diff --git a/demo/ernie-classification/README.md b/demo/ernie-classification/README.md index 8d72ebd06338358ea5c66875dd62f2b55a6e3998..f2ad1ff12ea5d3af395a1c90b4bd9d600b6f7e21 100644 --- a/demo/ernie-classification/README.md +++ b/demo/ernie-classification/README.md @@ -1,3 +1,59 @@ # ERNIE Classification -本示例如果使用PaddleHub Finetune API快速的完成Transformer类模型ERNIE或BERT完成文本分类任务。 +本示例将展示如何使用PaddleHub Finetune API利用ERNIE完成分类任务。 + +其中分类任务可以分为两大类 + +* 单句分类 +- 中文情感分析任务 ChnSentiCorp + + +* 句对分类 +- 语义相似度 LCQMC +- 检索式问答任务 nlpcc-dbqa + +## 如何开始Finetune + +在完成安装PaddlePaddle与PaddleHub后，通过执行脚本`sh run_sentiment_cls.sh`即可开始使用ERNIE对ChnSentiCorp数据集进行Finetune。 + +其中脚本参数说明如下： + +```bash +--batch_size: 批处理大小，请结合显存情况进行调整，若出现显存不足错误，请调低这一参数值 +--weight_decay: +--checkpoint_dir: 模型保存路径，PaddleHub会自动保存验证集上表现最好的模型 +--num_epoch: Finetune迭代的轮数 +--max_seq_len: ERNIE模型使用的最大序列长度，最大不能超过512, + 若出现显存不足错误，请调低这一参数 +``` + +## 代码步骤 + +使用PaddleHub Finetune API进行Finetune可以分为一下4个步骤 + +### Step1: 加载预训练模型 + +```python + module = hub.Module(name="ernie") + inputs, outputs, program = module.context( + trainable=True, max_seq_len=128) +``` +其中最大序列长度`max_seq_len`是可以调整的参数，建议值128，根据任务文本长度不同可以调整该值，但最大不超过512。 + +如果想尝试BERT模型，例如BERT中文模型，只需要更换Module中的参数即可. +PaddleHub除了ERNIE，还提供以下BERT模型: + +BERT模型名 | PaddleHub Module name +---------------------------------- | :------: +BERT-Base, Uncased | bert_uncased_L-12_H-768_A-12 +BERT-Large, Uncased | bert_uncased_L-24_H-1024_A-16 +BERT-Base, Cased | bert_cased_L-12_H-768_A-12 +BERT-Large, Cased | bert_cased_L-24_H-1024_A-16 +BERT-Base, Multilingual Cased | bert_multi_cased_L-12_H-768_A-12 +BERT-Base, Chinese | bert_chinese_L-12_H-768_A-12 + + +```python + # 即可无缝切换BERT中文模型 + module = hub.Module(name="bert_chinese_L-12_H-768_A-12") +``` diff --git a/demo/ernie-classification/question_answering.py b/demo/ernie-classification/question_answering.py index d9e7548a3299e251ac11f4ab59825e78cf0abb24..3429b488860427bc74df8ea6adf7bed97c30dcb9 100644 --- a/demo/ernie-classification/question_answering.py +++ b/demo/ernie-classification/question_answering.py @@ -22,7 +22,6 @@ import paddlehub as hub parser = argparse.ArgumentParser(__doc__) parser.add_argument("--num_epoch", type=int, default=3, help="Number of epoches for fine-tuning.") parser.add_argument("--learning_rate", type=float, default=5e-5, help="Learning rate used to train with warmup.") -parser.add_argument("--hub_module_dir", type=str, default=None, help="PaddleHub module directory") parser.add_argument("--weight_decay", type=float, default=0.01, help="Weight decay rate for L2 regularizer.") parser.add_argument("--data_dir", type=str, default=None, help="Path to training data.") parser.add_argument("--checkpoint_dir", type=str, default=None, help="Directory to model checkpoint") diff --git a/demo/ernie-classification/question_matching.py b/demo/ernie-classification/question_matching.py index a9d24c4ea4540017f69988a121afd1602b395832..af2a0f6492c4ea169fec7906e49231d26d65acc2 100644 --- a/demo/ernie-classification/question_matching.py +++ b/demo/ernie-classification/question_matching.py @@ -22,7 +22,6 @@ import paddlehub as hub parser = argparse.ArgumentParser(__doc__) parser.add_argument("--num_epoch", type=int, default=3, help="Number of epoches for fine-tuning.") parser.add_argument("--learning_rate", type=float, default=5e-5, help="Learning rate used to train with warmup.") -parser.add_argument("--hub_module_dir", type=str, default=None, help="PaddleHub module directory") parser.add_argument("--weight_decay", type=float, default=0.01, help="Weight decay rate for L2 regularizer.") parser.add_argument("--data_dir", type=str, default=None, help="Path to training