Merge branch 'main' into modular-flux-i2i

Author: sayakpaul

.github/workflows/pr_modular_tests.yml

@@ -0,0 +1,141 @@
+name: Fast PR tests for Modular
+
+on:
+  pull_request:
+    branches: [main]
+    paths:
+      - "src/diffusers/modular_pipelines/**.py"
+      - "src/diffusers/models/modeling_utils.py"
+      - "src/diffusers/models/model_loading_utils.py"
+      - "src/diffusers/pipelines/pipeline_utils.py"
+      - "src/diffusers/pipeline_loading_utils.py"
+      - "src/diffusers/loaders/lora_base.py"
+      - "src/diffusers/loaders/lora_pipeline.py"
+      - "src/diffusers/loaders/peft.py"
+      - "tests/modular_pipelines/**.py"
+      - ".github/**.yml"
+      - "utils/**.py"
+      - "setup.py"
+  push:
+    branches:
+      - ci-*
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.head_ref || github.run_id }}
+  cancel-in-progress: true
+
+env:
+  DIFFUSERS_IS_CI: yes
+  HF_HUB_ENABLE_HF_TRANSFER: 1
+  OMP_NUM_THREADS: 4
+  MKL_NUM_THREADS: 4
+  PYTEST_TIMEOUT: 60
+
+jobs:
+  check_code_quality:
+    runs-on: ubuntu-22.04
+    steps:
+      - uses: actions/checkout@v3
+      - name: Set up Python
+        uses: actions/setup-python@v4
+        with:
+          python-version: "3.10"
+      - name: Install dependencies
+        run: |
+          python -m pip install --upgrade pip
+          pip install .[quality]
+      - name: Check quality
+        run: make quality
+      - name: Check if failure
+        if: ${{ failure() }}
+        run: |
+          echo "Quality check failed. Please ensure the right dependency versions are installed with 'pip install -e .[quality]' and run 'make style && make quality'" >> $GITHUB_STEP_SUMMARY
+
+  check_repository_consistency:
+    needs: check_code_quality
+    runs-on: ubuntu-22.04
+    steps:
+      - uses: actions/checkout@v3
+      - name: Set up Python
+        uses: actions/setup-python@v4
+        with:
+          python-version: "3.10"
+      - name: Install dependencies
+        run: |
+          python -m pip install --upgrade pip
+          pip install .[quality]
+      - name: Check repo consistency
+        run: |
+          python utils/check_copies.py
+          python utils/check_dummies.py
+          python utils/check_support_list.py
+          make deps_table_check_updated
+      - name: Check if failure
+        if: ${{ failure() }}
+        run: |
+          echo "Repo consistency check failed. Please ensure the right dependency versions are installed with 'pip install -e .[quality]' and run 'make fix-copies'" >> $GITHUB_STEP_SUMMARY
+
+  run_fast_tests:
+    needs: [check_code_quality, check_repository_consistency]
+    strategy:
+      fail-fast: false
+      matrix:
+        config:
+          - name: Fast PyTorch Modular Pipeline CPU tests
+            framework: pytorch_pipelines
+            runner: aws-highmemory-32-plus
+            image: diffusers/diffusers-pytorch-cpu
+            report: torch_cpu_modular_pipelines
+
+    name: ${{ matrix.config.name }}
+
+    runs-on:
+      group: ${{ matrix.config.runner }}
+
+    container:
+      image: ${{ matrix.config.image }}
+      options: --shm-size "16gb" --ipc host -v /mnt/hf_cache:/mnt/cache/
+
+    defaults:
+      run:
+        shell: bash
+
+    steps:
+    - name: Checkout diffusers
+      uses: actions/checkout@v3
+      with:
+        fetch-depth: 2
+
+    - name: Install dependencies
+      run: |
+        python -m venv /opt/venv && export PATH="/opt/venv/bin:$PATH"
+        python -m uv pip install -e [quality,test]
+        pip uninstall transformers -y && python -m uv pip install -U transformers@git+https://github.com/huggingface/transformers.git --no-deps
+        pip uninstall accelerate -y && python -m uv pip install -U accelerate@git+https://github.com/huggingface/accelerate.git --no-deps
+
+    - name: Environment
+      run: |
+        python -m venv /opt/venv && export PATH="/opt/venv/bin:$PATH"
+        python utils/print_env.py
+
+    - name: Run fast PyTorch Pipeline CPU tests
+      if: ${{ matrix.config.framework == 'pytorch_pipelines' }}
+      run: |
+        python -m venv /opt/venv && export PATH="/opt/venv/bin:$PATH"
+        python -m pytest -n 8 --max-worker-restart=0 --dist=loadfile \
+          -s -v -k "not Flax and not Onnx" \
+          --make-reports=tests_${{ matrix.config.report }} \
+          tests/modular_pipelines
+
+    - name: Failure short reports
+      if: ${{ failure() }}
+      run: cat reports/tests_${{ matrix.config.report }}_failures_short.txt
+
+    - name: Test suite reports artifacts
+      if: ${{ always() }}
+      uses: actions/upload-artifact@v4
+      with:
+        name: pr_${{ matrix.config.framework }}_${{ matrix.config.report }}_test_reports
+        path: reports
+
+

docs/source/en/quantization/gguf.md

@@ -77,3 +77,44 @@ Once installed, set `DIFFUSERS_GGUF_CUDA_KERNELS=true`  to use optimized kernels
 - Q5_K
 - Q6_K
 
