From fa7297855d2cde0cf5aeacc0991ae8025333942c Mon Sep 17 00:00:00 2001 From: "shixian.shi" Date: Fri, 5 May 2023 15:31:06 +0800 Subject: [PATCH] update contextual finetune --- .../finetune.py | 13 ++++++------- 1 file changed, 6 insertions(+), 7 deletions(-) diff --git a/egs_modelscope/asr/paraformer/speech_paraformer-large-contextual_asr_nat-zh-cn-16k-common-vocab8404/finetune.py b/egs_modelscope/asr/paraformer/speech_paraformer-large-contextual_asr_nat-zh-cn-16k-common-vocab8404/finetune.py index 676c943af..e4d6682b0 100644 --- a/egs_modelscope/asr/paraformer/speech_paraformer-large-contextual_asr_nat-zh-cn-16k-common-vocab8404/finetune.py +++ b/egs_modelscope/asr/paraformer/speech_paraformer-large-contextual_asr_nat-zh-cn-16k-common-vocab8404/finetune.py @@ -3,7 +3,6 @@ import os from modelscope.metainfo import Trainers from modelscope.trainers import build_trainer -import funasr from funasr.datasets.ms_dataset import MsDataset from funasr.utils.modelscope_param import modelscope_args @@ -27,11 +26,11 @@ def modelscope_finetune(params): if __name__ == '__main__': params = modelscope_args(model="damo/speech_paraformer-large-contextual_asr_nat-zh-cn-16k-common-vocab8404", data_path="./data") - params.output_dir = "./checkpoint" # m模型保存路径 + params.output_dir = "./checkpoint" # 模型保存路径 params.data_path = "./example_data/" # 数据路径 - params.dataset_type = "large" # 小数据量设置small，若数据量大于1000小时，请使用large - params.batch_bins = 2000 # batch size，如果dataset_type="small"，batch_bins单位为fbank特征帧数，如果dataset_type="large"，batch_bins单位为毫秒， - params.max_epoch = 50 # 最大训练轮数 + params.dataset_type = "large" # finetune contextual paraformer模型只能使用large dataset + params.batch_bins = 200000 # batch size，如果dataset_type="small"，batch_bins单位为fbank特征帧数，如果dataset_type="large"，batch_bins单位为毫秒， + params.max_epoch = 20 # 最大训练轮数 params.lr = 0.00005 # 设置学习率 - - modelscope_finetune(params) + + modelscope_finetune(params) \ No newline at end of file