Add extra_requires to reduce dependencies (#580)

* update reqs

* update docs

* resolve comments

* upgrade pydantic

* fix rebase

* update doc

* update

* update

* update readme

* update

* add flash-attn

README.md

@@ -103,6 +103,14 @@ Install lmdeploy with pip ( python 3.8+) or [from source](./docs/en/build.md)
 pip install lmdeploy
 ```
 
+> **Note**<br />
+> `pip install lmdeploy` can only install the runtime required packages. If users want to run codes from modules like `lmdeploy.lite` and `lmdeploy.serve`, they need to install the extra required packages.
+> For instance, running `pip install lmdeploy[lite]` would install extra dependencies for `lmdeploy.lite` module.
+>
+> - `all`: Install lmdeploy with all dependencies in `requirements.txt`
+> - `lite`: Install lmdeploy with extra dependencies in `requirements/lite.txt`
+> - `serve`: Install lmdeploy with dependencies in `requirements/serve.txt`
+
 ### Deploy InternLM
 
 #### Get InternLM model
@@ -140,6 +148,9 @@ lmdeploy chat turbomind ./workspace
 #### Serving with gradio
 
 ```shell
+# install lmdeploy with extra dependencies
+pip install lmdeploy[serve]
+
 lmdeploy serve gradio ./workspace
 ```
 
@@ -150,6 +161,9 @@ lmdeploy serve gradio ./workspace
 Launch inference server by:
 
 ```shell
+# install lmdeploy with extra dependencies
+pip install lmdeploy[serve]
+
 lmdeploy serve api_server ./workspace --instance_num 32 --tp 1
 ```
 
@@ -182,6 +196,7 @@ bash workspace/service_docker_up.sh
 Then, you can communicate with the inference server by command line,
 
 ```shell
+python3 -m pip install tritonclient[grpc]
 lmdeploy serve triton_client {server_ip_addresss}:33337
 ```
 

README_zh-CN.md

@@ -104,6 +104,13 @@ TurboMind 的 output token throughput 超过 2000 token/s, 整体比 DeepSpeed
 pip install lmdeploy
 ```
 
+> **Note**<br />
+> `pip install lmdeploy`默认安装runtime依赖包，使用lmdeploy的lite和serve功能时，用户需要安装额外依赖包。例如: `pip install lmdeploy[lite]` 会额外安装`lmdeploy.lite`模块的依赖包
+>
+> - `all`: 安装`lmdeploy`所有依赖包，具体可查看`requirements.txt`
+> - `lite`: 额外安装`lmdeploy.lite`模块的依赖包，具体可查看`requirements/lite.txt`
+> - `serve`: 额外安装`lmdeploy.serve`模块的依赖包，具体可查看`requirements/serve.txt`
+
 ### 部署 InternLM
 
 #### 获取 InternLM 模型
@@ -140,6 +147,9 @@ lmdeploy chat turbomind ./workspace
 #### 启动 gradio server
 
 ```shell
+# 安装lmdeploy额外依赖
+pip install lmdeploy[serve]
+
 lmdeploy serve gradio ./workspace
 ```
 
@@ -150,6 +160,9 @@ lmdeploy serve gradio ./workspace
 使用下面的命令启动推理服务：
 
 ```shell
+# 安装lmdeploy额外依赖
+pip install lmdeploy[serve]
+
 lmdeploy serve api_server ./workspace --server_name 0.0.0.0 --server_port ${server_port} --instance_num 32 --tp 1
 ```
 
@@ -182,6 +195,7 @@ bash workspace/service_docker_up.sh
 你可以通过命令行方式与推理服务进行对话：
 
 ```shell
+python3 -m pip install tritonclient[grpc]
 lmdeploy serve triton_client {server_ip_addresss}:33337
 ```
 

docs/en/faq.md

@@ -17,7 +17,7 @@ It may have been caused by the following reasons.
 1. You haven't installed lmdeploy's precompiled package. `_turbomind` is the pybind package of c++ turbomind, which involves compilation. It is recommended that you install the precompiled one.
 
 ```shell
-pip install lmdeploy
+pip install lmdeploy[all]
 ```
 
 2. If you have installed it and still encounter this issue, it is probably because you are executing turbomind-related command in the root directory of lmdeploy source code. Switching to another directory will fix it
@@ -26,7 +26,7 @@ pip install lmdeploy
 
 ### libnccl.so.2 not found
 
-Make sure you have install lmdeploy (>=v0.0.5) through `pip install lmdeploy`.
+Make sure you have install lmdeploy (>=v0.0.5) through `pip install lmdeploy[all]`.
 
