changes to the intra-layer API for the GPT benchmark #53

	name: nvidia-rtx-3090 tests

	on:
	push:
	branches: [ develop ]
	pull_request:
	branches: [ develop ]

	jobs:
	inter-layer:
	runs-on: [ nvidia ]

	strategy:
	matrix:
	ginter: [ 1, 2 ]
	memopt: [ '0', '1' ]
	steps:
	- uses: actions/checkout@v3
	- name: Install AxoNN
	run: \|
	pip install -r requirements.txt
	- name: Download dataset
	run: \|
	python -c "import torchvision; torchvision.datasets.MNIST(root=\"./axonn/tests\", download=True, train=True)"
	- name: Train
	run: \|
	export G_inter=${{ matrix.ginter }}
	export G_data=$(( 2 / G_inter ))
	export memopt=${{ matrix.memopt }}
	echo "training with G_inter = ${G_inter}, G_data = $(( 2 / G_inter )) ${{ matrix.memopt }}"
	mpirun -n 2 pytest --with-mpi ./tests/test_vit.py
	- name: Uninstall AxoNN
	run: \|
	pip uninstall --yes axonn


	intra-layer:
	runs-on: [ nvidia ]
	steps:
	- uses: actions/checkout@v3
	- name: Install AxoNN
	run: \|
	pip install -r requirements.txt
	- name: Run unit intra-layer unit tests
	run: \|
	mpirun -n 2 pytest --with-mpi ./tests/test_intra_layer_fc.py
	- name: Uninstall AxoNN
	run: \|
	pip uninstall --yes axonn

Provide feedback