Flux quantized with lora (huggingface#10990)

* Flux quantized with lora

* fix

* changes

* Apply suggestions from code review

Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>

* Apply style fixes

* enable model cpu offload()

* Update src/diffusers/loaders/lora_pipeline.py

Co-authored-by: hlky <hlky@hlky.ac>

* update

* Apply suggestions from code review

* update

* add peft as an additional dependency for gguf

---------

Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
Co-authored-by: Dhruv Nair <dhruv.nair@gmail.com>

Author: 4 people

.github/workflows/nightly_tests.yml

@@ -417,7 +417,7 @@ jobs:
             additional_deps: ["peft"]
           - backend: "gguf"
             test_location: "gguf"
-            additional_deps: []
+            additional_deps: ["peft"]
           - backend: "torchao"
             test_location: "torchao"
             additional_deps: []

src/diffusers/loaders/lora_pipeline.py

@@ -22,6 +22,8 @@
     USE_PEFT_BACKEND,
     deprecate,
     get_submodule_by_name,
+    is_bitsandbytes_available,
+    is_gguf_available,
     is_peft_available,
     is_peft_version,
     is_torch_version,
@@ -68,6 +70,49 @@
 _MODULE_NAME_TO_ATTRIBUTE_MAP_FLUX = {"x_embedder": "in_channels"}
 
 
+def _maybe_dequantize_weight_for_expanded_lora(model, module):
+    if is_bitsandbytes_available():
+        from ..quantizers.bitsandbytes import dequantize_bnb_weight
+
+    if is_gguf_available():
+        from ..quantizers.gguf.utils import dequantize_gguf_tensor
+
+    is_bnb_4bit_quantized = module.weight.__class__.__name__ == "Params4bit"
+    is_gguf_quantized = module.weight.__class__.__name__ == "GGUFParameter"
+
+    if is_bnb_4bit_quantized and not is_bitsandbytes_available():
+        raise ValueError(
+            "The checkpoint seems to have been quantized with `bitsandbytes` (4bits). Install `bitsandbytes` to load quantized checkpoints."
+        )
+    if is_gguf_quantized and not is_gguf_available():
+        raise ValueError(
+            "The checkpoint seems to have been quantized with `gguf`. Install `gguf` to load quantized checkpoints."
+        )
+
+    weight_on_cpu = False
+    if not module.weight.is_cuda:
+        weight_on_cpu = True
+
+    if is_bnb_4bit_quantized:
+        module_weight = dequantize_bnb_weight(
+            module.weight.cuda() if weight_on_cpu else module.weight,
+            state=module.weight.quant_state,
+            dtype=model.dtype,
+        ).data
+    elif is_gguf_quantized:
+        module_weight = dequantize_gguf_tensor(
+            module.weight.cuda() if weight_on_cpu else module.weight,
+        )
+        module_weight = module_weight.to(model.dtype)
+    else:
+        module_weight = module.weight.data
+
+    if weight_on_cpu:
+        module_weight = module_weight.cpu()
+
+    return module_weight
+
+
 class StableDiffusionLoraLoaderMixin(LoraBaseMixin):
     r"""
     Load LoRA layers into Stable Diffusion [`UNet2DConditionModel`] and
@@ -2267,6 +2312,7 @@ def _maybe_expand_transformer_param_shape_or_error_(
         overwritten_params = {}
 
         is_peft_loaded = getattr(transformer, "peft_config", None) is not None
+        is_quantized = hasattr(transformer, "hf_quantizer")
         for name, module in transformer.named_modules():
             if isinstance(module, torch.nn.Linear):
                 module_weight = module.weight.data
@@ -2291,9 +2337,7 @@ def _maybe_expand_transformer_param_shape_or_error_(
                 if tuple(module_weight_shape) == (out_features, in_features):
                     continue
 
