Skip to content

Local UI to run and train LLMs and diffusion models. Supports GGUF, MLX, Qwen3.8, DeepSeek-V4.1, MiniMax-H3, Gemma 4, FLUX and more.

Tests: gate the new Q-GaLore offload CUDA case on a real CUDA device (#13994)

Daniel Han06690987170e
diff --git a/tests/test_optimizer_state_offload.py b/tests/test_optimizer_state_offload.py
index 5688df879a7..0977c065c05 100644
--- a/tests/test_optimizer_state_offload.py
+++ b/tests/test_optimizer_state_offload.py
@@ -587,7 +587,7 @@ def test_weight_quant_q_galore_steps_unchunked():
assert not module._chunk_safe(QGaLoreAdamW8bit([{"params": p, "weight_quant": True}]))
[email protected](not torch.cuda.is_available(), reason = "needs CUDA")
[email protected](not has_real_cuda(), reason = "needs CUDA")
@pytest.mark.parametrize("weight_quant", [False, True])
def test_q_galore_multi_chunk_offload(monkeypatch, weight_quant):
module = _module()