Fix batch prediction bug and FAQ README of Pipelines (PaddlePaddle#3712)

* Fix batch prediction bug of Pipelines * Change faq dir of pipelines
huchuqi12 · Nov 10, 2022 · 72cc9e1 · 72cc9e1
1 parent bbf0c39
commit 72cc9e1
Show file tree

Hide file tree

Showing 3 changed files with 7 additions and 9 deletions.
diff --git a/pipelines/examples/FAQ/Install_windows.md b/pipelines/examples/FAQ/Install_windows.md
@@ -26,9 +26,9 @@ python setup.py install
 
 ```bash
 # 我们建议在 GPU 环境下运行本示例，运行速度较快
-python examples/frequently-asked-question/dense_faq_example.py --device gpu
+python examples/FAQ/dense_faq_example.py --device gpu
 # 如果只有 CPU 机器，安装CPU版本的Paddle后，可以通过 --device 参数指定 cpu 即可, 运行耗时较长
-python examples/frequently-asked-question/dense_faq_example.py --device cpu
+python examples/FAQ/dense_faq_example.py --device cpu
 ```
 
 ### 1.4 构建 Web 可视化FAQ系统

diff --git a/pipelines/examples/FAQ/README.md b/pipelines/examples/FAQ/README.md
@@ -66,10 +66,10 @@ FAQ智能问答数据库的数据来自于[8000 多条保险行业问答数据](
 # 我们建议在 GPU 环境下运行本示例，运行速度较快
 # 设置 1 个空闲的 GPU 卡，此处假设 0 卡为空闲 GPU
 export CUDA_VISIBLE_DEVICES=0
-python examples/frequently-asked-question/dense_faq_example.py --device gpu
+python examples/FAQ/dense_faq_example.py --device gpu
 # 如果只有 CPU 机器，可以通过 --device 参数指定 cpu 即可, 运行耗时较长
 unset CUDA_VISIBLE_DEVICES
-python examples/frequently-asked-question/dense_faq_example.py --device cpu
+python examples/FAQ/dense_faq_example.py --device cpu
 ```
 `dense_faq_example.py`中`DensePassageRetriever`和`ErnieRanker`的模型介绍请参考[API介绍](../../API.md)
 
@@ -136,7 +136,7 @@ python rest_api/application.py 8891
 Linux 用户推荐采用 Shell 脚本来启动服务：：
 
 ```bash
-sh examples/frequently-asked-question/run_faq_server.sh
+sh examples/FAQ/run_faq_server.sh
 ```
 启动后可以使用curl命令验证是否成功运行：
 
@@ -155,7 +155,7 @@ python -m streamlit run ui/webapp_faq.py --server.port 8502
 Linux 用户推荐采用 Shell 脚本来启动服务：：
 
 ```bash
-sh examples/frequently-asked-question/run_faq_web.sh
+sh examples/FAQ/run_faq_web.sh
 ```
 
 到这里您就可以打开浏览器访问 http://127.0.0.1:8502 地址体验FAQ智能问答系统服务了。

diff --git a/pipelines/pipelines/nodes/ranker/ernie_ranker.py b/pipelines/pipelines/nodes/ranker/ernie_ranker.py
@@ -187,8 +187,7 @@ def predict_batch(
             right_idx = 0
             for number in number_of_docs:
                 right_idx = left_idx + number
-                grouped_predictions.append(
-                    similarity_scores[left_idx:right_idx])
+                grouped_predictions.append(preds[left_idx:right_idx])
                 left_idx = right_idx
             result = []
             for pred_group, doc_group in zip(grouped_predictions, documents):
@@ -202,7 +201,6 @@ def predict_batch(
                     doc for _, doc in sorted_scores_and_documents
                 ]
                 result.append(sorted_documents[:top_k])
-
             return result
 
     def _preprocess_batch_queries_and_docs(