Fix flaky FLUX LoRA unit test that fails occasionally due to numerica…

…l precision.
invoke-ai · Sep 20, 2024 · 1bd33f1 · 1bd33f1
1 parent e690364
commit 1bd33f1
Showing 1 changed file with 3 additions and 1 deletion.
diff --git a/tests/backend/lora/test_lora_patcher.py b/tests/backend/lora/test_lora_patcher.py
@@ -192,4 +192,6 @@ def test_apply_lora_sidecar_patches_matches_apply_lora_patches(num_layers: int):
     with LoRAPatcher.apply_lora_sidecar_patches(model=model, patches=lora_models, prefix="", dtype=dtype):
         output_lora_sidecar_patches = model(input)
 
-    assert torch.allclose(output_lora_patches, output_lora_sidecar_patches)
+    # Note: We set atol=1e-5 because the test failed occasionally with the default atol=1e-8. Slight numerical
+    # differences are tolerable and expected due to the difference between sidecar vs. patching.
+    assert torch.allclose(output_lora_patches, output_lora_sidecar_patches, atol=1e-5)