real acceptance test - test_simple_neuronpedia_runner()
runs gpt2-small res-jb and compares it with expected results
remove additional level of unnecessary batching
new settings
dtype override
shuffle tokens
minibatch device fix in feature_data_generator.py line 73. fixes #4
You should be able to run pytest -s tests/acceptance/test_neuronpedia_runner.py::test_simple_neuronpedia_runner now which will check that the dashboard is correctly generated and that the output is (approximately) correct.
Updates and Fixes:
You should be able to run
pytest -s tests/acceptance/test_neuronpedia_runner.py::test_simple_neuronpedia_runner
now which will check that the dashboard is correctly generated and that the output is (approximately) correct.