Merge branch 'main' into move-to-uv

* copy the state dict in load lora weights

* fixup

.github/ISSUE_TEMPLATE/bug-report.yml

@@ -66,32 +66,32 @@ body:
         Questions on DiffusionPipeline (Saving, Loading, From pretrained, ...):
 
         Questions on pipelines:
-        - Stable Diffusion @yiyixuxu @DN6 @sayakpaul @patrickvonplaten
-        - Stable Diffusion XL @yiyixuxu @sayakpaul @DN6 @patrickvonplaten
-        - Kandinsky @yiyixuxu @patrickvonplaten
-        - ControlNet @sayakpaul @yiyixuxu @DN6 @patrickvonplaten
-        - T2I Adapter @sayakpaul @yiyixuxu @DN6 @patrickvonplaten
-        - IF @DN6 @patrickvonplaten
-        - Text-to-Video / Video-to-Video @DN6 @sayakpaul @patrickvonplaten
-        - Wuerstchen @DN6 @patrickvonplaten
+        - Stable Diffusion @yiyixuxu @DN6 @sayakpaul 
+        - Stable Diffusion XL @yiyixuxu @sayakpaul @DN6 
+        - Kandinsky @yiyixuxu 
+        - ControlNet @sayakpaul @yiyixuxu @DN6 
+        - T2I Adapter @sayakpaul @yiyixuxu @DN6 
+        - IF @DN6 
+        - Text-to-Video / Video-to-Video @DN6 @sayakpaul 
+        - Wuerstchen @DN6 
         - Other: @yiyixuxu @DN6
 
         Questions on models:
-        - UNet @DN6 @yiyixuxu @sayakpaul @patrickvonplaten
-        - VAE @sayakpaul @DN6 @yiyixuxu @patrickvonplaten
-        - Transformers/Attention @DN6 @yiyixuxu @sayakpaul @DN6 @patrickvonplaten
+        - UNet @DN6 @yiyixuxu @sayakpaul 
+        - VAE @sayakpaul @DN6 @yiyixuxu 
+        - Transformers/Attention @DN6 @yiyixuxu @sayakpaul @DN6 
 
-        Questions on Schedulers: @yiyixuxu @patrickvonplaten
+        Questions on Schedulers: @yiyixuxu 
 
-        Questions on LoRA: @sayakpaul @patrickvonplaten
+        Questions on LoRA: @sayakpaul 
 
-        Questions on Textual Inversion: @sayakpaul @patrickvonplaten
+        Questions on Textual Inversion: @sayakpaul 
 
         Questions on Training: 
-        - DreamBooth @sayakpaul @patrickvonplaten
-        - Text-to-Image Fine-tuning @sayakpaul @patrickvonplaten
-        - Textual Inversion @sayakpaul @patrickvonplaten
-        - ControlNet @sayakpaul @patrickvonplaten
+        - DreamBooth @sayakpaul 
+        - Text-to-Image Fine-tuning @sayakpaul 
+        - Textual Inversion @sayakpaul 
+        - ControlNet @sayakpaul 
 
         Questions on Tests: @DN6 @sayakpaul @yiyixuxu 
 
@@ -99,7 +99,7 @@ body:
 
         Questions on JAX- and MPS-related things: @pcuenca
 
-        Questions on audio pipelines: @DN6 @patrickvonplaten
+        Questions on audio pipelines: @DN6 
         
 
         

.github/PULL_REQUEST_TEMPLATE.md

@@ -38,13 +38,13 @@ members/contributors who may be interested in your PR.
 