-                # TODO (sayakpaul): We still need to consider if the module we're expanding is
-                # quantized and handle it accordingly if that is the case.
-                module_out_features, module_in_features = module_weight.shape
+                module_out_features, module_in_features = module_weight_shape
                 debug_message = ""
                 if in_features > module_in_features:
                     debug_message += (
@@ -2316,6 +2360,10 @@ def _maybe_expand_transformer_param_shape_or_error_(
                     parent_module_name, _, current_module_name = name.rpartition(".")
                     parent_module = transformer.get_submodule(parent_module_name)
 
+                    if is_quantized:
+                        module_weight = _maybe_dequantize_weight_for_expanded_lora(transformer, module)
+
+                    # TODO: consider if this layer needs to be a quantized layer as well if `is_quantized` is True.
                     with torch.device("meta"):
                         expanded_module = torch.nn.Linear(
                             in_features, out_features, bias=bias, dtype=module_weight.dtype
@@ -2327,7 +2375,7 @@ def _maybe_expand_transformer_param_shape_or_error_(
                     new_weight = torch.zeros_like(
                         expanded_module.weight.data, device=module_weight.device, dtype=module_weight.dtype
                     )
-                    slices = tuple(slice(0, dim) for dim in module_weight.shape)
+                    slices = tuple(slice(0, dim) for dim in module_weight_shape)
                     new_weight[slices] = module_weight
                     tmp_state_dict = {"weight": new_weight}
                     if module_bias is not None:
@@ -2416,7 +2464,12 @@ def _calculate_module_shape(
         base_weight_param_name: str = None,
     ) -> "torch.Size":
         def _get_weight_shape(weight: torch.Tensor):
-            return weight.quant_state.shape if weight.__class__.__name__ == "Params4bit" else weight.shape
+            if weight.__class__.__name__ == "Params4bit":
+                return weight.quant_state.shape
+            elif weight.__class__.__name__ == "GGUFParameter":
+                return weight.quant_shape
+            else:
+                return weight.shape
 
         if base_module is not None:
             return _get_weight_shape(base_module.weight)

src/diffusers/quantizers/gguf/utils.py

@@ -400,6 +400,8 @@ def __new__(cls, data, requires_grad=False, quant_type=None):
         data = data if data is not None else torch.empty(0)
         self = torch.Tensor._make_subclass(cls, data, requires_grad)
         self.quant_type = quant_type
+        block_size, type_size = GGML_QUANT_SIZES[quant_type]
+        self.quant_shape = _quant_shape_from_byte_shape(self.shape, type_size, block_size)
 
         return self
 

tests/quantization/bnb/test_4bit.py

@@ -21,8 +21,15 @@
 import pytest
 import safetensors.torch
 from huggingface_hub import hf_hub_download
-
-from diffusers import BitsAndBytesConfig, DiffusionPipeline, FluxTransformer2DModel, SD3Transformer2DModel
+from PIL import Image
+
+from diffusers import (
+    BitsAndBytesConfig,
+    DiffusionPipeline,
+    FluxControlPipeline,
+    FluxTransformer2DModel,
+    SD3Transformer2DModel,
+)
 from diffusers.utils import is_accelerate_version, logging
 from diffusers.utils.testing_utils import (
     CaptureLogger,
@@ -696,6 +703,42 @@ def test_lora_loading(self):
         self.assertTrue(max_diff < 1e-3)
 
