KO
|
EN
gitlite — search
Search
#python
#java
#python3
#arduino
#golang
#machine-learning
#rust
#html
#flask
#javascript
#seismology
#nodejs
DeepLinkExt
★ 12
Open GitHub ↗
No description available.
Download README (.md)
Explore Similar Repositories
tedu-blog
:
TEDU CMS Blog
AI-Upscaling-Models
:
A collection of upscaling models I've trained.
Reminders
:
This project offers a unique approach to scheduling and managing tasks through a behavior-driven solution rather than relying on static and external services like traditional job schedulers.
Bluetooth-lyrics-speaker
:
ESP32S3-16M,LVGL,platform,lyrics speaker
Code-Llama-QA-Bot
:
This is a QA bot based on Code Llama. It is a new LLM for code by Meta AI.
// repository documentation
Was this content helpful?
★ 0
(0 ratings)
Select Rating:
★
★
★
★
★
Submit Feedback
Recent Feedback
×
Download README
Do you want to download the
README.md
file for
DeepLinkExt
?
Download (.md)
# DeepLinkExt 该仓库提供了一套在国产硬件上进行大模型训练、微调、推理的解决方案。对下调用[DIOPI](https://github.com/DeepLink-org/DIOPI)支持的大模型算子(如flash_attention和rms_norm),对上承接大模型训练、微调、推理框架;并对应提供了对应的大模型算子的torch组合实现。 DIOPI的具体内容请参考[DIOPI INTRODUCTION](https://deeplink.readthedocs.io/zh-cn/latest/doc/DIOPI/Introduction.html). 目前支持的框架以及对应的算子可以查看[框架与算子](https://github.com/DeepLink-org/DeepLinkExt/tree/main/deeplink_ext). ## Install DeepLinkExt依赖deeplink.framework/dipu,需要先完成dipu的编译安装,具体请参考[dipu quick_start](https://deeplink.readthedocs.io/zh-cn/latest/doc/DIPU/quick_start.html) 完成dipu的编译后,请参考如下代码,设置必要的环境变量。 ```bash export PYTHONPATH=$WORKDIR/deeplink.framework/dipu/:$PYTHONPATH export DIPU_ROOT=$WORKDIR/deeplink.framework/dipu/torch_dipu export DIOPI_PATH=$WORKDIR/deeplink.framework/dipu/third_party/DIOPI/proto export VENDOR_INCLUDE_DIRS=${PATH_TO_VENDOR_INCLUDE} # 底层软件栈的include路径,例如/usr/local/Ascend/ascend-toolkit/latest/include ``` 完成上述准备工作后,使用如下命令即可安装DeepLinkExt。提供两种安装实例如下: ### 1. inplace安装 ``` cd $WORKDIR/DeepLinkExt python3 setup.py build_ext --inplace ``` ### 2. 安装到指定目录 ```bash cd $WORKDIR/DeepLinkExt pip install -e . -t $TARGET_INSTALL_DIR ``` ## Usage 以InternEvo、LightLLM大模型训练框架为例,参考如下代码,即可实现在训练/推理时使用DeepLinkExt的扩展算子。 ### InternEvo DeepLinkExt已完全接入InternEvo,在完成DeepLinkExt的编译安装后,将其添加到PYTHONPATH,使用InternEvo进行训练即可。 ### LightLLM 对于LightLLM,在启动推理的脚本中,需要添加如下代码,即可实现使用DeepLinkExt扩展算子。 ```python import deeplink_ext.patch_lightllm.py import lightllm ```