improvement(service): rename feature_store.py to store.py and move to services directory (#438)

* Rename feature_store.py to store.py and move to services directory

---------

Co-authored-by: openhands <[email protected]>

Author: tpoisonooo

README.md

@@ -63,7 +63,7 @@ The Web version's API for Android also supports other devices. See [Python sampl
 - \[2025/03\] Simplify deployment and removing `--standalone`
 - \[2025/03\] [Forwarding multiple wechat group message](./docs/zh/doc_merge_wechat_group.md)
 - \[2024/09\] [Inverted indexer](https://github.com/InternLM/HuixiangDou/pull/387) makes LLM prefer knowledge base🎯
-- \[2024/09\] [Code retrieval](./huixiangdou/service/parallel_pipeline.py)
+- \[2024/09\] [Code retrieval](./huixiangdou/services/parallel_pipeline.py)
 - \[2024/08\] [chat_with_readthedocs](https://huixiangdou.readthedocs.io/en/latest/), see [how to integrate](./docs/zh/doc_add_readthedocs.md) 👍
 - \[2024/07\] Image and text retrieval & Removal of `langchain` 👍
 - \[2024/07\] [Hybrid Knowledge Graph and Dense Retrieval](./docs/en/doc_knowledge_graph.md) improve 1.7% F1 score 🎯
@@ -136,7 +136,7 @@ The Web version's API for Android also supports other devices. See [Python sampl
 - Dense for Document
 - Sparse for Code 
 - [Knowledge Graph](./docs/en/doc_knowledge_graph.md)
-- [Internet Search](./huixiangdou/service/web_search.py)
+- [Internet Search](./huixiangdou/services/web_search.py)
 - [SourceGraph](https://sourcegraph.com)
 - Image and Text
 
@@ -211,14 +211,14 @@ cp -rf resource/data* repodir/
 
 # Build knowledge base, this will save the features of repodir to workdir, and update the positive and negative example thresholds into `config.ini`
 mkdir workdir
-python3 -m huixiangdou.service.feature_store
+python3 -m huixiangdou.services.store
 ```
 
 ## III. Setup LLM API and test
 Set the model and `api-key` in `config.ini`. If running LLM locally, we recommend using `vllm`.
 
 ```text
-vllm serve /path/to/Qwen-2.5-7B-Instruct --enable-prefix-caching --served-model-name Qwen-2.5-7B-Instruct
+vllm serve /path/to/Qwen-2.5-7B-Instruct --served-model-name vllm --enable-prefix-caching --served-model-name Qwen-2.5-7B-Instruct
 ```
 
 Here is an example of the configured `config.ini`:
@@ -327,7 +327,7 @@ apt update
 apt install python-dev libxml2-dev libxslt1-dev antiword unrtf poppler-utils pstotext tesseract-ocr flac ffmpeg lame libmad0 libsox-fmt-mp3 sox libjpeg-dev swig libpulse-dev
 python3 -m pip install -r requirements-cpu.txt
 # Establish knowledge base
-python3 -m huixiangdou.service.feature_store --config_path config-cpu.ini
+python3 -m huixiangdou.services.store --config_path config-cpu.ini
 # Q&A test
 python3 -m huixiangdou.main --config_path config-cpu.ini
 # gradio UI

README_zh.md

@@ -63,7 +63,7 @@ Web 版给 android 的接口，也支持非 android 调用，见[python 样例
 - \[2025/03\] 简化运行流程，移除 `--standalone`
 - \[2025/03\] [在多个微信群中转发消息](./docs/zh/doc_merge_wechat_group.md)
 - \[2024/09\] [倒排索引](https://github.com/InternLM/HuixiangDou/pull/387)让 LLM 更偏向使用领域知识 🎯
-- \[2024/09\] 稀疏方法实现[代码检索](./huixiangdou/service/parallel_pipeline.py)
+- \[2024/09\] 稀疏方法实现[代码检索](./huixiangdou/services/parallel_pipeline.py)
 - \[2024/08\] ["chat_with readthedocs"](https://huixiangdou.readthedocs.io/zh-cn/latest/) ，见[集成说明](./docs/zh/doc_add_readthedocs.md)
 - \[2024/07\] 图文检索 & 移除 `langchain` 👍
 - \[2024/07\] [混合知识图谱和稠密检索，F1 提升 1.7%](./docs/zh/doc_knowledge_graph.md) 🎯
@@ -136,7 +136,7 @@ Web 版给 android 的接口，也支持非 android 调用，见[python 样例
 
 - 文档用稠密，代码用稀疏
 - [知识图谱](./docs/zh/doc_knowledge_graph.md)
-- [联网搜索](./huixiangdou/service/web_search.py)
+- [联网搜索](./huixiangdou/services/web_search.py)
 - [SourceGraph](https://sourcegraph.com)
 - 图文混合
 
@@ -210,13 +210,13 @@ cp -rf resource/data* repodir/
 
 # 建立知识库，repodir 的特征会保存到 workdir，拒答阈值也会自动更新进 `config.ini`
 mkdir workdir
-python3 -m huixiangdou.service.feature_store
+python3 -m huixiangdou.services.store
 ```
 
