From 7166dc47c9bd764e2cd1d7c3d9e046cdc439b58c Mon Sep 17 00:00:00 2001 From: CaptainJack2491 Date: Mon, 17 Aug 2026 16:35:28 +0100 Subject: perf(cuda): add AMP fp16 autocast with GradScaler and torch.compile reduce-overhead mode --- tests/test_cuda_scaling.py | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) (limited to 'tests/test_cuda_scaling.py') diff --git a/tests/test_cuda_scaling.py b/tests/test_cuda_scaling.py index 5f0befc..73cda66 100644 --- a/tests/test_cuda_scaling.py +++ b/tests/test_cuda_scaling.py @@ -15,9 +15,11 @@ def test_device_resolution(): def test_config_device_args(): - cfg = parse_args(["rnn", "0", "--device", "cpu", "--eval_batch_size", "256"]) + cfg = parse_args(["rnn", "0", "--device", "cpu", "--eval_batch_size", "256", "--use_amp", "True", "--compile_model", "True"]) assert cfg.device == "cpu" assert cfg.eval_batch_size == 256 + assert cfg.use_amp is True + assert cfg.compile_model is True def test_bisect_next_prime_large_range(): -- cgit v1.2.3