|
Download gpu_utils/utils_readme.md from hulehule/pllm2-full-dump: direct link, hf CLI and curl.
- Browser
- Download file 613 Bytes
-
https://huggingface.co/hulehule/pllm2-full-dump/resolve/main/gpu_utils/utils_readme.md
- Command line
-
hf download hf://hulehule/pllm2-full-dump/gpu_utils/utils_readme.md
-
curl -L -o utils_readme.md https://huggingface.co/hulehule/pllm2-full-dump/resolve/main/gpu_utils/utils_readme.md
613 Bytes
直接在PLLM1.0的目录下运行:
python3 gpu_utils/gpu_stress.py --duration 2.0 --target_utilization 0.8
python3 gpu_utils/gpu_stress.py --info_only
使用所有可用GPU,运行30分钟,目标利用率50%
python3 gpu_utils/gpu_stress.py --duration 0.5 --target_utilization 0.5
指定使用2个GPU,运行10分钟,目标利用率80%
cd /Users/bytedance/Desktop/PLLM1.0-Sep23 && python3 gpu_utils/gpu_stress.py --duration 0.1667 --target_utilization 0.8 --num_gpus 2
gpu_utils/python3 clean_gpu.py
重复清理3次,每次间隔2秒
gpu_utils/python3 clean_gpu.py --repeat 3 --interval 2