feat: 文档调整 (#432)
* feat: 文档调整
* feat: 图片调整
diff --git a/docs/docs-cn/source/1.guide_cn.md b/docs/docs-cn/source/1.guide.md
similarity index 100%
rename from docs/docs-cn/source/1.guide_cn.md
rename to docs/docs-cn/source/1.guide.md
diff --git a/docs/docs-cn/source/3.quick_start/1.quick_start.md b/docs/docs-cn/source/3.quick_start/1.quick_start.md
index a34f120..cd34e8b 100644
--- a/docs/docs-cn/source/3.quick_start/1.quick_start.md
+++ b/docs/docs-cn/source/3.quick_start/1.quick_start.md
@@ -19,15 +19,19 @@
```
## 本地运行流图作业
+
下面介绍如何在本地环境运行一个实时环路查找的图计算作业。
-### Demo1 从本地文件读取数据[1.quick_start_copy.md](..%2F..%2F..%2Fdocs-en%2Fsource%2F3.quick_start%2F1.quick_start_copy.md)
+### Demo1 从本地文件读取数据[1.quick_start_copy.md]
+
1. 直接运行脚本即可:
+
```shell
bin/gql_submit.sh --gql geaflow/geaflow-examples/gql/loop_detection_file_demo.sql
```
其中 loop_detection_file_demo.sql 是一段实时查询图中所有四度环路的 DSL 计算作业，其内容如下：
+
```sql
set geaflow.dsl.window.size = 1;
set geaflow.dsl.ignore.exception = true;
@@ -102,11 +106,13 @@
RETURN a.id as a_id, b.id as b_id, c.id as c_id, d.id as d_id, a.id as a1_id
);
```
-该 DSL 会从项目中的resource文件 **demo_job_data.txt** 中读取点边数据，进行构图，然后计算图中所有的 4 度的环路, 并将环路上的点 id 输出到
+
+该 DSL 会从项目中的 resource 文件 **demo_job_data.txt** 中读取点边数据，进行构图，然后计算图中所有的 4 度的环路, 并将环路上的点 id 输出到
/tmp/geaflow/demo_job_result,
用户也可通过修改 `geaflow.dsl.file.path` 参数自定义输出路径。
2. 输出结果如下
+
```
2,3,4,1,2
4,1,2,3,4
@@ -114,15 +120,17 @@
1,2,3,4,1
```
-### Demo2 交互式使用socket读取数据
+### Demo2 交互式使用 socket 读取数据
+
用户也可自己在命令台输入数据，实时进行构图。
+
1. 运行脚本:
```shell
bin/gql_submit.sh --gql geaflow/geaflow-examples/gql/loop_detection_socket_demo.sql
```
-loop_detection_socket_demo.sql 主要区别是source表是通过socket进行读取：
+loop_detection_socket_demo.sql 主要区别是 source 表是通过 socket 进行读取：
```sql
CREATE TABLE IF NOT EXISTS tbl_source (
@@ -198,15 +206,15 @@
-4. 访问可视化dashboard页面
+4. 访问可视化 dashboard 页面
-本地模式的进程会占用本地的8090和8088端口，附带一个可视化页面。
+本地模式的进程会占用本地的 8090 和 8088 端口，附带一个可视化页面。
在浏览器中输入 http://localhost:8090 即可访问前端页面。
-关于更多dashboard相关的内容，请参考文档：
+关于更多 dashboard 相关的内容，请参考文档：
[文档](../7.deploy/3.dashboard.md)
## GeaFlow Console 快速上手
@@ -214,10 +222,11 @@
GeaFlow Console 是 GeaFlow 提供的图计算研发平台，我们将介绍如何在 Docker 容器里面启动 GeaFlow Console 平台，提交流图计算作业。文档地址：
[文档](2.quick_start_docker.md)
-## GeaFlow Kubernetes Operator快速上手
-Geaflow Kubernetes Operator是一个可以快速将Geaflow应用部署到kubernetes集群中的部署工具。
-我们将介绍如何通过Helm安装geaflow-kubernetes-operator，通过yaml文件快速提交geaflow作业，
-并访问operator的dashboard页面查看集群下的作业状态。文档地址：
+## GeaFlow Kubernetes Operator 快速上手
+
+Geaflow Kubernetes Operator 是一个可以快速将 Geaflow 应用部署到 kubernetes 集群中的部署工具。
+我们将介绍如何通过 Helm 安装 geaflow-kubernetes-operator，通过 yaml 文件快速提交 geaflow 作业，
+并访问 operator 的 dashboard 页面查看集群下的作业状态。文档地址：
[文档](../7.deploy/2.quick_start_operator.md)
## 使用 G6VP 进行流图计算作业可视化
diff --git a/docs/docs-en/source/7.deploy/5.install_llm.md b/docs/docs-en/source/7.deploy/5.install_llm.md
index 60d56ef..c660b82 100644
--- a/docs/docs-en/source/7.deploy/5.install_llm.md
+++ b/docs/docs-en/source/7.deploy/5.install_llm.md
@@ -1,13 +1,16 @@
# LLM Local Deployment
-The users have the capability to locally deploy extensive models as a service. The complete process, encompassing downloading pre-trained models, deploying them as a service, and debugging, is described in the following steps. It is essential for the user's machine to have Docker installed and be granted access to the repository containing these large models.
