From 550947a7d00c9a2cbb079295760a926e1d90bda7 Mon Sep 17 00:00:00 2001 From: wuzewu Date: Thu, 13 Jun 2019 16:06:57 +0800 Subject: [PATCH] Add an option to control whether to load ERNIE using PaddleHub (#2401) * Add an option to control whether to load ERNIE using PaddleHub * Update README --- PaddleNLP/emotion_detection/README.md | 18 +++++++++++++++--- PaddleNLP/emotion_detection/run_ernie.sh | 3 +++ .../emotion_detection/run_ernie_classifier.py | 16 +++++++++++++--- PaddleNLP/models/representation/ernie.py | 3 +-- 4 files changed, 32 insertions(+), 8 deletions(-) diff --git a/PaddleNLP/emotion_detection/README.md b/PaddleNLP/emotion_detection/README.md index 11bb00a4..71de7b3e 100644 --- a/PaddleNLP/emotion_detection/README.md +++ b/PaddleNLP/emotion_detection/README.md @@ -19,10 +19,9 @@ ## 快速开始 -本项目依赖于 Python2.7、Paddlepaddle Fluid 1.4.0以及PaddleHub 0.5.0，请确保相关依赖都已安装正确 +本项目依赖于 Paddlepaddle 1.3.2 及以上版本，请参考 [安装指南](http://www.paddlepaddle.org/#quick-start) 进行安装 -[PaddlePaddle安装指南](http://www.paddlepaddle.org/#quick-start) -[PaddleHub安装指南](https://github.com/PaddlePaddle/PaddleHub) +python版本依赖python 2.7 #### 安装代码 @@ -169,6 +168,19 @@ python tokenizer.py --test_data_dir ./test.txt.utf8 --batch_size 1 > test.txt.ut --init_checkpoint ./models/ernie_finetune/params ``` +我们也提供了使用PaddleHub加载ERNIE模型的选项，PaddleHub是PaddlePaddle的预训练模型管理工具，可以一行代码完成预训练模型的加载，简化预训练模型的使用和迁移学习。更多相关的介绍，可以查看[PaddleHub](https://github.com/PaddlePaddle/PaddleHub) + +如果想使用该功能，需要修改run_ernie.sh中的配置如下： +```shell +# 在train()函数中，修改--use_paddle_hub选项 +--use_paddle_hub true +``` + +注意：使用该选项需要先安装PaddleHub，安装命令如下 +```shell +$ pip install paddlehub +``` + ## 如何贡献代码 如果你可以修复某个issue或者增加一个新功能，欢迎给我们提交PR。如果对应的PR被接受了，我们将根据贡献的质量和难度进行打分（0-5分，越高越好）。如果你累计获得了10分，可以联系我们获得面试机会或者为你写推荐信。 diff --git a/PaddleNLP/emotion_detection/run_ernie.sh b/PaddleNLP/emotion_detection/run_ernie.sh index e90faf2f..56c9bc7a 100644 --- a/PaddleNLP/emotion_detection/run_ernie.sh +++ b/PaddleNLP/emotion_detection/run_ernie.sh @@ -12,6 +12,7 @@ train() { --verbose true \ --do_train true \ --do_val true \ + --use_paddle_hub false \ --batch_size 32 \ --init_checkpoint ${MODEL_PATH}/params \ --train_set ${TASK_DATA_PATH}/train.tsv \ @@ -35,6 +36,7 @@ evaluate() { --use_cuda true \ --verbose true \ --do_val true \ + --use_paddle_hub false \ --batch_size 32 \ --init_checkpoint ${MODEL_PATH}/params \ --test_set ${TASK_DATA_PATH}/test.tsv \ @@ -50,6 +52,7 @@ infer() { --use_cuda true \ --verbose true \ --do_infer true \ + --use_paddle_hub false \ --batch_size 32 \ --init_checkpoint ${MODEL_PATH}/params \ --infer_set ${TASK_DATA_PATH}/infer.tsv \ diff --git a/PaddleNLP/emotion_detection/run_ernie_classifier.py b/PaddleNLP/emotion_detection/run_ernie_classifier.py index 8a996740..2a5c20f1 100644 --- a/PaddleNLP/emotion_detection/run_ernie_classifier.py +++ b/PaddleNLP/emotion_detection/run_ernie_classifier.py @@ -28,6 +28,7 @@ model_g.add_arg("ernie_config_path", str, None, "Path to the json file for ernie model_g.add_arg("senta_config_path", str, None, "Path to the json file for senta model config.") model_g.add_arg("init_checkpoint", str, None, "Init checkpoint to resume training from.") model_g.add_arg("output_dir", str, "checkpoints", "Path to save checkpoints") +model_g.add_arg("use_paddle_hub", bool, False, "Whether to load ERNIE using PaddleHub") train_g = utils.ArgumentGroup(parser, "training", "training options.") train_g.add_arg("epoch", int, 10, "Number of epoches for training.") @@ -201,7 +202,10 @@ def main(args): pyreader_name='train_reader') # get ernie_embeddings - embeddings = ernie.ernie_encoder_with_paddle_hub(ernie_inputs, args.max_seq_len) + if args.use_paddle_hub: + embeddings = ernie.ernie_encoder_with_paddle_hub(ernie_inputs, args.max_seq_len) + else: + embeddings = ernie.ernie_encoder(ernie_inputs, ernie_config=ernie_config) # user defined model based on ernie embeddings loss, accuracy, num_seqs = create_model( @@ -233,7 +237,10 @@ def main(args): pyreader_name='eval_reader') # get ernie_embeddings - embeddings = ernie.ernie_encoder_with_paddle_hub(ernie_inputs, args.max_seq_len) + if args.use_paddle_hub: + embeddings = ernie.ernie_encoder_with_paddle_hub(ernie_inputs, args.max_seq_len) + else: + embeddings = ernie.ernie_encoder(ernie_inputs, ernie_config=ernie_config) # user defined model based on ernie embeddings loss, accuracy, num_seqs = create_model( @@ -253,7 +260,10 @@ def main(args): pyreader_name='infer_reader') # get ernie_embeddings - embeddings = ernie.ernie_encoder_with_paddle_hub(ernie_inputs, args.max_seq_len) + if args.use_paddle_hub: + embeddings = ernie.ernie_encoder_with_paddle_hub(ernie_inputs, args.max_seq_len) + else: + embeddings = ernie.ernie_encoder(ernie_inputs, ernie_config=ernie_config) probs = create_model(args, embeddings, diff --git a/PaddleNLP/models/representation/ernie.py b/PaddleNLP/models/representation/ernie.py index 8dbc5c6e..90061d9d 100644 --- a/PaddleNLP/models/representation/ernie.py +++ b/PaddleNLP/models/representation/ernie.py @@ -57,8 +57,7 @@ def ernie_encoder_with_paddle_hub(ernie_inputs, max_seq_len): pre_program=main_program, next_program=program, input_dict=input_dict, - inplace=True, - need_log=False) + inplace=True) enc_out = outputs["sequence_output"] unpad_enc_out = fluid.layers.sequence_unpad( -- GitLab