mirror of
https://github.com/hiyouga/LLaMA-Factory.git
synced 2025-08-02 03:32:50 +08:00
16 lines
366 B
Bash
16 lines
366 B
Bash
#!/bin/bash
|
|
|
|
NPROC_PER_NODE=4
|
|
NNODES=2
|
|
RANK=0
|
|
MASTER_ADDR=192.168.0.1
|
|
MASTER_PORT=29500
|
|
|
|
CUDA_VISIBLE_DEVICES=0,1,2,3 python -m torch.distributed.run \
|
|
--nproc_per_node $NPROC_PER_NODE \
|
|
--nnodes $NNODES \
|
|
--node_rank $RANK \
|
|
--master_addr $MASTER_ADDR \
|
|
--master_port $MASTER_PORT \
|
|
src/train.py examples/full_multi_gpu/llama3_full_sft.yaml
|