-
- ## Step 1: Download the Model File
- The pre-trained large model file has been uploaded to the [Hugging Face repository](https://huggingface.co/tugraph/CodeLlama-7b-GQL-hf). Please proceed with downloading and locally unzipping the model file.
-
- ## Step 2: Prepare the Docker Container Environment
+The users have the capability to locally deploy extensive models as a service. The complete process, encompassing downloading pre-trained models, deploying them as a service, and debugging, is described in the following steps. It is essential for the user's machine to have Docker installed and be granted access to the repository containing these large models.
+
+## Step 1: Download the Model File
+
+The pre-trained large model file has been uploaded to the [Hugging Face repository](https://huggingface.co/tugraph/CodeLlama-7b-GQL-hf). Please proceed with downloading and locally unzipping the model file.
+
+
+## Step 2: Prepare the Docker Container Environment
+
1. Run the following command on the terminal to download the Docker image required for model servicing:
-
+
```
docker pull tugraph/llam_infer_service:0.0.1
@@ -15,23 +18,25 @@
docker images
```
-
+
2. Run the following command to start the Docker container:
-
+
```
-docker run -it --name ${Container name} -v ${Local model path}:${Container model path} -p ${Local port}:${Container service port} -d ${Image name}
+docker run -it --name ${Container name} -v ${Local model path}:${Container model path} -p ${Local port}:${Container service port} -d ${Image name}
// Such as
docker run -it --name my-model-container -v /home/huggingface:/opt/huggingface -p 8000:8000 -d llama_inference_server:v1
// Check whether the container is running properly
-docker ps
+docker ps
```
Here, we map the container's port 8000 to the local machine's port 8000, mount the directory where the local model (/home/huggingface) resides to the container's path (/opt/huggingface), and set the container name to my-model-container.
## Step 3: Model Service Deployment
+
1. Model transformation
+
```
// Enter the container you just created
docker exec -it ${container_id} bash
@@ -40,11 +45,12 @@
cd /opt/llama_cpp
python3 ./convert.py ${Container model path}
```
+
When the execution is complete, a file with the prefix ggml-model is generated under the container model path.
-
+
2. Model quantization (optional)
-Take the llam2-7B model as an example: By default, the accuracy of the model converted by convert.py is F16 and the model size is 13.0GB. If the current machine resources cannot satisfy such a large model inference, the converted model can be further quantized by./quantize.
+ Take the llam2-7B model as an example: By default, the accuracy of the model converted by convert.py is F16 and the model size is 13.0GB. If the current machine resources cannot satisfy such a large model inference, the converted model can be further quantized by./quantize.
```
// As shown below, q4_0 quantizes the original model to int4 and compresses the model size to 3.5GB
@@ -52,11 +58,13 @@
cd /opt/llama_cpp
./quantize ${Default generated F16 model path} ${Quantized model path} q4_0
```
+
The following are reference indicators such as the size and reasoning speed of the quantized model:
-
+
3. Model servicing
-Run the following command to deploy the above generated model as a service, and specify the address and port of the service binding through the parameters:
+ Run the following command to deploy the above generated model as a service, and specify the address and port of the service binding through the parameters:
+
```
// ./server -h. You can view parameter details
// ${ggml-model...file} The file name prefixes the generated ggml-model
@@ -69,7 +77,7 @@
```
4. Debugging service
-Send an http request to the service address, where "prompt" is the query statement and "content" is the inference result.
+ Send an http request to the service address, where "prompt" is the query statement and "content" is the inference result.
```
curl --request POST \
@@ -77,8 +85,7 @@
--header "Content-Type: application/json" \
--data '{"prompt": "请返回小红的10个年龄大于20的朋友","n_predict": 128}'
```
+
Debugging service
The following is the model inference result after service deployment:
-
-
-
\ No newline at end of file
+
