Skip to content
/ ml_scripts Public

A collection of my own scripts to run pytorch models, debugging pytorch compiler, and filter the results

1 star 2 forks Branches Tags Activity
Star
Notifications You must be signed in to change notification settings

FindHao/ml_scripts

BranchesTags

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

356 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
 
 

Repository files navigation

ml_scripts

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

cuda installation

CUDA_VERSION=12.1 CUDA_INSTALL_PREFIX=/install/path  ./install_cuda.sh

It will automatically download and install cuda 12.1, cudnn, and nccl 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.

About

A collection of my own scripts to run pytorch models, debugging pytorch compiler, and filter the results

Resources

Readme
Activity

Stars

1 star

Watchers

3 watching

Forks

2 forks
Report repository

Releases

No releases published

Packages

No packages published

Contributors 2

  •  
  •  

Languages