llama3-sft 提交脚本

#!/bin/bash

pip install transformers==4.41.2 -i https://mirrors.aliyun.com/pypi/simple/

pip install flash-attn==2.6.1 -i https://mirrors.aliyun.com/pypi/simple/

pip install tensorboard -i https://mirrors.aliyun.com/pypi/simple/

export NCCL_BLOCKING_WAIT=1
export NCCL_COMM_TIMEOUT=3600   # 将超时时长设置为3600秒（即1小时）
export WANDB_MODE=offline
export WANDB_SILENT=true

cd /LLaMA-Factory
python -m torch.distributed.run \
    --master_addr $MASTER_ADDR \
    --master_port $MASTER_PORT \
    --nproc_per_node $NPROC_PER_NODE \
    --nnodes $WORLD_SIZE \
    --node_rank $RANK \
    src/train.py \
    examples/full_multi_gpu/llama3p2_1B_part_sft_multi_TaskQueryTranslateEsFrPt.yaml

# llama3p2_1B_fu