 ## 三、配置 LLM，运行测试
 设置 `config.ini` 中的模型和 api-key。如果本地运行 LLM，我们推荐使用 `vllm`
 ```text
-vllm serve /path/to/Qwen-2.5-7B-Instruct --enable-prefix-caching --served-model-name Qwen-2.5-7B-Instruct
+vllm serve /path/to/Qwen-2.5-7B-Instruct --served-model-name vllm --enable-prefix-caching --served-model-name Qwen-2.5-7B-Instruct
 ```
 
 配置好的 `config.ini` 样例如下：
@@ -322,7 +322,7 @@ apt update
 apt install python-dev libxml2-dev libxslt1-dev antiword unrtf poppler-utils pstotext tesseract-ocr flac ffmpeg lame libmad0 libsox-fmt-mp3 sox libjpeg-dev swig libpulse-dev
 python3 -m pip install -r requirements-cpu.txt
 # 建立知识库
-python3 -m huixiangdou.service.feature_store  --config_path config-cpu.ini
+python3 -m huixiangdou.services.store --config_path config-cpu.ini
 # 问答测试
 python3 -m huixiangdou.main --config_path config-cpu.ini
 # gradio UI

docs/en/doc_architecture.md

@@ -30,7 +30,7 @@ The module contains only 3 parts:
 .
 ├── frontend # Frontends like Lark, WeChat, etc., are part of the algorithm
 ├── main.py # main provides an example program
-├── service # service is the implementation of the algorithm
+├── services # services is the implementation of the algorithm
 ```
 
 **service** In our [paper](https://arxiv.org/abs/2401.08772), we introduced HuixiangDou as a pipeline structure; in implementation, it may include functions, a local LLM, or an RPC. All these foundational capabilities are regarded as services.
@@ -45,7 +45,7 @@ This is where the main body of the HuixiangDou pipeline is.
 
 ```bash
 .
-├── feature_store.py # Manages the creation and query of text features. In the future, "creation" and "query" will be separated
+├── store.py # Manages the creation and query of text features. In the future, "creation" and "query" will be separated
 ├── helper.py # Contains some helper tools
 ├── llm_client.py # LLM might be an RPC, so a client is needed
 ├── llm_server_hybrid.py # There might be more than one LLM, hence the name hybrid
@@ -54,7 +54,7 @@ This is where the main body of the HuixiangDou pipeline is.
 └── worker.py # The main logic as mentioned in the paper, calling the components above
 ```
 
-**1. feature_store.py** In the era of facial recognition, the storage and retrieval of facial features are called a feature store, which is the origin of the name.
+**1. store.py** In the era of facial recognition, the storage and retrieval of facial features are called a feature store, which is the origin of the name.
 
 1. When extracting features, the text will be partitionally split (the construction technique affects accuracy), the text2vec model extracts features, and saves them locally;
 2. During retrieval, in addition to directly using text2vec matching, a re-rank model will adjust the order

docs/en/doc_knowledge_graph.md

@@ -52,10 +52,10 @@ python3 -m huixiangdou.service.kg --help
 
 ## 3. Build Dense Retrieval Feature Library
 
-This step is the `feature_store` in the README. Since you need to calculate the optimal threshold under hybrid retrieval, do not skip it.
+This step is the `store` in the README. Since you need to calculate the optimal threshold under hybrid retrieval, do not skip it.
 