 Core library:
 
-- Schedulers: @yiyixuxu and @patrickvonplaten
-- Pipelines:  @patrickvonplaten and @sayakpaul
-- Training examples: @sayakpaul and @patrickvonplaten
-- Docs: @stevhliu and @yiyixuxu
+- Schedulers: @yiyixuxu 
+- Pipelines:  @sayakpaul @yiyixuxu @DN6
+- Training examples: @sayakpaul 
+- Docs: @stevhliu and @sayakpaul
 - JAX and MPS: @pcuenca
 - Audio: @sanchit-gandhi
-- General functionalities: @patrickvonplaten and @sayakpaul
+- General functionalities: @sayakpaul @yiyixuxu @DN6
 
 Integrations:
 

.github/workflows/push_tests.yml

@@ -61,7 +61,7 @@ jobs:
       max-parallel: 1
       matrix:
         module: ${{ fromJson(needs.setup_torch_cuda_pipeline_matrix.outputs.pipeline_test_matrix) }}
-    runs-on: [single-gpu, nvidia-gpu, t4, ci]
+    runs-on: docker-gpu
     container:
       image: diffusers/diffusers-pytorch-cuda
       options: --shm-size "16gb" --ipc host -v /mnt/hf_cache:/mnt/cache/ --gpus 0

docker/diffusers-flax-cpu/Dockerfile

@@ -23,13 +23,13 @@ ENV PATH="/opt/venv/bin:$PATH"
 
 # pre-install the heavy dependencies (these can later be overridden by the deps from setup.py)
 # follow the instructions here: https://cloud.google.com/tpu/docs/run-in-container#train_a_jax_model_in_a_docker_container
-RUN python3 -m pip install --no-cache-dir --upgrade pip && \
-    python3 -m pip install --upgrade --no-cache-dir \
+RUN python3 -m pip install --no-cache-dir --upgrade pip uv && \
+    python3 -m uv pip install --upgrade --no-cache-dir \
         clu \
         "jax[cpu]>=0.2.16,!=0.3.2" \
         "flax>=0.4.1" \
         "jaxlib>=0.1.65" && \
-    python3 -m pip install --no-cache-dir \
+    python3 -m uv pip install --no-cache-dir \
         accelerate \
         datasets \
         hf-doc-builder \

docker/diffusers-flax-tpu/Dockerfile

@@ -23,15 +23,15 @@ ENV PATH="/opt/venv/bin:$PATH"
 
 # pre-install the heavy dependencies (these can later be overridden by the deps from setup.py)
 # follow the instructions here: https://cloud.google.com/tpu/docs/run-in-container#train_a_jax_model_in_a_docker_container
-RUN python3 -m pip install --no-cache-dir --upgrade pip && \
-    python3 -m pip install --no-cache-dir \
+RUN python3 -m pip install --no-cache-dir --upgrade pip uv && \
+    python3 -m uv pip install --no-cache-dir \
         "jax[tpu]>=0.2.16,!=0.3.2" \
         -f https://storage.googleapis.com/jax-releases/libtpu_releases.html && \
-    python3 -m pip install --upgrade --no-cache-dir \
+    python3 -m uv pip install --upgrade --no-cache-dir \
         clu \
         "flax>=0.4.1" \
         "jaxlib>=0.1.65" && \
-    python3 -m pip install --no-cache-dir \
+    python3 -m uv pip install --no-cache-dir \
         accelerate \
         datasets \
         hf-doc-builder \

docker/diffusers-onnxruntime-cpu/Dockerfile

@@ -22,14 +22,14 @@ RUN python3 -m venv /opt/venv
 ENV PATH="/opt/venv/bin:$PATH"
 
 # pre-install the heavy dependencies (these can later be overridden by the deps from setup.py)
-RUN python3 -m pip install --no-cache-dir --upgrade pip && \
-    python3 -m pip install --no-cache-dir \
+RUN python3 -m pip install --no-cache-dir --upgrade pip uv && \
+    python3 -m uv pip install --no-cache-dir \
         torch==2.1.2 \
         torchvision==0.16.2 \
         torchaudio==2.1.2 \
         onnxruntime \
         --extra-index-url https://download.pytorch.org/whl/cpu && \
-    python3 -m pip install --no-cache-dir \
+    python3 -m uv pip install --no-cache-dir \
         accelerate \
         datasets \
         hf-doc-builder \

docker/diffusers-onnxruntime-cuda/Dockerfile

@@ -22,14 +22,14 @@ RUN python3 -m venv /opt/venv
 ENV PATH="/opt/venv/bin:$PATH"
 