+## Convert to GGUF
+
+Use the Space below to convert a Diffusers checkpoint into the GGUF format for inference.
+run conversion:
+
+<iframe
+	src="https://diffusers-internal-dev-diffusers-to-gguf.hf.space"
+	frameborder="0"
+	width="850"
+	height="450"
+></iframe>
+
+
+```py
+import torch
+
+from diffusers import FluxPipeline, FluxTransformer2DModel, GGUFQuantizationConfig
+
+ckpt_path = (
+    "https://huggingface.co/sayakpaul/different-lora-from-civitai/blob/main/flux_dev_diffusers-q4_0.gguf"
+)
+transformer = FluxTransformer2DModel.from_single_file(
+    ckpt_path,
+    quantization_config=GGUFQuantizationConfig(compute_dtype=torch.bfloat16),
+    config="black-forest-labs/FLUX.1-dev",
+    subfolder="transformer",
+    torch_dtype=torch.bfloat16,
+)
+pipe = FluxPipeline.from_pretrained(
+    "black-forest-labs/FLUX.1-dev",
+    transformer=transformer,
+    torch_dtype=torch.bfloat16,
+)
+pipe.enable_model_cpu_offload()
+prompt = "A cat holding a sign that says hello world"
+image = pipe(prompt, generator=torch.manual_seed(0)).images[0]
+image.save("flux-gguf.png")
+```
+
+When using Diffusers format GGUF checkpoints, it's a must to provide the model `config` path. If the
+model config resides in a `subfolder`, that needs to be specified, too.

src/diffusers/loaders/lora_conversion_utils.py

@@ -817,7 +817,11 @@ def _convert(original_key, diffusers_key, state_dict, new_state_dict):
     # has both `peft` and non-peft state dict.
     has_peft_state_dict = any(k.startswith("transformer.") for k in state_dict)
     if has_peft_state_dict:
-        state_dict = {k: v for k, v in state_dict.items() if k.startswith("transformer.")}
+        state_dict = {
+            k.replace("lora_down.weight", "lora_A.weight").replace("lora_up.weight", "lora_B.weight"): v
+            for k, v in state_dict.items()
+            if k.startswith("transformer.")
+        }
         return state_dict
 
     # Another weird one.

src/diffusers/loaders/single_file_model.py

@@ -153,9 +153,17 @@
         "checkpoint_mapping_fn": convert_cosmos_transformer_checkpoint_to_diffusers,
         "default_subfolder": "transformer",
     },
+    "QwenImageTransformer2DModel": {
+        "checkpoint_mapping_fn": lambda x: x,
+        "default_subfolder": "transformer",
+    },
 }
 
 
+def _should_convert_state_dict_to_diffusers(model_state_dict, checkpoint_state_dict):
+    return not set(model_state_dict.keys()).issubset(set(checkpoint_state_dict.keys()))
+
+
 def _get_single_file_loadable_mapping_class(cls):
     diffusers_module = importlib.import_module(__name__.split(".")[0])
     for loadable_class_str in SINGLE_FILE_LOADABLE_CLASSES:
@@ -381,19 +389,23 @@ def from_single_file(cls, pretrained_model_link_or_path_or_dict: Optional[str] =
             model_kwargs = {k: kwargs.get(k) for k in kwargs if k in expected_kwargs or k in optional_kwargs}
             diffusers_model_config.update(model_kwargs)
 
+        ctx = init_empty_weights if is_accelerate_available() else nullcontext
+        with ctx():
+            model = cls.from_config(diffusers_model_config)
+
         checkpoint_mapping_kwargs = _get_mapping_function_kwargs(checkpoint_mapping_fn, **kwargs)
-        diffusers_format_checkpoint = checkpoint_mapping_fn(
-            config=diffusers_model_config, checkpoint=checkpoint, **checkpoint_mapping_kwargs
-        )
+
+        if _should_convert_state_dict_to_diffusers(model.state_dict(), checkpoint):
+            diffusers_format_checkpoint = checkpoint_mapping_fn(
+                config=diffusers_model_config, checkpoint=checkpoint, **checkpoint_mapping_kwargs
+            )
+        else:
+            diffusers_format_checkpoint = checkpoint
+
         if not diffusers_format_checkpoint:
             raise SingleFileComponentError(
                 f"Failed to load {mapping_class_name}. Weights for this component appear to be missing in the checkpoint."
             )
-
-        ctx = init_empty_weights if is_accelerate_available() else nullcontext
-        with ctx():
-            model = cls.from_config(diffusers_model_config)
-
         # Check if `_keep_in_fp32_modules` is not None
         use_keep_in_fp32_modules = (cls._keep_in_fp32_modules is not None) and (
             (torch_dtype == torch.float16) or hasattr(hf_quantizer, "use_keep_in_fp32_modules")

