Support per token measurements through logits processor #45

	name: CLI CUDA Torch-ORT Tests

	on:
	workflow_dispatch:
	push:
	branches: [main]
	pull_request:
	types: [opened, reopened, synchronize]

	concurrency:
	group: ${{ github.workflow }}-${{ github.event.pull_request.number \|\| github.ref }}
	cancel-in-progress: true

	jobs:
	build_image_and_run_cli_cuda_torch_ort_tests:
	runs-on: nvidia-gpu
	steps:
	- name: Checkout
	uses: actions/checkout@v3

	- name: Build image
	run: docker build
	--file docker/cuda.dockerfile
	--build-arg USER_ID=$(id -u)
	--build-arg GROUP_ID=$(id -g)
	--build-arg CUDA_VERSION=11.8.0
	--build-arg TORCH_CUDA=cu118
	--tag opt-bench-cuda:11.8.0
	.

	- name: Get GPUs with most free memory
	id: get_devices
	run: \|
	echo "::set-output name=devices::$(nvidia-smi --query-gpu=memory.free,index --format=csv,noheader,nounits \| sort -n -k1 \| tail -n 2 \| awk -F', ' '{print $2}' \| xargs echo -n \| sed 's/ /,/g' \| awk '{print $0}')"

	- name: Run tests
	run: docker run
	--rm
	--pid host
	--shm-size 64G
	--env USE_CUDA="1"
	--entrypoint /bin/bash
	--gpus '"device=${{ steps.get_devices.outputs.devices }}"'
	--volume $(pwd):/workspace/optimum-benchmark
	--workdir /workspace/optimum-benchmark
	opt-bench-cuda:11.8.0
	-c "pip install -e .[testing,torch-ort,peft] && python -m torch_ort.configure && pytest -k 'cli and cuda and torch_ort' -x"

Provide feedback