m训练部署

conda config --show channels

conda config --add channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/
conda config --add channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free/

conda config --add channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/conda-forge/
conda config --add channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/pytorch/

conda config --set show_channel_urls yes

conda config --set ssl_verify false

conda config --remove channels conda-forge

conda clean -i
conda config --remove-key channels 

conda clean --all -y

conda create -n llama_factory python=3.10
conda activate llama_factory
mkdir project
cd project

git clone --depth 1 https://github.com/hiyouga/LLaMA-Factory.git
cd LLaMA-Factory

git fetch --tags

git checkout -b v0.9.3 tags/v0.9.3

pip install -e ".[torch,metrics,liger-kernel,bitsandbytes,vllm,modelscope]" 

quickstart


llama-factory

docker build -f ./docker/docker-cuda/Dockerfile \
    --build-arg PIP_INDEX=https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple \
    --build-arg EXTRAS=metrics \
    -t llamafactory:latest .

docker build -f ./docker/docker-cuda/Dockerfile     --build-arg PIP_INDEX=https://pypi.org/simple     --build-arg EXTRAS=metrics     -t llamafactory:latest .

docker run -dit --ipc=host      -p 7860:7860     -p 8000:8000     --name llamafactory     llamafactory:latest
docker exec -it llamafactory bash

# 给如下目录赋予权限 chmod -R 777 /app/data 
# 启动web服务，命令如下： nohup llamafactory-cli webui > /app/webui.log 2>&1 &
访问 http://localhost:7860/

https://zhuanlan.zhihu.com/p/695287607

docker 镜像下载
https://www.rockylinux.cn/notes/docker-image-accelerator-and-configuration-guide.html
试用资源

https://gallery.pai-ml.com/#/preview/deepLearning/nlp/llama_factory

 

微调8bit量化配置

CUDA_VISIBLE_DEVICES=0 llamafactory-cli train \
  --model_name_or_path meta-llama/Meta-Llama-3-8B \
  --stage sft \
  --dataset your_dataset \
  --quantization_bit 8 \
  --quantization_method bitsandbytes \
  --finetuning_type lora \
  --lora_target q_proj,v_proj \
  --lora_rank 8 \
  --lora_alpha 32 \
  --per_device_train_batch_size 1 \
  --max_source_length 256 \
  --gradient_checkpointing \
  --flash_attn \
  --output_dir ./output

CUDA_VISIBLE_DEVICES=0 llamafactory-cli train \
    --stage sft \
    --do_train \
    --model_name_or_path /root/.cache/modelscope/hub/models/LLM-Research/Meta-Llama-3-8B-Instruct \
    --dataset alpaca_gpt4_zh,identity,adgen_local \
    --dataset_dir ./data \
    --template llama3 \
    --finetuning_type lora \
    --output_dir ./saves/LLaMA3-8B/lora/sft \
    --quantization_bit 4 \
    --overwrite_cache \
    --overwrite_output_dir \
    --cutoff_len 1024 \
    --preprocessing_num_workers 16 \
    --per_device_train_batch_size 2 \
  --per_device_eval_batch_size 1 \
  --gradient_accumulation_steps 8 \
  --lr_scheduler_type cosine \
  --logging_steps 50 \
  --warmup_steps 20 \
  --save_steps 100 \
  --eval_steps 50 \
  --fp16 True \
  --use_unsloth

　　

 

watch -n 1 nvidia-smi

## 清洗
python WikiExtractor.py --infn  zhwiki-20250720-pages-articles-multistream.xml.bz2