From 2deb74ec94170f6807bf5a96a1132005294b7152 Mon Sep 17 00:00:00 2001 From: Feiyu Chan Date: Tue, 29 Oct 2019 14:29:16 +0800 Subject: [PATCH] Dv3: add link for trained model (#3796) dv3: add trained model link for downloading --- PaddleSpeech/DeepVoice3/README.md | 5 ++++- PaddleSpeech/DeepVoice3/README_cn.md | 4 +++- 2 files changed, 7 insertions(+), 2 deletions(-) diff --git a/PaddleSpeech/DeepVoice3/README.md b/PaddleSpeech/DeepVoice3/README.md index 40ca7af3..b090b861 100644 --- a/PaddleSpeech/DeepVoice3/README.md +++ b/PaddleSpeech/DeepVoice3/README.md @@ -119,7 +119,10 @@ For more details about `train.py`, see `python train.py --help`. #### load checkpoints -You can load saved checkpoint and resume training with `--checkpoint`, if you wan to reset optimizer states, pass `--reset-optimizer` in addition. +We provide a trained model ([dv3.single_frame](https://paddlespeech.bj.bcebos.com/Parakeet/dv3.single_frame.tar.gz)) for downloading, which is trained with the default preset. Unzip the downloaded file with `tar xzvf dv3.single_frame.tar.gz`, you will get `config.json`, `model.pdparams` and `README.md`. `config.json` is the preset json with which the model is trained, `model.pdparams` is the parameter file, and `README.md` is a brief introduction of the model. + +You can load saved checkpoint and resume training with `--checkpoint` (You only need to provide the base name of the parameter file, eg. if you want to load `model.pdparams`, just use `--checkpoint=model`). If there is also a file with the same basename and extension name `.pdopt` in the same folder with the model file (i.e. `model.pdopt`, which is the optimizer file), it is also loaded automatically. If you wan to reset optimizer states, pass `--reset-optimizer` in addition. + #### train a part of the model diff --git a/PaddleSpeech/DeepVoice3/README_cn.md b/PaddleSpeech/DeepVoice3/README_cn.md index 9574b7fc..0828c983 100644 --- a/PaddleSpeech/DeepVoice3/README_cn.md +++ b/PaddleSpeech/DeepVoice3/README_cn.md @@ -118,7 +118,9 @@ python train.py --data-root=${data-root} --use-gpu \ #### 加载保存的模型 -用户可以通过 `--checkpoint` 参数加载保存的模型并恢复训练。如果你想要重置优化器的状态,在训练脚本加入 `--reset-optimizer` 参数。 +我们提供了使用默认的配置文件训练的模型 [dv3.single_frame](https://paddlespeech.bj.bcebos.com/Parakeet/dv3.single_frame.tar.gz) 供用户下载。使用 `tar xzvf dv3.single_frame.tar.gz` 解压下载的文件,会得到 `config.json`, `model.pdparams` and `README.md`。其中 `config.json` 是模型训练时使用的配置文件,`model.pdparams` 是参数文件,`README.md` 是模型的简要说明。 + +用户可以通过 `--checkpoint` 参数加载保存的模型并恢复训练(注意:只需要传基础文件名,不需要扩展名,例如需要加载 `model.pdparams` 那么,只需要使用 `--checkpoint=model`)。如果同一个文件夹内有一个和参数文件基础文件名相同,而后缀为 `.pdopt` 的文件,(如 `model.pdopt`,即优化器文件),那么该文件也会被自动加载。如果你想要重置优化器的状态,在训练脚本加入 `--reset-optimizer` 参数。 #### 训练模型的一部分 -- GitLab