finetuning_wav2vec.json

{
    "model_name_or_path": "facebook/wav2vec2-large-xlsr-53",
    "dataset_name" : "mozilla-foundation/common_voice_7_0",
    "dataset_config_name" : "lg",
    "dataset_data_dir": "cv-corpus-16.1-2023-12-06",
    "dataset_min_filesize": 10000,
    "dataset_min_textlength": 10,
    "output_dir": "output/wav2vec2-common_voice-lg",
    "overwrite_output_dir": true,
    "num_train_epochs": 200,
    "per_device_eval_batch_size": 8,
    "per_device_train_batch_size": 8,
    "evaluation_strategy": "steps",
    "learning_rate": 1e-4,
    "warmup_steps": 300,
    "fp16": true,
    "freeze_feature_extractor": true,
    "save_steps": 2000,
    "eval_steps": 2000,
    "save_total_limit": 1,
    "logging_steps": 2000,
    "group_by_length": true,
    "feat_proj_dropout": 0.04,
    "layerdrop": 0.041,
    "attention_dropout": 0.094,
    "activation_dropout": 0.055,
    "hidden_dropout": 0.047,
    "mask_time_prob": 0.4,
    "do_train": true,
    "do_eval": true,
    "gradient_accumulation_steps": 2,
    "dataloader_num_workers": 8,
    "train_split_name": "other",
    "chars_to_ignore": [",","?",".","!","-",";",":","\""]
}