Merge branch 'main' of https://github.com/laszewsk/mlcommons

.gitignore

@@ -114,4 +114,8 @@ ENV/
 .vscode/
 
 # Source Model Data
-benchmarks/earthquake/feb-2022/data/EarthquakeDec2020/
+benchmarks/earthquake/feb-2022/data/EarthquakeDec2020/
+**/blob/blobs/*
+*.err
+*.out
+*-output.ipynb

benchmarks/cloudmask/target/green_overlay/Makefile

@@ -0,0 +1,128 @@
+SHELL=/bin/bash
+AWS_S3=aws s3 --no-sign-request --endpoint-url https://s3.echo.stfc.ac.uk
+USER_SCRATCH=/scratch/${USER}
+PROJECT_DIR=${USER_SCRATCH}/mlcommons/benchmarks/cloudmask
+PROJECT_DATA=${USER_SCRATCH}/data
+NAME=cloudmask
+
+.PHONY: image-singularity image-docker project
+
+all: requirements data
+
+
+download:
+	git clone git@github.com:laszewsk/mlcommons.git
+
+requirements:
+	time pip install -r ${PROJECT_DIR}/experiments/rivanna/requirements.txt
+
+data: 
+	mkdir -p ${PROJECT_DATA}/ssts
+	mkdir -p ${PROJECT_DATA}/one-day
+	echo -n "Downloading first portion of data..."
+	cd ${PROJECT_DATA}; ${AWS_S3} sync s3://sciml-datasets/es/cloud_slstr_ds1/one-day ./one-day --cli-read-timeout 0
+	echo -n "Downloading second portion of data..."
+	cd ${PROJECT_DATA}; ${AWS_S3} sync s3://sciml-datasets/es/cloud_slstr_ds1/ssts ./ssts --cli-read-timeout 0
+
+
+project: clean project.json generate
+
+setup:
+	python setup_env_and_yaml.py
+	source ~/ENV3/bin/activate && pip install -r /scratch/${USER}/mlcommons/benchmarks/cloudmask/experiments/rivanna/requirements.txt
+
+generate: jobs-project.sh 
+
+run: submit
+
+submit:
+	-sh jobs-project.sh
+
+localscratch: localscratch.json
+
+
+jobs-%.sh: %.json
+	cms sbatch generate submit --name=$<  > $@
+
+
+simple:
+	cms sbatch generate \
+	           --source=simple.in.slurm \
+	           --config=$< \
+	           --name=$(basename $@) \
+	           --noos \
+	           --os=USER,HOME \
+		   	   --nocm \
+	           --output_dir=./$(basename $@) \
+               --source_dir=. \
+               --verbose
+
+
+%.json: config.in.yaml
+	cms sbatch generate \
+	           --source=cloudmask_v2.in.slurm \
+	           --config=$< \
+	           --name=$(basename $@) \
+	           --noos \
+	           --os=USER,HOME \
+		       --nocm \
+	           --output_dir=./$(basename $@) \
+               --source_dir=. \
+		--copycode="cloudmask_v2.py,data_loader.py,model.py" \
+               --verbose
+
+kill: stop
+
+stop:
+	for i in "$$(squeue --user $$USER | awk 'NR>1{print $$1}')"; do scancel $$i ; done
+
+inspect:
+	$(eval D=$(shell ls project/$(ls -1) | head -n 1))
+	echo ${D}
+	$(shell emacs project/${D}/config.yaml project/${D}/job.slurm)
+
+watch: status
+
+status:
+	watch squeue --format=\"%.18i %.9P %.50j %.8u %.8T %.10M %.9l %.6D %R\" --me
+
+
+clean:
+	@-rm -rf localscratch localscratch.json jobs-localscratch.sh
+	@-rm -rf project project.json jobs-project.sh
+	@-rm -f rivanna.slurm
+	@-rm -rf '__pycache__'
+	@-rm -rf *~
+
+
+# image
+
+
+image-singularity:
+	cms rivanna singularity build image-singularity/cloudmask.def
+
+run-singularity:
+	cd image-singularity; make run
+
+shell-singularity:
+	cd image-singularity; make shell
+
+run-localscratch:
+	cd image-singularity; make run-localscratch
+
+shell-localscratch:
+	cd image-singularity; make shell-localscratch
+
+shell-rivanna:
+	cd image-singularity; make shell-rivanna
+
+run: run-singularity
+
+image: image-singularity
+
+push:
+	-git push
+	ssh -tt rivanna "cd /scratch/thf2bn/mlcommons/benchmarks/cloudmask; ssh-add; git pull"
+
+
+shell: shell-singularity