Artificial Intelligence

nvidia 驱动：https://wiki.debian.org/NvidiaGraphicsDrivers & https://www.nvidia.cn/drivers/lookup/ & https://docs.nvidia.com/datacenter/tesla/driver-installation-guide/debian.html & https://cnblogs.com/niuben/p/18397121

sudo apt remove --autoremove --purge -V \*nvidia\* libxnvctrl\* \*nvidia\*:i386
sudo apt install linux-headers-$(uname -r)
sudo dpkg -i nvidia-driver-local-repo-$distro-$version*_$arch.deb
sudo cp /var/nvidia-driver-local-repo-$distro-$version/nvidia-driver-*-keyring.gpg /usr/share/keyrings/
sudo dpkg -i nvidia-driver-local-repo-$distro-$version*_$arch.deb
sudo apt update
sudo apt -V install cuda-drivers
sudo apt -V install nvidia-driver-cuda nvidia-kernel-dkms
sudo reboot
sudo systemctl restart nvidia-persistenced
cat /proc/driver/nvidia/version
apt remove --purge nvidia-driver-local-repo\*
watch -n 1 nvidia-smi

nvcc：https://developer.nvidia.com/cuda-toolkit-archive & https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html#post-installation-actions & https://docs.nvidia.com/cuda/cuda-quick-start-guide/index.html

sudo apt remove --autoremove --purge "*cuda*" "*cublas*" "*cufft*" "*cufile*" "*curand*" "*cusolver*" "*cusparse*" "*gds-tools*" "*npp*" "*nvjpeg*" "nsight*" "*nvvm*" "*nvptx*"
curl -LOJ  https://developer.download.nvidia.com/compute/cuda/12.9.1/local_installers/cuda-repo-debian12-12-9-local_12.9.1-575.57.08-1_amd64.deb
sudo dpkg -i cuda-repo-debian12-12-9-local_12.9.1-575.57.08-1_amd64.deb
sudo cp /var/cuda-repo-debian12-12-9-local/cuda-*-keyring.gpg /usr/share/keyrings/
sudo apt update
sudo apt -y install cuda-toolkit-12-9
apt list | grep cuda-repo-
sudo apt remove --purge "cuda-repo-<distro>-X-Y-local*"
export PATH=${PATH}:/usr/local/cuda/bin
export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:/usr/local/cuda/lib64
nvcc -v

NVIDIA Container Toolkit：https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/latest/install-guide.html

 

Python：https://cnblogs.com/jhxxb/p/11288575.html

[project]
name = "Qwen3-VL"
version = "0.0.1"
requires-python = "==3.12.*"
dependencies = [
#    "torch",
#    "torchvision",

#    "transformers",
    #"datasets",
    #"evaluate",
#    "accelerate",
#    "transformers-stream-generator",

    #"triton-windows",
    #"sentence_transformers",
    #"faiss-gpu",
    #"chromadb",
    #"pymilvus",
    "vllm>=0.11.0",
    "qwen-vl-utils>=0.0.14",
    "modelscope",

#    "pillow",
#    "addict",
#    "matplotlib",
#    "easydict",
#    "einops",

#    "gradio",
#    "gradio_client",
]

[tool.uv.sources]
torch = { index = "pytorch" }
torchvision = { index = "pytorch" }
torchaudio = { index = "pytorch" }

[[tool.uv.index]]
name = "pytorch"
#url = "https://download.pytorch.org/whl/cu130"
url = "https://mirror.nju.edu.cn/pytorch/whl/cu129"
#url = "https://mirror.sjtu.edu.cn/pytorch-wheels/cu130"
# https://developer.aliyun.com/mirror/pytorch-wheels
#url = "https://mirrors.aliyun.com/pytorch-wheels/cu130"
explicit = true

[tool.uv]
package = false
no-build = true

 

Model

https://huggingface.co

import os
# os.environ['http_proxy'] = 'http://127.0.0.1:10808'
# os.environ['https_proxy'] = 'http://127.0.0.1:10808'
# https://huggingface.co/docs/huggingface_hub/package_reference/environment_variables
os.environ['HF_ENDPOINT'] = 'https://hf-mirror.com'
# os.environ['HF_ENDPOINT'] = 'https://aifasthub.com'
# os.environ['HF_ENDPOINT'] = 'https://hf-cdn.sufy.com'
# os.environ['HF_ENDPOINT'] = 'https://huggingface.lolicp.com'
os.environ['HF_HUB_CACHE'] = 'HF_HOME'
os.environ['HF_DEBUG'] = 'TRUE'
# os.environ['HF_HUB_OFFLINE'] = 'TRUE'

https://modelscope.cn/models

pip install modelscope
modelscope download --model Qwen/Qwen3-VL-2B-Instruct --local_dir Qwen/Qwen3-VL-2B-Instruct
modelscope download --model deepseek-ai/DeepSeek-OCR --local_dir deepseek-ai/DeepSeek-OCR
CUDA_VISIBLE_DEVICES=0,1 vllm serve Qwen/Qwen3-VL-2B-Instruct --host 0.0.0.0 --port 8000 --max-model-len 44640 --tensor-parallel-size 2

https://aistudio.baidu.com/modelsoverview

 

vLLM

https://docs.vllm.ai/projects/recipes & https://docs.vllm.com.cn

 

llama.cpp/Ollama

https://ggml.ai

 

TensorRT/onnxruntime + Triton inference server

 

UI

https://docs.openwebui.com/getting-started/quick-start/

docker volume rm open-webui
docker rm -f open-webui
docker run -d -p 3000:8080 --gpus all --name open-webui -e WEBUI_AUTH=False -e OPENAI_API_BASE_URL=http://192.168.1.130:8000/v1 -v open-webui:/app/backend/data ghcr.nju.edu.cn/open-webui/open-webui:cuda
docker logs -f open-webui

https://labelstud.io/guide/start & https://labelstud.io/guide/ml_create

label-studio start -b --port 80 --data-dir D:\label-studio --username l@l.ll --password l

 

---

https://pytorch.org/get-started/locally/

https://www.paddlepaddle.org.cn & https://paddleocr.ai & https://paddlenlp.readthedocs.io & https://paddlepaddle.github.io/PaddleX