 ```bash
-python3 -m huixiangdou.service.feature_store
+python3 -m huixiangdou.services.store
 ```
 
 Test it.

docs/zh/doc_architecture.md

@@ -30,22 +30,22 @@ module 内只有 3 个部分：
 .
 ├── frontend       # 飞书、微信这些，都是茴香豆算法的前端
 ├── main.py        #  main 提供示例程序
-├── service        # service 就是算法实现
+├── services       # services 就是算法实现
 ```
 
-**service** 我们在[论文](https://arxiv.org/abs/2401.08772)里介绍豆哥是套 pipeline。在实现里，可能包含函数、本地 LLM 或者 RPC。把这些基础能力都视做 service。
+**services** 我们在[论文](https://arxiv.org/abs/2401.08772)里介绍豆哥是套 pipeline。在实现里，可能包含函数、本地 LLM 或者 RPC。把这些基础能力都视做 service。
 
 **frontend** 既然豆哥是套算法 pipeline，那么微信、飞书、web 这些，都是它的前端。这个目录放调用前端的工具类和函数，目前里面是飞书的 API 用法
 
 **main.py** 现在有算法、有前端，需要个入口函数实现业务逻辑。你在 `config.ini` 配置了飞书，就应该发给飞书 qaq
 
-## 第三层：service
+## 第三层：services
 
 这里是 HuixiangDou 算法主体。
 
 ```bash
 .
-├── feature_store.py # 管理文本特征的建立和查询，未来会把 “建立” 和 “查询” 分开
+├── store.py # 管理文本特征的建立和查询，未来会把 “建立” 和 “查询” 分开
 ├── helper.py        # 放一些辅助工具
 ├── llm_client.py    # LLM 可能是个 RPC，所以需要个 client
 ├── llm_server_hybrid.py  # LLM 可能不止一个，所以是 hybrid
@@ -54,7 +54,7 @@ module 内只有 3 个部分：
 └── worker.py        # 论文所说的主逻辑，调用上面的组件
 ```
 
-**1. feature_store.py** 人脸识别时代，面部特征的存储和检索叫 feature_store，这是名字来源。
+**1. store.py** 人脸识别时代，面部特征的存储和检索叫 feature_store，这是名字来源。
 
     - 提取特征时，会花式分割文本（构造技巧会影响精度）、text2vec 模型提取特征、保存到本地；
     - 检索时，除了直接用 text2vec 匹配，还会 rerank 模型调整顺序

docs/zh/doc_knowledge_graph.md

@@ -48,10 +48,10 @@ python3 -m huixiangdou.service.kg --dump-neo4j --neo4j-uri ${URI} --neo4j-user $
 
 ## 三、建立稠密检索特征库
 
-这步就是 README 里的 `feature_store`，因为要算混合检索下的最佳阈值，不要跳过。
+这步就是 README 里的 `store`，因为要算混合检索下的最佳阈值，不要跳过。
 
 ```bash
-python3 -m huixiangdou.service.feature_store
+python3 -m huixiangdou.services.store
 ```
 
 测试效果

docs/zh/doc_rag_annotate_sft_data.md

@@ -15,7 +15,7 @@ RAG 标注训练数据是否有用，请参考论文：
 
 基于 [config-alignment-example.json](../../config-alignment-example.ini) 做几处修改：
 
-1. 设置 bce 模型路径，执行 `python3 -m huixiangdou.service.feature_store --config_path config-alignment-example.ini`，用自己的知识库提取特征
+1. 设置 bce 模型路径，执行 `python3 -m huixiangdou.services.store --config_path config-alignment-example.ini`，用自己的知识库提取特征
 2. 配置 config 中网络搜索 key
 3. 配置 sourcegraph key。可能需要私有化部署一套 sourcegraph
 4. 选择使用的 remote LLM 并配置 RPM，一般标注用 GPT。xi-api 是国内的一个代理

evaluation/end2end/main.py

@@ -1,4 +1,4 @@
-from huixiangdou.service import ParallelPipeline
+from huixiangdou.services import ParallelPipeline
 from huixiangdou.primitive import Query
 import json
 import asyncio

evaluation/rejection/build_fs_and_filter.py

@@ -10,7 +10,7 @@
 from sklearn.metrics import f1_score, precision_score, recall_score
 from tqdm import tqdm
 
-from huixiangdou.service import CacheRetriever, FeatureStore
+from huixiangdou.services import CacheRetriever, FeatureStore
 from huixiangdou.primitive import FileOperation
 save_hardcase = False
 

evaluation/rejection/kg_filter.py

@@ -10,7 +10,8 @@
 from sklearn.metrics import f1_score, precision_score, recall_score
 from tqdm import tqdm
 
-from huixiangdou.service import KnowledgeGraph, histogram
+from huixiangdou.services import KnowledgeGraph
+from huixiangdou.services import histogram
 
 
 def load_dataset():