 If the issue still exists after lmdeploy installation, add the path of `libnccl.so.2` to environment variable LD_LIBRARY_PATH.
 

docs/en/supported_models/codellama.md

@@ -26,7 +26,7 @@ Based on the above table, download the model that meets your requirements. Execu
 
 ```shell
 # install lmdeploy
-python3 -m pip install lmdeploy
+python3 -m pip install lmdeploy[all]
 
 # convert weight layout
 lmdeploy convert codellama /the/path/of/codellama/model

docs/en/w4a16.md

@@ -5,7 +5,7 @@ LMDeploy supports LLM model inference of 4-bit weight, with the minimum requirem
 Before proceeding with the inference, please ensure that lmdeploy is installed.
 
 ```shell
-pip install lmdeploy
+pip install lmdeploy[all]
 ```
 
 ## 4-bit LLM model Inference

docs/zh_cn/faq.md

@@ -17,7 +17,7 @@ pip install --upgrade mmengine
 1. 您没有安装 lmdeploy 的预编译包。`_turbomind`是 turbomind c++ 的 pybind部分，涉及到编译。推荐您直接安装预编译包。
 
 ```
-pip install lmdeploy
+pip install lmdeploy[all]
 ```
 
 2. 如果已经安装了，还是出现这个问题，请检查下执行目录。不要在 lmdeploy 的源码根目录下执行 python -m lmdeploy.turbomind.\*下的package，换到其他目录下执行。
@@ -26,7 +26,7 @@ pip install lmdeploy
 
 ### libnccl.so.2 not found
 
-确保通过 `pip install lmdeploy` 安装了 lmdeploy (>=v0.0.5)。
+确保通过 `pip install lmdeploy[all]` 安装了 lmdeploy (>=v0.0.5)。
 
 如果安装之后，问题还存在，那么就把`libnccl.so.2`的路径加入到环境变量 LD_LIBRARY_PATH 中。
 

docs/zh_cn/supported_models/codellama.md

@@ -26,7 +26,7 @@
 
 ```shell
 # 安装 lmdeploy
-python3 -m pip install lmdeploy
+python3 -m pip install lmdeploy[all]
 
 # 转模型格式
 lmdeploy convert codellama /path/of/codellama/model

docs/zh_cn/w4a16.md

@@ -5,7 +5,7 @@ LMDeploy 支持 4bit 权重模型的推理，**对 NVIDIA 显卡的最低要求
 在推理之前，请确保安装了 lmdeploy
 
 ```shell
-pip install lmdeploy
+pip install lmdeploy[all]
 ```
 
 ## 4bit 权重模型推理

requirements.txt

-accelerate
-datasets
-fastapi
-fire
-gradio<4.0.0
-mmengine
-numpy
-pybind11
-safetensors
-sentencepiece
-setuptools
-shortuuid
-tiktoken
-torch
-transformers>=4.33.0
-tritonclient[all]
-uvicorn
+-r requirements/build.txt
+-r requirements/runtime.txt
+-r requirements/lite.txt
+-r requirements/serve.txt

requirements/build.txt

+pybind11
+setuptools

requirements/lite.txt

+accelerate
+datasets
+flash-attn

requirements/runtime.txt

+fire
+mmengine
+numpy
+safetensors
+sentencepiece
+tiktoken
+torch
+transformers>=4.33.0

requirements/serve.txt

+fastapi
+gradio<4.0.0
+pydantic>2.0.0
+shortuuid
+uvicorn

requirements/test.txt

+coverage
+pytest

setup.py

@@ -134,7 +134,14 @@ if __name__ == '__main__':
             'lmdeploy': lmdeploy_package_data,
         },
         include_package_data=True,
-        install_requires=parse_requirements('requirements.txt'),
+        setup_requires=parse_requirements('requirements/build.txt'),
+        tests_require=parse_requirements('requirements/test.txt'),
+        install_requires=parse_requirements('requirements/runtime.txt'),
+        extras_require={
+            'all': parse_requirements('requirements.txt'),
+            'lite': parse_requirements('requirements/lite.txt'),
+            'serve': parse_requirements('requirements/serve.txt')
+        },
         has_ext_modules=check_ext_modules,
         classifiers=[
             'Programming Language :: Python :: 3.8',