Fix Wan AccVideo/CausVid fuse_lora (huggingface#11856)

* fix

* actually, better fix

* empty commit; trigger tests again

* mark wanvace test as flaky

Author: a-r-r-o-w

src/diffusers/loaders/lora_conversion_utils.py

@@ -1825,24 +1825,22 @@ def _convert_non_diffusers_wan_lora_to_diffusers(state_dict):
     is_i2v_lora = any("k_img" in k for k in original_state_dict) and any("v_img" in k for k in original_state_dict)
     lora_down_key = "lora_A" if any("lora_A" in k for k in original_state_dict) else "lora_down"
     lora_up_key = "lora_B" if any("lora_B" in k for k in original_state_dict) else "lora_up"
+    has_time_projection_weight = any(
+        k.startswith("time_projection") and k.endswith(".weight") for k in original_state_dict
+    )
 
-    diff_keys = [k for k in original_state_dict if k.endswith((".diff_b", ".diff"))]
-    if diff_keys:
-        for diff_k in diff_keys:
-            param = original_state_dict[diff_k]
-            # The magnitudes of the .diff-ending weights are very low (most are below 1e-4, some are upto 1e-3,
-            # and 2 of them are about 1.6e-2 [the case with AccVideo lora]). The low magnitudes mostly correspond
-            # to norm layers. Ignoring them is the best option at the moment until a better solution is found. It
-            # is okay to ignore because they do not affect the model output in a significant manner.
-            threshold = 1.6e-2
-            absdiff = param.abs().max() - param.abs().min()
-            all_zero = torch.all(param == 0).item()
-            all_absdiff_lower_than_threshold = absdiff < threshold
-            if all_zero or all_absdiff_lower_than_threshold:
-                logger.debug(
-                    f"Removed {diff_k} key from the state dict as it's all zeros, or values lower than hardcoded threshold."
-                )
-                original_state_dict.pop(diff_k)
+    for key in list(original_state_dict.keys()):
+        if key.endswith((".diff", ".diff_b")) and "norm" in key:
+            # NOTE: we don't support this because norm layer diff keys are just zeroed values. We can support it
+            # in future if needed and they are not zeroed.
+            original_state_dict.pop(key)
+            logger.debug(f"Removing {key} key from the state dict as it is a norm diff key. This is unsupported.")
+
+        if "time_projection" in key and not has_time_projection_weight:
+            # AccVideo lora has diff bias keys but not the weight keys. This causes a weird problem where
+            # our lora config adds the time proj lora layers, but we don't have the weights for them.
+            # CausVid lora has the weight keys and the bias keys.
+            original_state_dict.pop(key)
 
     # For the `diff_b` keys, we treat them as lora_bias.
     # https://huggingface.co/docs/peft/main/en/package_reference/lora#peft.LoraConfig.lora_bias

tests/lora/test_lora_layers_wanvace.py

@@ -28,6 +28,7 @@
 from diffusers.utils.import_utils import is_peft_available
 from diffusers.utils.testing_utils import (
     floats_tensor,
+    is_flaky,
     require_peft_backend,
     require_peft_version_greater,
     skip_mps,
@@ -215,3 +216,7 @@ def test_lora_exclude_modules_wanvace(self):
                 np.allclose(output_lora_exclude_modules, output_lora_pretrained, atol=1e-3, rtol=1e-3),
                 "Lora outputs should match.",
             )
+
+    @is_flaky
+    def test_simple_inference_with_text_denoiser_lora_and_scale(self):
+        super().test_simple_inference_with_text_denoiser_lora_and_scale()