update

Author: DN6

src/diffusers/quantizers/gguf/gguf_quantizer.py

@@ -49,6 +49,7 @@ def validate_environment(self, *args, **kwargs):
                 "To load GGUF format files you must have `gguf` installed in your environment: `pip install gguf>=0.10.0`"
             )
 
+    # Copied from diffusers.quantizers.bitsandbytes.bnb_quantizer.BnB4BitDiffusersQuantizer.adjust_max_memory
     def adjust_max_memory(self, max_memory: Dict[str, Union[int, str]]) -> Dict[str, Union[int, str]]:
         # need more space for buffers that are created during quantization
         max_memory = {key: val * 0.90 for key, val in max_memory.items()}

src/diffusers/utils/testing_utils.py

@@ -32,6 +32,7 @@
     is_bitsandbytes_available,
     is_compel_available,
     is_flax_available,
+    is_gguf_available,
     is_note_seq_available,
     is_onnx_available,
     is_opencv_available,
@@ -476,6 +477,18 @@ def decorator(test_case):
     return decorator
 
 
+def require_gguf_version_greater_or_equal(gguf_version):
+    def decorator(test_case):
+        correct_gguf_version = is_gguf_available() and version.parse(
+            version.parse(importlib.metadata.version("gguf")).base_version
+        ) >= version.parse(gguf_version)
+        return unittest.skipUnless(
+            correct_gguf_version, f"Test requires gguf with the version greater than {gguf_version}."
+        )(test_case)
+
+    return decorator
+
+
 def deprecate_after_peft_backend(test_case):
     """
     Decorator marking a test that will be skipped after PEFT backend

tests/quantization/gguf/test_gguf.py

@@ -4,16 +4,24 @@
 import torch
 
 from diffusers import FluxTransformer2DModel, GGUFQuantizationConfig
-from diffusers.quantizers.gguf.utils import GGUFParameter
 from diffusers.utils.testing_utils import (
+    is_gguf_available,
     nightly,
+    require_accelerate,
     require_big_gpu_with_torch_cuda,
+    require_gguf_version_greater_or_equal,
     torch_device,
 )
 
 
+if is_gguf_available():
+    from diffusers.quantizers.gguf.utils import GGUFParameter
+
+
 @nightly
 @require_big_gpu_with_torch_cuda
+@require_accelerate
+@require_gguf_version_greater_or_equal("0.10.0")
 class GGUFSingleFileTests(unittest.TestCase):
     ckpt_path = "https://huggingface.co/city96/FLUX.1-dev-gguf/blob/main/flux1-dev-Q2_K.gguf"
     torch_dtype = torch.bfloat16