 # pre-install the heavy dependencies (these can later be overridden by the deps from setup.py)
-RUN python3 -m pip install --no-cache-dir --upgrade pip && \
-    python3 -m pip install --no-cache-dir \
+RUN python3 -m pip install --no-cache-dir --upgrade pip uv && \
+    python3 -m uv pip install --no-cache-dir \
         torch==2.1.2 \
         torchvision==0.16.2 \
         torchaudio==2.1.2 \
         "onnxruntime-gpu>=1.13.1" \
         --extra-index-url https://download.pytorch.org/whl/cu117 && \
-    python3 -m pip install --no-cache-dir \
+    python3 -m uv pip install --no-cache-dir \
         accelerate \
         datasets \
         hf-doc-builder \

docker/diffusers-pytorch-compile-cuda/Dockerfile

@@ -24,8 +24,8 @@ RUN python3.9 -m venv /opt/venv
 ENV PATH="/opt/venv/bin:$PATH"
 
 # pre-install the heavy dependencies (these can later be overridden by the deps from setup.py)
-RUN python3.9 -m pip install --no-cache-dir --upgrade pip && \
-    python3.9 -m pip install --no-cache-dir \
+RUN python3.9 -m pip install --no-cache-dir --upgrade pip uv && \
+    python3.9 -m uv pip install --no-cache-dir \
     torch \
     torchvision \
     torchaudio \

docker/diffusers-pytorch-cpu/Dockerfile

@@ -23,14 +23,14 @@ RUN python3 -m venv /opt/venv
 ENV PATH="/opt/venv/bin:$PATH"
 
 # pre-install the heavy dependencies (these can later be overridden by the deps from setup.py)
-RUN python3 -m pip install --no-cache-dir --upgrade pip && \
-    python3 -m pip install --no-cache-dir \
+RUN python3 -m pip install --no-cache-dir --upgrade pip uv && \
+    python3 -m uv pip install --no-cache-dir \
         torch \
         torchvision \
         torchaudio \
         invisible_watermark \
         --extra-index-url https://download.pytorch.org/whl/cpu && \
-    python3 -m pip install --no-cache-dir \
+    python3 -m uv pip install --no-cache-dir \
         accelerate \
         datasets \
         hf-doc-builder \

docker/diffusers-pytorch-cuda/Dockerfile

@@ -23,8 +23,8 @@ RUN python3 -m venv /opt/venv
 ENV PATH="/opt/venv/bin:$PATH"
 
 # pre-install the heavy dependencies (these can later be overridden by the deps from setup.py)
-RUN python3 -m pip install --no-cache-dir --upgrade pip && \
-    python3 -m pip install --no-cache-dir \
+RUN python3 -m pip install --no-cache-dir --upgrade pip uv && \
+    python3 -m uv pip install --no-cache-dir \
     torch \
     torchvision \
     torchaudio \

docker/diffusers-pytorch-xformers-cuda/Dockerfile

@@ -23,13 +23,13 @@ RUN python3 -m venv /opt/venv
 ENV PATH="/opt/venv/bin:$PATH"
 
 # pre-install the heavy dependencies (these can later be overridden by the deps from setup.py)
-RUN python3 -m pip install --no-cache-dir --upgrade pip && \
+RUN python3 -m pip install --no-cache-dir --upgrade pip uv && \
     python3 -m pip install --no-cache-dir \
         torch \
         torchvision \
         torchaudio \
         invisible_watermark && \
-    python3 -m pip install --no-cache-dir \
+    python3 -m uv pip install --no-cache-dir \
         accelerate \
         datasets \
         hf-doc-builder \

examples/text_to_image/train_text_to_image_sdxl.py

@@ -951,6 +951,9 @@ def main(args):
         unet, optimizer, train_dataloader, lr_scheduler
     )
 
+    if args.use_ema:
+        ema_unet.to(accelerator.device)
+
     # We need to recalculate our total training steps as the size of the training dataloader may have changed.
     num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
     if overrode_max_train_steps:
@@ -1126,6 +1129,8 @@ def main(args):
 
