BranchCore work #533

Workflow file for this run

.github/workflows/act-backend-float32.yml at 9b4036a

	name: ACT Backend Tests (float32)

	on:
	push:
	branches: [ main ]
	pull_request:
	branches: [ main ]
	workflow_dispatch:

	env:
	PYTHONPATH: ${{ github.workspace }}
	# Optional: make TF deterministic-ish and reduce noisy logs
	TF_ENABLE_ONEDNN_OPTS: "0"
	# Enable CI mode for lightweight installs
	ACT_CI_MODE: "true"

	jobs:
	backend-float32:
	name: Backend Tests (float32)
	runs-on: ubuntu-latest
	steps:
	- name: Free disk space
	run: \|
	sudo rm -rf /usr/local/lib/android
	sudo rm -rf /usr/share/dotnet
	sudo rm -rf /opt/hostedtoolcache
	df -h

	- name: Clean pip cache
	run: \|
	pip cache purge

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

	- name: Set up Python
	uses: actions/setup-python@v5
	with:
	python-version: "3.12"

	- name: Install dependencies
	run: \|
	python -m pip install --upgrade pip
	pip install coverage torch torchvision onnx onnx2torch "onnx-simplifier" "onnxsim==0.6.5" tensorflow pandas numpy scipy pyyaml tqdm psutil transformers scikit-learn

	# Network Generation & Basic Tests

	- name: Generate Networks (float32)
	run: \|
	cd ${{ github.workspace }}
	coverage run -p -m act.back_end --generate --device cpu --dtype float32

	- name: Run Serialization Tests (float32)
	run: \|
	cd ${{ github.workspace }}
	coverage run -p -m act.back_end --test-serialization --device cpu --dtype float32

	- name: Run ACT2Torch Tests (float32)
	run: \|
	cd ${{ github.workspace }}
	coverage run -p -m act.pipeline --verify act2torch --device cpu --dtype float32

	- name: Constraint exporter — torchlp LP export over all layer_testing nets
	run: \|
	cd ${{ github.workspace }}
	for f in act/back_end/examples/nets/layer_testing_*.json; do
	coverage run -p -m act.back_end --verify --network "$f" \
	--solver torchlp --device cpu --dtype float32
	done

	# ─────────────────────────────────────────────────────────────────
	# Soundness check (TF-agnostic): runs once before per-solver matrix.
	# See act-backend-float64.yml for the full rationale.
	# ─────────────────────────────────────────────────────────────────
	- name: Soundness — counterexample validation (float32)
	run: \|
	cd ${{ github.workspace }}
	coverage run -p -m act.pipeline --verify netfactory --validate-soundness --device cpu --dtype float32

	# ─────────────────────────────────────────────────────────────────
	# Per-solver matrix. All steps below run on ACT nets generated by
	# NetFactory (act/back_end/net_factory.py), not PyTorch models —
	# TorchVision E2E lives in act-pipeline-verify.yml.
	# `--verify netfactory --validate-soundness` runs unified two-level
	# validation for each solver: Level 1 (counterexample cross-check)
	# + Level 2 (per-neuron bounds via analyze facts for interval/hybridz,
	# dual-forward bounds for dual). All solvers now exercise both levels.
	# ─────────────────────────────────────────────────────────────────

	# Solver 1: IntervalTF (forward box) + LP cascade
	- name: Solver = interval + LP cascade (float32)
	run: \|
	cd ${{ github.workspace }}
	coverage run -p -m act.pipeline --verify netfactory --validate-soundness --device cpu --dtype float32 --tf-modes interval --solvers torchlp

	# Solver 2: HybridzTF (zonotope forward) + LP cascade
	- name: Solver = hybridz + LP cascade (float32)
	run: \|
	cd ${{ github.workspace }}
	coverage run -p -m act.pipeline --verify netfactory --validate-soundness --device cpu --dtype float32 --tf-modes hybridz --solvers torchlp

	# Solver 3: DualSolver (Wong-Kolter dual backward, no LP cascade)
	- name: Solver = dual (Wong-Kolter dual backward, float32)
	run: \|
	cd ${{ github.workspace }}
	coverage run -p -m act.pipeline --verify netfactory --validate-soundness --device cpu --dtype float32 --solvers dual

	# Spec-kind coverage: RANGE / MARGIN_ROBUST / UNSAFE_LINEAR explicit runs.
	- name: Spec-kind coverage (RANGE / MARGIN / UNSAFE_LINEAR via dual + LP)
	run: \|
	cd ${{ github.workspace }}
	coverage run -p -m act.pipeline --verify netfactory --validate-soundness --device cpu --dtype float32 --tf-modes interval --solvers torchlp --networks layer_testing_range,layer_testing_margin_robust,resnet_4x3x16x16_8x2x1_1277797607
	coverage run -p -m act.pipeline --verify netfactory --validate-soundness --device cpu --dtype float32 --solvers dual --networks layer_testing_range,layer_testing_margin_robust,layer_testing_unsafe_linear,resnet_4x3x16x16_8x2x1_1277797607

	- name: List example networks (back_end)
	run: \|
	cd ${{ github.workspace }}
	coverage run -p -m act.back_end --list-examples

	- name: Combine coverage data
	if: always()
	run: \|
	cd ${{ github.workspace }}
	coverage combine
	coverage xml -o coverage.xml
	coverage report

	- name: Upload coverage to Codecov
	if: always()
	uses: codecov/codecov-action@v5
	with:
	files: ./coverage.xml
	flags: backend-float32
	name: backend-float32
	fail_ci_if_error: false
	env:
	CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}

Navigation Menu

BranchCore work #533

Workflow file

BranchCore work #533

Uh oh!

Workflow file for this run