Refactor llm perf backend handling #123

	name: Update LLM Perf Benchmarks - CUDA PyTorch

	on:
	workflow_dispatch:
	schedule:
	- cron: "0 0 * * *"
	push:
	branches:
	- main
	pull_request:
	branches:
	- main
	types:
	- opened
	- reopened
	- synchronize
	- labeled
	- unlabeled

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

	env:
	IMAGE: ghcr.io/huggingface/optimum-benchmark:latest-cuda

	jobs:
	run_benchmarks:
	if: ${{
	(github.event_name == 'push') \|\|
	(github.event_name == 'workflow_dispatch') \|\|
	contains( github.event.pull_request.labels.*.name, 'leaderboard')}}

	strategy:
	fail-fast: false
	matrix:
	subset: [unquantized, bnb, awq, gptq]

	machine:
	[
	{ name: 1xA10, runs-on: { group: "aws-g5-4xlarge-plus" } },
	{ name: 1xT4, runs-on: { group: "aws-g4dn-2xlarge" } },
	]

	runs-on: ${{ matrix.machine.runs-on }}

	steps:
	- name: Checkout
	uses: actions/checkout@v4

	- name: Run benchmarks
	uses: addnab/docker-run-action@v3
	env:
	SUBSET: ${{ matrix.subset }}
	MACHINE: ${{ matrix.machine.name }}
	HF_TOKEN: ${{ secrets.HF_TOKEN }}
	with:
	image: ${{ env.IMAGE }}
	options: \|
	--rm
	--gpus all
	--shm-size 64G
	--env SUBSET
	--env MACHINE
	--env HF_TOKEN
	--env MKL_THREADING_LAYER=GNU
	--env HF_HUB_ENABLE_HF_TRANSFER=1
	--volume ${{ github.workspace }}:/workspace
	--workdir /workspace
	run: \|
	pip install packaging && pip install flash-attn einops scipy auto-gptq optimum bitsandbytes autoawq codecarbon
	pip install -U transformers huggingface_hub[hf_transfer]
	pip install -e .
	python llm_perf/benchmark_runners/update_llm_perf_cuda_pytorch.py

Provide feedback