LLaMA-Factory-Mirror/examples/lora_multi_gpu/ds_zero3.sh

16 lines
344 B
Bash

#!/bin/bash
NPROC_PER_NODE=4
NNODES=1
RANK=0
MASTER_ADDR=127.0.0.1
MASTER_PORT=29500
CUDA_VISIBLE_DEVICES=0,1,2,3 torchrun \
--nproc_per_node $NPROC_PER_NODE \
--nnodes $NNODES \
--node_rank $RANK \
--master_addr $MASTER_ADDR \
--master_port $MASTER_PORT \
src/train.py examples/lora_multi_gpu/llama3_lora_sft_ds.yaml