AI开发平台MODELARTS-LLaVA模型基于DevServer适配PyTorch NPU预训练指导(6.3.906):Step4 准备训练环境

时间:2024-12-09 20:36:09

Step4 准备训练环境

  1. 获取LLaVA模型代码。
    cd ${container_work_dir}
    unzip AscendCloud-6.3.906-xxx.zip
    unzip AscendCloud-AIGC-6.3.906-xxx.zip
    cd multimodal_algorithm/LLAVA/llava-train/5d8f1760c08b7dfba3ae97b71cbd4c6f17d12dbd
    bash build.sh
    cd LLaVA
    mkdir ./playground/data/LLaVA-Pretrain
  2. 下载vicuna-13b-v1.5模型。下载地址:lmsys/vicuna-13b-v1.5 · Hugging Face
    图1 下载vicuna-13b-v1.5模型
support.huaweicloud.com/bestpractice-modelarts/modelarts_10_01207.html