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",

src/diffusers/modular_pipelines/modular_pipeline.py

@@ -493,6 +493,22 @@ def combine_outputs(*named_output_lists: List[Tuple[str, List[OutputParam]]]) ->
 
         return list(combined_dict.values())
 
+    @property
+    def input_names(self) -> List[str]:
+        return [input_param.name for input_param in self.inputs]
+
+    @property
+    def intermediate_input_names(self) -> List[str]:
+        return [input_param.name for input_param in self.intermediate_inputs]
+
+    @property
+    def intermediate_output_names(self) -> List[str]:
+        return [output_param.name for output_param in self.intermediate_outputs]
+
+    @property
+    def output_names(self) -> List[str]:
+        return [output_param.name for output_param in self.outputs]
+
 
 class PipelineBlock(ModularPipelineBlocks):
     """
@@ -2839,3 +2855,8 @@ def _dict_to_component_spec(
             type_hint=type_hint,
             **spec_dict,
         )
+
+    def set_progress_bar_config(self, **kwargs):
+        for sub_block_name, sub_block in self.blocks.sub_blocks.items():
+            if hasattr(sub_block, "set_progress_bar_config"):
+                sub_block.set_progress_bar_config(**kwargs)

src/diffusers/modular_pipelines/stable_diffusion_xl/before_denoise.py

@@ -744,8 +744,6 @@ def prepare_latents_inpaint(
         timestep=None,
         is_strength_max=True,
         add_noise=True,
-        return_noise=False,
-        return_image_latents=False,
     ):
         shape = (
             batch_size,
@@ -768,7 +766,7 @@ def prepare_latents_inpaint(
         if image.shape[1] == 4:
             image_latents = image.to(device=device, dtype=dtype)
             image_latents = image_latents.repeat(batch_size // image_latents.shape[0], 1, 1, 1)
-        elif return_image_latents or (latents is None and not is_strength_max):
+        elif latents is None and not is_strength_max:
             image = image.to(device=device, dtype=dtype)
             image_latents = self._encode_vae_image(components, image=image, generator=generator)
             image_latents = image_latents.repeat(batch_size // image_latents.shape[0], 1, 1, 1)
@@ -786,13 +784,7 @@ def prepare_latents_inpaint(
             noise = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
             latents = image_latents.to(device)
 
-        outputs = (latents,)
-
-        if return_noise:
-            outputs += (noise,)
-
-        if return_image_latents:
-            outputs += (image_latents,)
+        outputs = (latents, noise, image_latents)
 
         return outputs
 
@@ -864,7 +856,7 @@ def __call__(self, components: StableDiffusionXLModularPipeline, state: Pipeline
         block_state.height = block_state.image_latents.shape[-2] * components.vae_scale_factor
         block_state.width = block_state.image_latents.shape[-1] * components.vae_scale_factor
 
-        block_state.latents, block_state.noise = self.prepare_latents_inpaint(
+        block_state.latents, block_state.noise, block_state.image_latents = self.prepare_latents_inpaint(
             components,
             block_state.batch_size * block_state.num_images_per_prompt,
             components.num_channels_latents,
@@ -878,8 +870,6 @@ def __call__(self, components: StableDiffusionXLModularPipeline, state: Pipeline
             timestep=block_state.latent_timestep,
             is_strength_max=block_state.is_strength_max,
             add_noise=block_state.add_noise,
-            return_noise=True,
-            return_image_latents=False,
         )
 
         # 7. Prepare mask latent variables

tests/lora/utils.py

@@ -2400,7 +2400,6 @@ def _test_group_offloading_inference_denoiser(self, offload_type, use_stream):
 
             components, _, _ = self.get_dummy_components(self.scheduler_classes[0])
             pipe = self.pipeline_class(**components)
-            pipe = pipe.to(torch_device)
             pipe.set_progress_bar_config(disable=None)
             denoiser = pipe.transformer if self.unet_kwargs is None else pipe.unet
 
@@ -2416,6 +2415,10 @@ def _test_group_offloading_inference_denoiser(self, offload_type, use_stream):
                 num_blocks_per_group=1,
                 use_stream=use_stream,
             )
+            # Place other model-level components on `torch_device`.
+            for _, component in pipe.components.items():
+                if isinstance(component, torch.nn.Module):
+                    component.to(torch_device)
             group_offload_hook_1 = _get_top_level_group_offload_hook(denoiser)
             self.assertTrue(group_offload_hook_1 is not None)
             output_1 = pipe(**inputs, generator=torch.manual_seed(0))[0]