llama-factory

1.在启智平台上找到没有安装tensorflow的镜像作为基础镜像

把llama-factory的github仓库进行下载，得到zip压缩包，上传到启智平台中，如下：

2. 执行命令如下

进入文件夹

cd LLaMA-Factory-main

更新pip

python -m pip install --upgrade pip

安装依赖：

pip install -e '.[torch,metrics]' -i pypi.tuna.tsinghua.edu /simple/

解决依赖包冲突：

pip install --no-deps -e

进行环境验证：

lamafactory-cli train -h

输出：

oot@i0435935b1bb4582a32b2a2767606073-task0-0:/tmp/code/cats2/LLaMA-Factory-main# lamafactory-cli train -h bash: lamafactory-cli: command not found root@i0435935b1bb4582a32b2a2767606073-task0-0:/tmp/code/cats2/LLaMA-Factory-main# llamafactory-cli train -h usage: llamafactory-cli [-h] [--ray_run_name RAY_RUN_NAME] [--ray_storage_path RAY_STORAGE_PATH] [--ray_num_workers RAY_NUM_WORKERS] [--resources_per_worker RESOURCES_PER_WORKER] [--placement_strategy {SPREAD,PACK,STRICT_SPREAD,STRICT_PACK}] options: -h, --help show this help message and exit --ray_run_name RAY_RUN_NAME, --ray-run-name RAY_RUN_NAME The training results will be saved at `<ray_storage_path>/ray_run_name`. (default: None) --ray_storage_path RAY_STORAGE_PATH, --ray-storage-path RAY_STORAGE_PATH The storage path to save training results to (default: ./saves) --ray_num_workers RAY_NUM_WORKERS, --ray-num-workers RAY_NUM_WORKERS The number of workers for Ray training. Default is 1 worker. (default: 1) --resources_per_worker RESOURCES_PER_WORKER, --resources-per-worker RESOURCES_PER_WORKER The resources per worker for Ray training. Default is to use 1 GPU per worker. (default: {'GPU': 1}) --placement_strategy {SPREAD,PACK,STRICT_SPREAD,STRICT_PACK}, --placement-strategy {SPREAD,PACK,STRICT_SPREAD,STRICT_PACK} The placement strategy for Ray training. Default is PACK. (default: PACK)