 
+@require_transformers_version_greater("4.44.0")
+@require_peft_backend
+class SlowBnb4BitFluxControlWithLoraTests(Base4bitTests):
+    def setUp(self) -> None:
+        gc.collect()
+        torch.cuda.empty_cache()
+
+        self.pipeline_4bit = FluxControlPipeline.from_pretrained("eramth/flux-4bit", torch_dtype=torch.float16)
+        self.pipeline_4bit.enable_model_cpu_offload()
+
+    def tearDown(self):
+        del self.pipeline_4bit
+
+        gc.collect()
+        torch.cuda.empty_cache()
+
+    def test_lora_loading(self):
+        self.pipeline_4bit.load_lora_weights("black-forest-labs/FLUX.1-Canny-dev-lora")
+
+        output = self.pipeline_4bit(
+            prompt=self.prompt,
+            control_image=Image.new(mode="RGB", size=(256, 256)),
+            height=256,
+            width=256,
+            max_sequence_length=64,
+            output_type="np",
+            num_inference_steps=8,
+            generator=torch.Generator().manual_seed(42),
+        ).images
+        out_slice = output[0, -3:, -3:, -1].flatten()
+        expected_slice = np.array([0.1636, 0.1675, 0.1982, 0.1743, 0.1809, 0.1936, 0.1743, 0.2095, 0.2139])
+
+        max_diff = numpy_cosine_similarity_distance(expected_slice, out_slice)
+        self.assertTrue(max_diff < 1e-3, msg=f"{out_slice=} != {expected_slice=}")
+
+
 @slow
 class BaseBnb4BitSerializationTests(Base4bitTests):
     def tearDown(self):

tests/quantization/gguf/test_gguf.py

@@ -8,19 +8,22 @@
 from diffusers import (
     AuraFlowPipeline,
     AuraFlowTransformer2DModel,
+    FluxControlPipeline,
     FluxPipeline,
     FluxTransformer2DModel,
     GGUFQuantizationConfig,
     SD3Transformer2DModel,
     StableDiffusion3Pipeline,
 )
+from diffusers.utils import load_image
 from diffusers.utils.testing_utils import (
     is_gguf_available,
     nightly,
     numpy_cosine_similarity_distance,
     require_accelerate,
     require_big_gpu_with_torch_cuda,
     require_gguf_version_greater_or_equal,
+    require_peft_backend,
     torch_device,
 )
 
@@ -456,3 +459,46 @@ def test_pipeline_inference(self):
         )
         max_diff = numpy_cosine_similarity_distance(expected_slice, output_slice)
         assert max_diff < 1e-4
+
+
+@require_peft_backend
+@nightly
+@require_big_gpu_with_torch_cuda
+@require_accelerate
+@require_gguf_version_greater_or_equal("0.10.0")
+class FluxControlLoRAGGUFTests(unittest.TestCase):
+    def test_lora_loading(self):
+        ckpt_path = "https://huggingface.co/city96/FLUX.1-dev-gguf/blob/main/flux1-dev-Q2_K.gguf"
+        transformer = FluxTransformer2DModel.from_single_file(
+            ckpt_path,
+            quantization_config=GGUFQuantizationConfig(compute_dtype=torch.bfloat16),
+            torch_dtype=torch.bfloat16,
+        )
+        pipe = FluxControlPipeline.from_pretrained(
+            "black-forest-labs/FLUX.1-dev",
+            transformer=transformer,
+            torch_dtype=torch.bfloat16,
+        ).to("cuda")
+        pipe.load_lora_weights("black-forest-labs/FLUX.1-Canny-dev-lora")
+
+        prompt = "A robot made of exotic candies and chocolates of different kinds. The background is filled with confetti and celebratory gifts."
+        control_image = load_image(
+            "https://huggingface.co/datasets/sayakpaul/sample-datasets/resolve/main/control_image_robot_canny.png"
+        )
+
+        output = pipe(
+            prompt=prompt,
+            control_image=control_image,
+            height=256,
+            width=256,
+            num_inference_steps=10,
+            guidance_scale=30.0,
+            output_type="np",
+            generator=torch.manual_seed(0),
+        ).images
+
+        out_slice = output[0, -3:, -3:, -1].flatten()
+        expected_slice = np.array([0.8047, 0.8359, 0.8711, 0.6875, 0.7070, 0.7383, 0.5469, 0.5820, 0.6641])
+
+        max_diff = numpy_cosine_similarity_distance(expected_slice, out_slice)
+        self.assertTrue(max_diff < 1e-3)