Add a workflow to run benchmarks #3

	name: Benchmarks

	on:
	push:
	branches:
	- main
	pull_request:
	branches:
	- main

	jobs:
	multiprocess_gpu_test:
	runs-on: ubuntu-latest # For self-hosted runners with GPUs, update accordingly.
	strategy:
	matrix:
	gpu-count: [2, 4, 8] # Adjust based on the runner's available GPUs
	steps:
	- uses: actions/checkout@v3
	- name: Install dependencies
	run: \|
	python -m pip install --upgrade pip
	pip install -U jax
	pip install -U "jax[cuda12]"
	- name: Run Multiprocess GPU Test
	run: \|
	export CUDA_VISIBLE_DEVICES="0-$((${matrix.gpu-count} - 1))"
	python -m pytest -n ${matrix.gpu-count} tests/multiprocess_gpu_test.py

Provide feedback