             # Checks if the accelerator has performed an optimization step behind the scenes
             if accelerator.sync_gradients:
+                if args.use_ema:
+                    ema_unet.step(unet.parameters())
                 progress_bar.update(1)
                 global_step += 1
                 accelerator.log({"train_loss": train_loss}, step=global_step)

examples/textual_inversion/textual_inversion_sdxl.py

@@ -546,6 +546,8 @@ class TextualInversionDataset(Dataset):
 
         example["original_size"] = (image.height, image.width)
 
+        image = image.resize((self.size, self.size), resample=self.interpolation)
+
         if self.center_crop:
             y1 = max(0, int(round((image.height - self.size) / 2.0)))
             x1 = max(0, int(round((image.width - self.size) / 2.0)))
@@ -576,7 +578,6 @@ class TextualInversionDataset(Dataset):
         img = np.array(image).astype(np.uint8)
 
         image = Image.fromarray(img)
-        image = image.resize((self.size, self.size), resample=self.interpolation)
 
         image = self.flip_transform(image)
         image = np.array(image).astype(np.uint8)

scripts/convert_dance_diffusion_to_diffusers.py

@@ -4,6 +4,7 @@ import math
 import os
 from copy import deepcopy
 
+import requests
 import torch
 from audio_diffusion.models import DiffusionAttnUnet1D
 from diffusion import sampling
@@ -73,9 +74,14 @@ class DiffusionUncond(nn.Module):
 
 def download(model_name):
     url = MODELS_MAP[model_name]["url"]
-    os.system(f"wget {url} ./")
+    r = requests.get(url, stream=True)
 
-    return f"./{model_name}.ckpt"
+    local_filename = f"./{model_name}.ckpt"
+    with open(local_filename, "wb") as fp:
+        for chunk in r.iter_content(chunk_size=8192):
+            fp.write(chunk)
+
+    return local_filename
 
 
 DOWN_NUM_TO_LAYER = {

src/diffusers/loaders/lora.py

@@ -106,6 +106,10 @@ class LoraLoaderMixin:
         if not USE_PEFT_BACKEND:
             raise ValueError("PEFT backend is required for this method.")
 
+        # if a dict is passed, copy it instead of modifying it inplace
+        if isinstance(pretrained_model_name_or_path_or_dict, dict):
+            pretrained_model_name_or_path_or_dict = pretrained_model_name_or_path_or_dict.copy()
+
         # First, ensure that the checkpoint is a compatible one and can be successfully loaded.
         state_dict, network_alphas = self.lora_state_dict(pretrained_model_name_or_path_or_dict, **kwargs)
 
@@ -1229,6 +1233,10 @@ class StableDiffusionXLLoraLoaderMixin(LoraLoaderMixin):
         # it here explicitly to be able to tell that it's coming from an SDXL
         # pipeline.
 
+        # if a dict is passed, copy it instead of modifying it inplace
+        if isinstance(pretrained_model_name_or_path_or_dict, dict):
+            pretrained_model_name_or_path_or_dict = pretrained_model_name_or_path_or_dict.copy()
+
         # First, ensure that the checkpoint is a compatible one and can be successfully loaded.
         state_dict, network_alphas = self.lora_state_dict(
             pretrained_model_name_or_path_or_dict,

src/diffusers/models/lora.py

@@ -361,16 +361,19 @@ class LoRACompatibleConv(nn.Conv2d):
         self.w_down = None
 
     def forward(self, hidden_states: torch.Tensor, scale: float = 1.0) -> torch.Tensor:
-        if self.lora_layer is None:
-            # make sure to the functional Conv2D function as otherwise torch.compile's graph will break
-            # see: https://github.com/huggingface/diffusers/pull/4315
-            return F.conv2d(
-                hidden_states, self.weight, self.bias, self.stride, self.padding, self.dilation, self.groups
-            )
+        if self.padding_mode != "zeros":
+            hidden_states = F.pad(hidden_states, self._reversed_padding_repeated_twice, mode=self.padding_mode)
+            padding = (0, 0)
+        else:
+            padding = self.padding
+
+        original_outputs = F.conv2d(
+            hidden_states, self.weight, self.bias, self.stride, padding, self.dilation, self.groups
+        )
+
+        if self.lora_layer is None:
+            return original_outputs
         else:
-            original_outputs = F.conv2d(
-                hidden_states, self.weight, self.bias, self.stride, self.padding, self.dilation, self.groups
-            )
             return original_outputs + (scale * self.lora_layer(hidden_states))
 
