Update README

README.md

@@ -20,29 +20,25 @@
 -v 路径、docker_name和imageID根据实际情况修改
 
 ### Docker（方法一）
-
 ```bash
 docker pull image.sourcefind.cn:5000/dcu/admin/base/pytorch:2.1.0-centos7.6-dtk24.04-py310
 
 docker run -it -v /path/your_code_data/:/path/your_code_data/ -v /opt/hyhal/:/opt/hyhal/:ro --shm-size=32G --privileged=true --device=/dev/kfd --device=/dev/dri/ --group-add video --name docker_name imageID bash
 
 cd /your_code_path/sentence-bert_pytorch
-pip install -e .
+pip install -r requirements.txt
 pip install -U huggingface_hub hf_transfer
 export HF_ENDPOINT=https://hf-mirror.com
 ```
 
 ### Dockerfile（方法二）
-
 ```bash
 cd ./docker
-cp ../requirements.txt requirements.txt
-
 docker build --no-cache -t sbert:latest .
 docker run -it -v /path/your_code_data/:/path/your_code_data/ -v /opt/hyhal/:/opt/hyhal/:ro --shm-size=32G --privileged=true --device=/dev/kfd --device=/dev/dri/ --group-add video --name docker_name imageID bash
 
 cd /your_code_path/sentence-bert_pytorch
-pip install -e .
+pip install -r requirements.txt
 pip install -U huggingface_hub hf_transfer
 export HF_ENDPOINT=https://hf-mirror.com
 ```
@@ -54,6 +50,7 @@ export HF_ENDPOINT=https://hf-mirror.com
 DTK软件栈：dtk24.04
 python：python3.10
 torch：2.1.0
+torchvision: 0.16.0
 ```
 
 Tips：以上dtk软件栈、python、torch等DCU相关工具版本需要严格一一对应
@@ -62,7 +59,7 @@ Tips：以上dtk软件栈、python、torch等DCU相关工具版本需要严格
 
 ```bash
 cd /your_code_path/sentence-bert_pytorch
-pip install -e .
+pip install -r requirements.txt
 pip install -U huggingface_hub hf_transfer
 export HF_ENDPOINT=https://hf-mirror.com
 ```
@@ -97,11 +94,11 @@ bash finetune.sh
 ### 单机单卡
 - 训练
 ```bash
-python training_stsbenchmark.py
+python training_stsbenchmark.py --train_batch_size 64 --num_epochs 5
 ```
 - 微调
 ```bash
-python training_stsbenchmark_continue_training.py
+python training_stsbenchmark_continue_training.py --train_batch_size 64 --num_epochs 5
 ```
 
 ## 推理
@@ -109,7 +106,7 @@ python training_stsbenchmark_continue_training.py
 2. 执行以下命令，测试数据默认为`./datasets/simple_wikipedia_v1/simple_wiki_pair.txt`，可修改`--data_path`参数为其他待测文件地址，文件内容格式请参考[simple_wiki_pair.txt](./datasets/simple_wikipedia_v1/simple_wiki_pair.txt)。
 
 ```bash
-python infer.py --data_path ./datasets/simple_wikipedia_v1/simple_wiki_pair.txt --model_name_or_path all-MiniLM-L6-v2
+python infer.py --data_path ./dataset/simple_wikipedia_v1/simple_wiki_pair.txt --model_name_or_path all-MiniLM-L6-v2
 ```
 
 ## result
@@ -119,7 +116,11 @@ python infer.py --data_path ./datasets/simple_wikipedia_v1/simple_wiki_pair.txt
 </div>
 
 ### 精度
-暂无
+在sts-test数据集上评估模型，Cosine-Similarity得分对比
+| device | backbone | epoch | Pearson | Spearman  |
+| :------: | :------: | :------: | :------: | :------: |
+| K100 | bert-base-uncased | 5 | 0.8500 | 0.8460 |
+| A800 | bert-base-uncased | 5 | 0.8449 | 0.8385 |
 
 ## 应用场景
 ### 算法类别

finetune.sh

@@ -7,4 +7,4 @@ export USE_MIOPEN_BATCHNORM=1
 
 echo "Training start ..."
 
-torchrun --nproc_per_node=4 training_stsbenchmark_continue_training.py --train_batch_size 16 --num_epochs 5
+torchrun --nproc_per_node=4 training_stsbenchmark_continue_training.py --train_batch_size 64 --num_epochs 5

requirements.txt

@@ -6,4 +6,5 @@ scipy
 huggingface-hub>=0.15.1
 Pillow
 datasets
-accelerate>=0.20.3
\ No newline at end of file
+accelerate>=0.20.3
+pandas==2.2.2
\ No newline at end of file

train.sh

@@ -7,4 +7,4 @@ export USE_MIOPEN_BATCHNORM=1
 
 echo "Training start ..."
 
-torchrun --nproc_per_node=4 training_stsbenchmark.py --train_batch_size 16 --num_epochs 5
+torchrun --nproc_per_node=4 training_stsbenchmark.py --train_batch_size 64 --num_epochs 5

training_stsbenchmark.py

@@ -19,8 +19,8 @@ parser.add_argument('--num_epochs', type=int, default=10)
 parser.add_argument('--model_name_or_path', type=str, default="bert-base-uncased")
 parser.add_argument('--save_root_path', type=str, default="output", help='Model output folder')
 parser.add_argument('--lr', default=2e-05)
-parser.add_argument('--eval_steps', type=int, default=100)
-parser.add_argument('--save_steps', type=int, default=100)
+parser.add_argument('--eval_steps', type=int, default=-1)
+parser.add_argument('--save_steps', type=int, default=-1)
 parser.add_argument('--save_total_limit', type=int, default=2)
 parser.add_argument('--logging_steps', type=int, default=10)
 args = parser.parse_args()

training_stsbenchmark_continue_training.py

-"""
-This example loads the pre-trained SentenceTransformer model 'nli-distilroberta-base-v2' from Hugging Face.
-It then fine-tunes this model for some epochs on the STS benchmark dataset.
-
-Note: In this example, you must specify a SentenceTransformer model.
-If you want to fine-tune a huggingface/transformers model like bert-base-uncased, see training_nli.py and training_stsbenchmark.py
-"""
-
 import logging
 import argparse
 from datetime import datetime
@@ -26,8 +18,8 @@ parser.add_argument('--num_epochs', type=int, default=10)
 parser.add_argument('--model_name_or_path', type=str, default="all-MiniLM-L6-v2")
 parser.add_argument('--save_root_path', type=str, default="output", help='Model output folder')
 parser.add_argument('--lr', default=2e-05)
-parser.add_argument('--eval_steps', type=int, default=100)
-parser.add_argument('--save_steps', type=int, default=100)
+parser.add_argument('--eval_steps', type=int, default=-1)
+parser.add_argument('--save_steps', type=int, default=-1)
 parser.add_argument('--save_total_limit', type=int, default=2)
 parser.add_argument('--logging_steps', type=int, default=10)
 args = parser.parse_args()
@@ -37,7 +29,7 @@ model_name = args.model_name_or_path
 train_batch_size = args.train_batch_size
 num_epochs = args.num_epochs
 output_dir = (
-    args.save_root_path + "training_stsbenchmark_" + model_name.replace("/", "-") + "-" + datetime.now().strftime("%Y-%m-%d_%H-%M-%S")
+    args.save_root_path + "/training_stsbenchmark_" + model_name.replace("/", "-") + "-" + datetime.now().strftime("%Y-%m-%d_%H-%M-%S")
 )
 
 # 1. Here we define our SentenceTransformer model.