandb_run.py:_redirect():1689] Redirecting console.
+2022-02-01 15:30:26,647 INFO    MainThread:3265625 [wandb_run.py:_redirect():1745] Redirects installed.
+2022-02-01 15:30:26,648 INFO    MainThread:3265625 [wandb_init.py:init():633] run started, returning control to user process
+2022-02-01 15:30:26,671 INFO    MainThread:3265625 [wandb_run.py:_config_callback():956] 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': 31, '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-1b', 'transformers_version': '4.17.0.dev0', 'feat_extract_dropout': 0.0, 'model_type': 'wav2vec2', 'num_feat_extract_layers': 7, 'hidden_size': 1280, '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': 48, 'intermediate_size': 5120, 'hidden_act': 'gelu', 'num_attention_heads': 16, 'hidden_dropout': 0.047, 'attention_dropout': 0.094, 'activation_dropout': 0.055, 'feat_proj_dropout': 0.04, 'final_dropout': 0.0, 'layerdrop': 0.041, 'layer_norm_eps': 1e-05, 'initializer_range': 0.02, 'vocab_size': 34, 'do_stable_layer_norm': True, 'use_weighted_layer_sum': False, 'apply_spec_augment': True, 'mask_time_prob': 0.082, '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': 1024, 'proj_codevector_dim': 1024, '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': 1280, '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': 16, 'per_device_eval_batch_size': 16, 'per_gpu_train_batch_size': 'None', 'per_gpu_eval_batch_size': 'None', 'gradient_accumulation_steps': 2, 'eval_accumulation_steps': 'None', 'learning_rate': 0.0001, 'weight_decay': 0.0, 'adam_beta1': 0.9, 'adam_beta2': 0.999, 'adam_epsilon': 1e-08, 'max_grad_norm': 1.0, 'num_train_epochs': 15.0, '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/Feb01_15-29-26_dante', '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']", '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': 'NbAiLab/wav2vec2-xls-r-1b-npsc', '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': 16, 'eval_batch_size': 16}
+2022-02-01 15:30:26,674 INFO    MainThread:3265625 [wandb_watch.py:watch():43] Watching