update

Author: DN6

src/diffusers/loaders/single_file_utils.py

@@ -60,6 +60,7 @@
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
 
 CHECKPOINT_KEY_NAMES = {
+    "v1": "model.diffusion_model.output_blocks.11.0.skip_connection.weight",
     "v2": "model.diffusion_model.input_blocks.2.1.transformer_blocks.0.attn2.to_k.weight",
     "xl_base": "conditioner.embedders.1.model.transformer.resblocks.9.mlp.c_proj.bias",
     "xl_refiner": "conditioner.embedders.0.model.transformer.resblocks.9.mlp.c_proj.bias",
@@ -2196,12 +2197,6 @@ def convert_flux_transformer_checkpoint_to_diffusers(checkpoint, **kwargs):
         if "model.diffusion_model." in k:
             checkpoint[k.replace("model.diffusion_model.", "")] = checkpoint.pop(k)
 
-    original_flux = any(k.startswith("double_blocks.") for k in checkpoint) or any(
-        k.startswith("single_blocks.") for k in checkpoint
-    )
-    if not original_flux:
-        return checkpoint
-
     num_layers = list(set(int(k.split(".", 2)[1]) for k in checkpoint if "double_blocks." in k))[-1] + 1  # noqa: C401
     num_single_layers = list(set(int(k.split(".", 2)[1]) for k in checkpoint if "single_blocks." in k))[-1] + 1  # noqa: C401
     mlp_ratio = 4.0

tests/quantization/gguf/test_gguf.py

@@ -212,6 +212,7 @@ def _check_for_gguf_linear(model):
 
 class FluxGGUFSingleFileTests(GGUFSingleFileTesterMixin, unittest.TestCase):
     ckpt_path = "https://huggingface.co/city96/FLUX.1-dev-gguf/blob/main/flux1-dev-Q2_K.gguf"
+    diffusers_ckpt_path = "https://huggingface.co/sayakpaul/flux-diffusers-gguf/blob/main/model-Q4_0.gguf"
     torch_dtype = torch.bfloat16
     model_cls = FluxTransformer2DModel
     expected_memory_use_in_gb = 5
@@ -296,6 +297,16 @@ def test_pipeline_inference(self):
         max_diff = numpy_cosine_similarity_distance(expected_slice, output_slice)
         assert max_diff < 1e-4
 
+    def test_loading_gguf_diffusers_format(self):
+        model = self.model_cls.from_single_file(
+            self.diffusers_ckpt_path,
+            subfolder="transformer",
+            quantization_config=GGUFQuantizationConfig(compute_dtype=torch.bfloat16),
+            config="black-forest-labs/FLUX.1-dev",
+        )
+        model.to("cuda")
+        model(**self.get_dummy_inputs())
+
 
 class SD35LargeGGUFSingleFileTests(GGUFSingleFileTesterMixin, unittest.TestCase):
     ckpt_path = "https://huggingface.co/city96/stable-diffusion-3.5-large-gguf/blob/main/sd3.5_large-Q4_0.gguf"