当前位置：首页 > ai >正文

llama-factory微调报错：

ai 2025/7/25 9:52:17

报错信息

[INFO] [utils.py:789:see_memory_usage] CPU Virtual Memory: used =
81.51 GB, percent = 64.9%
W0419 10:14:27.573000 108354 site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 108373 closing signal SIGTERM
W0419 10:14:27.594000 108354 site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 108375 closing signal SIGTERM
W0419 10:14:27.594000 108354 site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 108376 closing signal SIGTERM
E0419 10:14:33.446000 108354 site-packages/torch/distributed/elastic/multiprocessing/api.py:869] failed (exitcode: -9) local_rank: 1 (pid: 108374) of binary: /opt/conda/envs/llamaf/bin/python

Traceback (most recent call last):
File “/opt/conda/envs/llamaf/bin/torchrun”, line 33, in
sys.exit(load_entry_point(‘torch==2.5.1’, ‘console_scripts’, ‘torchrun’)())
xxx
xxx
xxx
torch.distributed.elastic.multiprocessing.errors.ChildFailedError:
/home/LLaMA-Factory/src/llamafactory/launcher.py FAILED

由于看到CPU Virtual Memory 只加载到了64.9%，发现是内存不足。

解决方法：

# 创建分区路径
sudo mkdir -p /data/swap/
# 设置分区的大小
# bs=128M是块大小，count=64是块数量，
# 所以swap空间大小是bs*count=96GB
sudo dd if=/dev/zero of=/data/swap/swap0 bs=512M count=192
# 设置该目录权限
sudo chmod 0600 /data/swap/swap0
# 创建SWAP文件
sudo mkswap /data/swap/swap0
# 激活SWAP文件
sudo swapon /data/swap/swap0
# 查看SWAP信息是否正确
sudo swapon -s