data.") parser.add_argument("--checkpoint_dir", type=str, default=None, help="Directory to model checkpoint") diff --git a/demo/ernie-classification/run_question_matching.sh b/demo/ernie-classification/run_question_matching.sh index 2230d8e0b713fc285f083dd3fb26d08a98d744df..7f349b942f765610cdcc60f4d2877cabe6afe90a 100644 --- a/demo/ernie-classification/run_question_matching.sh +++ b/demo/ernie-classification/run_question_matching.sh @@ -1,4 +1,4 @@ -export CUDA_VISIBLE_DEVICES=0 +export CUDA_VISIBLE_DEVICES=5 CKPT_DIR="./ckpt_question_matching" python -u question_matching.py \ diff --git a/demo/ernie-classification/run_sentiment_cls.sh b/demo/ernie-classification/run_sentiment_cls.sh index 34203b1ae2e4d9c6230610f2b31759abb8c57930..246a4fa278dd132bc600435ba756631c04de5abc 100644 --- a/demo/ernie-classification/run_sentiment_cls.sh +++ b/demo/ernie-classification/run_sentiment_cls.sh @@ -1,4 +1,4 @@ -export CUDA_VISIBLE_DEVICES=3 +export CUDA_VISIBLE_DEVICES=5 CKPT_DIR="./ckpt_sentiment_cls" python -u sentiment_cls.py \ diff --git a/demo/ernie-classification/sentiment_cls.py b/demo/ernie-classification/sentiment_cls.py index c6bd5605d39d8d95b80d6e7ef05b1ca1e106d7ec..ec6b1b4be08d271effaa871c9d9a93e15192deab 100644 --- a/demo/ernie-classification/sentiment_cls.py +++ b/demo/ernie-classification/sentiment_cls.py @@ -22,7 +22,6 @@ import paddlehub as hub parser = argparse.ArgumentParser(__doc__) parser.add_argument("--num_epoch", type=int, default=3, help="Number of epoches for fine-tuning.") parser.add_argument("--learning_rate", type=float, default=5e-5, help="Learning rate used to train with warmup.") -parser.add_argument("--hub_module_dir", type=str, default=None, help="PaddleHub module directory") parser.add_argument("--weight_decay", type=float, default=0.01, help="Weight decay rate for L2 regularizer.") parser.add_argument("--data_dir", type=str, default=None, help="Path to training data.") parser.add_argument("--checkpoint_dir", type=str, default=None, help="Directory to model checkpoint") diff --git a/demo/ernie-seq-labeling/run_sequence_labeling.sh b/demo/ernie-seq-labeling/run_sequence_labeling.sh index 89aa22f683f97aedb14d37d46470d12fe0176651..ddd24d2a3d72318647e9ba4092cffccfb3498acc 100644 --- a/demo/ernie-seq-labeling/run_sequence_labeling.sh +++ b/demo/ernie-seq-labeling/run_sequence_labeling.sh @@ -1,4 +1,4 @@ -export CUDA_VISIBLE_DEVICES=0 +export CUDA_VISIBLE_DEVICES=6 CKPT_DIR="./ckpt_sequence_labeling" diff --git a/demo/ernie-seq-labeling/sequence_labeling.py b/demo/ernie-seq-labeling/sequence_labeling.py index ef4b51e9663400f2267a2129857ede1d7d021a70..702801149e9cf082eb8da2019fa4dc6ecd57d91e 100644 --- a/demo/ernie-seq-labeling/sequence_labeling.py +++ b/demo/ernie-seq-labeling/sequence_labeling.py @@ -13,7 +13,8 @@ # limitations under the License. """Finetuning on sequence labeling task.""" -import paddle +import argparse + import paddle.fluid as fluid import paddlehub as hub @@ -21,7 +22,6 @@ import paddlehub as hub parser = argparse.ArgumentParser(__doc__) parser.add_argument("--num_epoch", type=int, default=3, help="Number of epoches for fine-tuning.") parser.add_argument("--learning_rate", type=float, default=5e-5, help="Learning rate used to train with warmup.") -parser.add_argument("--hub_module_dir", type=str, default=None, help="PaddleHub module directory") parser.add_argument("--weight_decay", type=float, default=0.01, help="Weight decay rate for L2 regularizer.") parser.add_argument("--checkpoint_dir", type=str, default=None, help="Directory to model checkpoint") parser.add_argument("--max_seq_len", type=int, default=512, help="Number of words of the longest seqence.")