| New file |
| | |
|---|
| | | import os |
|---|
| | | |
|---|
| | | from modelscope.metainfo import Trainers |
|---|
| | | from modelscope.trainers import build_trainer |
|---|
| | | |
|---|
| | | from funasr.datasets.ms_dataset import MsDataset |
|---|
| | | from funasr.utils.modelscope_param import modelscope_args |
|---|
| | | |
|---|
| | | |
|---|
| | | def modelscope_finetune(params): |
|---|
| | | if not os.path.exists(params.output_dir): |
|---|
| | | os.makedirs(params.output_dir, exist_ok=True) |
|---|
| | | # dataset split ["train", "validation"] |
|---|
| | | ds_dict = MsDataset.load(params.data_path) |
|---|
| | | kwargs = dict( |
|---|
| | | model=params.model, |
|---|
| | | data_dir=ds_dict, |
|---|
| | | dataset_type=params.dataset_type, |
|---|
| | | work_dir=params.output_dir, |
|---|
| | | batch_bins=params.batch_bins, |
|---|
| | | max_epoch=params.max_epoch, |
|---|
| | | lr=params.lr) |
|---|
| | | trainer = build_trainer(Trainers.speech_asr_trainer, default_args=kwargs) |
|---|
| | | trainer.train() |
|---|
| | | |
|---|
| | | |
|---|
| | | if __name__ == '__main__': |
|---|
| | | params = modelscope_args(model="damo/speech_data2vec_pretrain-paraformer-zh-cn-aishell2-16k", |
|---|
| | | data_path="./data") |
|---|
| | | params.output_dir = "./checkpoint" |
|---|
| | | params.data_path = "./example_data/" |
|---|
| | | params.dataset_type = "small" |
|---|
| | | params.batch_bins = 16000 |
|---|
| | | params.max_epoch = 50 |
|---|
| | | params.lr = 0.00002 |
|---|
| | | |
|---|
| | | modelscope_finetune(params) |
|---|
