ml_scripts

This is a collection of scripts including cuda installation, nightly pytorch installation, torchbench tests, and more.

cuda installation

t_cuda=121 t_cuda_path=/install/path  ./install_cuda.sh

It will automatically download and install cuda 12.1, cudnn 8.8.1, and nccl 2.18.1 to the path /install/path/cuda-12.1.

PyTorch + TorchBench Compilation

# set cuda path before install
# set conda env with python 3.10
conda create -n pt_compiled python=3.10
conda activate pt_compiled
# compile pytorch and other packages
work_path=~/project/ compile_torch.sh

torchbench scripts

For scripts that source run_base.sh, the run command is like

work_path=/home/yhao/d/p8 env1=pt_jan02 env2=pt_oct26_cu117 cuda_env1=/home/yhao/setenvs/set11.6.sh mode=train  ./run_all_speedup_cuda.sh

The detailed explanation of the arguments is in run_base.sh.

Name		Name	Last commit message	Last commit date
Latest commit History 290 Commits
cases		cases
inductor		inductor
.gitignore		.gitignore
README.md		README.md
__init__.py		__init__.py
best_bs.csv		best_bs.csv
bigbird.patch		bigbird.patch
compile_torch.sh		compile_torch.sh
cu116.txt		cu116.txt
cu117.txt		cu117.txt
d2_mask.py		d2_mask.py
down_nightly_torch.py		down_nightly_torch.py
download.sh		download.sh
filter_bs.py		filter_bs.py
filter_gpu_activetime.py		filter_gpu_activetime.py
filter_niter.py		filter_niter.py
filter_overhead.py		filter_overhead.py
filter_ratios.py		filter_ratios.py
filter_roofline.py		filter_roofline.py
filter_speedup.py		filter_speedup.py
filter_time.py		filter_time.py
filter_time_w_guard_checks.py		filter_time_w_guard_checks.py
generate_profiling.sh		generate_profiling.sh
gpu_mem_monitor.sh		gpu_mem_monitor.sh
install.sh		install.sh
install_conda_env.sh		install_conda_env.sh
install_cuda.sh		install_cuda.sh
install_cuda_env.env		install_cuda_env.env
install_cudnn.sh		install_cudnn.sh
lt		lt
merge_two_coverage_logs.py		merge_two_coverage_logs.py
osp		osp
print_all_task.sh		print_all_task.sh
remove_models.txt		remove_models.txt
requirements.txt		requirements.txt
run_all.sh		run_all.sh
run_all_eg.sh		run_all_eg.sh
run_all_for_overhead.sh		run_all_for_overhead.sh
run_all_inductor.sh		run_all_inductor.sh
run_all_iterations.sh		run_all_iterations.sh
run_all_memory.sh		run_all_memory.sh
run_all_ncu.sh		run_all_ncu.sh
run_all_profile.sh		run_all_profile.sh
run_all_speedup.sh		run_all_speedup.sh
run_all_speedup_backend.sh		run_all_speedup_backend.sh
run_all_speedup_cuda.sh		run_all_speedup_cuda.sh
run_all_speedup_cuda_mem.sh		run_all_speedup_cuda_mem.sh
run_all_tflops.sh		run_all_tflops.sh
run_all_torchexpert.sh		run_all_torchexpert.sh
run_all_torchexpert_json.sh		run_all_torchexpert_json.sh
run_base.sh		run_base.sh
run_benchmark.py		run_benchmark.py
setcuda.sh		setcuda.sh
tb-output-eval-full.json		tb-output-eval-full.json
tb-output-eval-withfp16.json		tb-output-eval-withfp16.json
tb-output-eval.json		tb-output-eval.json

FindHao/ml_scripts

Folders and files

Latest commit

History

Repository files navigation

ml_scripts

cuda installation

PyTorch + TorchBench Compilation

torchbench scripts

About

Resources

Stars

Watchers

Forks

Languages