 

tests/lora/test_lora_layers_peft.py

@@ -26,11 +26,13 @@ import torch.nn as nn
 from huggingface_hub import hf_hub_download
 from huggingface_hub.repocard import RepoCard
 from packaging import version
+from safetensors.torch import load_file
 from transformers import CLIPTextModel, CLIPTextModelWithProjection, CLIPTokenizer
 
 from diffusers import (
     AutoencoderKL,
     AutoPipelineForImage2Image,
+    AutoPipelineForText2Image,
     ControlNetModel,
     DDIMScheduler,
     DiffusionPipeline,
@@ -1177,6 +1179,24 @@ class PeftLoraLoaderMixinTests:
             # Just makes sure it works..
             _ = pipe(**inputs, generator=torch.manual_seed(0)).images
 
+    def test_modify_padding_mode(self):
+        def set_pad_mode(network, mode="circular"):
+            for _, module in network.named_modules():
+                if isinstance(module, torch.nn.Conv2d):
+                    module.padding_mode = mode
+
+        for scheduler_cls in [DDIMScheduler, LCMScheduler]:
+            components, _, _ = self.get_dummy_components(scheduler_cls)
+            pipe = self.pipeline_class(**components)
+            pipe = pipe.to(self.torch_device)
+            pipe.set_progress_bar_config(disable=None)
+            _pad_mode = "circular"
+            set_pad_mode(pipe.vae, _pad_mode)
+            set_pad_mode(pipe.unet, _pad_mode)
+
+            _, _, inputs = self.get_dummy_inputs()
+            _ = pipe(**inputs).images
+
 
 class StableDiffusionLoRATests(PeftLoraLoaderMixinTests, unittest.TestCase):
     pipeline_class = StableDiffusionPipeline
@@ -1727,6 +1747,40 @@ class LoraIntegrationTests(PeftLoraLoaderMixinTests, unittest.TestCase):
         self.assertTrue(np.allclose(lora_images, lora_images_again, atol=1e-3))
         release_memory(pipe)
 
+    def test_not_empty_state_dict(self):
+        # Makes sure https://github.com/huggingface/diffusers/issues/7054 does not happen again
+        pipe = AutoPipelineForText2Image.from_pretrained(
+            "runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16
+        ).to("cuda")
+        pipe.scheduler = LCMScheduler.from_config(pipe.scheduler.config)
+
+        cached_file = hf_hub_download("hf-internal-testing/lcm-lora-test-sd-v1-5", "test_lora.safetensors")
+        lcm_lora = load_file(cached_file)
+
+        pipe.load_lora_weights(lcm_lora, adapter_name="lcm")
+        self.assertTrue(lcm_lora != {})
+        release_memory(pipe)
+
+    def test_load_unload_load_state_dict(self):
+        # Makes sure https://github.com/huggingface/diffusers/issues/7054 does not happen again
+        pipe = AutoPipelineForText2Image.from_pretrained(
+            "runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16
+        ).to("cuda")
+        pipe.scheduler = LCMScheduler.from_config(pipe.scheduler.config)
+
+        cached_file = hf_hub_download("hf-internal-testing/lcm-lora-test-sd-v1-5", "test_lora.safetensors")
+        lcm_lora = load_file(cached_file)
+        previous_state_dict = lcm_lora.copy()
+
+        pipe.load_lora_weights(lcm_lora, adapter_name="lcm")
+        self.assertDictEqual(lcm_lora, previous_state_dict)
+
+        pipe.unload_lora_weights()
+        pipe.load_lora_weights(lcm_lora, adapter_name="lcm")
+        self.assertDictEqual(lcm_lora, previous_state_dict)
+
+        release_memory(pipe)
+
 
 @slow
 @require_torch_gpu