Skip to content

DiFanrui/InfiniRepo

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

3 Commits
 
 
 
 
 
 
 
 

Repository files navigation

InfiniLM 个人优化项目

赛道 T2-1-2 服务能力优化 — InfiniLM 推理引擎性能优化。

目录结构

InfiniLM/     # InfiniTensor/InfiniLM 推理引擎
InfiniCore/   # InfiniTensor/InfiniCore C++ 核心库

快速搭建

详细步骤见 InfiniLM/docs_extern/环境搭建指南.md

速览

# 1. 确认 CUDA 版本匹配
nvidia-smi | grep "CUDA Version"   # 看驱动支持的最高版本
nvcc --version                      # 看 Toolkit 版本
# 两者必须匹配,不匹配则降级 Toolkit

# 2. 初始化第三方库
cd InfiniCore
git submodule update --init third_party/spdlog third_party/nlohmann_json
git clone --depth 1 https://github.com/NVIDIA/cutlass.git third_party/cutlass

# 3. 编译 InfiniCore(GPU 后端)
python3 scripts/install.py --nv-gpu=y --cuda=/usr/local/cuda

# 4. 安装 InfiniLM
cd ../InfiniLM
python3 -m venv .venv && source .venv/bin/activate
pip install -e .

# 5. 安装 InfiniCore Python 包 + 编译 Python 绑定
cd ../InfiniCore
pip install -e .
XMAKE_ROOT=y xmake build -y _infinicore
mkdir -p python/infinicore/lib/ && touch python/infinicore/lib/__init__.py
cp build/linux/x86_64/release/_infinicore.cpython-*.so python/infinicore/lib/

# 6. 安装 Python 依赖
pip install torch --index-url https://download.pytorch.org/whl/cu124
pip install numpy Pillow safetensors transformers modelscope uvicorn fastapi vllm

# 7. 下载模型
modelscope download --model LLM-Research/Meta-Llama-3.1-8B-Instruct \
  --local_dir InfiniLM/models/Llama-3.1-8B-Instruct

# 8. 启动服务 + 验证
cd InfiniLM
python python/infinilm/server/inference_server.py \
  --device nvidia --model=models/Llama-3.1-8B-Instruct \
  --port 8102 --tp 1 --max-new-tokens 4096 \
  --num-blocks 64 --max-batch-size 4 --enable-graph --enable-paged-attn

# 另开终端
curl http://localhost:8102/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"messages":[{"role":"user","content":"Hi"}],"max_tokens":10}'

在当前机器 Push

# 先在 GitHub 创建新 repo,然后:
cd /root/difanrui
git remote add origin https://github.com/<your-username>/<your-repo>.git
git add -A
git commit -m "Initial commit: InfiniLM + InfiniCore with setup fixes"
git push -u origin main

代码改动

相对于原版 InfiniTensor/InfiniLMInfiniTensor/InfiniCore,本仓库做了以下修改:

文件 改动
InfiniLM/setup.py build_cpp_module()XMAKE_ROOT=y 环境变量
InfiniCore/scripts/install.py run_cmd()XMAKE_ROOT=yinstall() build 命令加 -j4
InfiniCore/setup.py run_xmake_build()XMAKE_ROOT=y

相关文档

About

infiniTensor推理框架的比赛仓库,包含内核Core和推理框架LM

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

No releases published

Packages

 
 
 

Contributors