Merge branch 'main' into cogvideox-ddim-inversion

Author: linoytsaban

.github/workflows/pr_style_bot.yml

@@ -9,160 +9,43 @@ permissions:
   pull-requests: write
 
 jobs:
-  check-permissions:
-    if: >
-      contains(github.event.comment.body, '@bot /style') &&
-      github.event.issue.pull_request != null
-    runs-on: ubuntu-latest
-    outputs:
-      is_authorized: ${{ steps.check_user_permission.outputs.has_permission }}
-    steps:
-      - name: Check user permission
-        id: check_user_permission
-        uses: actions/github-script@v6
-        with:
-          script: |
-            const comment_user = context.payload.comment.user.login;
-            const { data: permission } = await github.rest.repos.getCollaboratorPermissionLevel({
-              owner: context.repo.owner,
-              repo: context.repo.repo,
-              username: comment_user
-            });
-            const authorized = permission.permission === 'admin';
-            console.log(`User ${comment_user} has permission level: ${permission.permission}, authorized: ${authorized} (only admins allowed)`);
-            core.setOutput('has_permission', authorized);
-
-  run-style-bot:
-    needs: check-permissions
-    if: needs.check-permissions.outputs.is_authorized == 'true'
-    runs-on: ubuntu-latest
-    steps:
-      - name: Extract PR details
-        id: pr_info
-        uses: actions/github-script@v6
-        with:
-          script: |
-            const prNumber = context.payload.issue.number;
-            const { data: pr } = await github.rest.pulls.get({
-              owner: context.repo.owner,
-              repo: context.repo.repo,
-              pull_number: prNumber
-            });
-            
-            // We capture both the branch ref and the "full_name" of the head repo
-            // so that we can check out the correct repository & branch (including forks).
-            core.setOutput("prNumber", prNumber);
-            core.setOutput("headRef", pr.head.ref);
-            core.setOutput("headRepoFullName", pr.head.repo.full_name);
-
-      - name: Check out PR branch
-        uses: actions/checkout@v3
-        env: 
-          HEADREPOFULLNAME: ${{ steps.pr_info.outputs.headRepoFullName }}
-          HEADREF: ${{ steps.pr_info.outputs.headRef }}
-        with:
-          # Instead of checking out the base repo, use the contributor's repo name
-          repository: ${{ env.HEADREPOFULLNAME }}
-          ref: ${{ env.HEADREF }}
-          # You may need fetch-depth: 0 for being able to push
-          fetch-depth: 0
-          token: ${{ secrets.GITHUB_TOKEN }}
-      
-      - name: Debug
-        env: 
-          HEADREPOFULLNAME: ${{ steps.pr_info.outputs.headRepoFullName }}
-          HEADREF: ${{ steps.pr_info.outputs.headRef }}
-          PRNUMBER: ${{ steps.pr_info.outputs.prNumber }}
-        run: |
-          echo "PR number: $PRNUMBER"
-          echo "Head Ref: $HEADREF"
-          echo "Head Repo Full Name: $HEADREPOFULLNAME"
-
-      - name: Set up Python
-        uses: actions/setup-python@v4
-
-      - name: Install dependencies
-        run: |
-          pip install .[quality]
-
-      - name: Download necessary files from main branch of Diffusers
-        run: |
-          curl -o main_Makefile https://raw.githubusercontent.com/huggingface/diffusers/main/Makefile
-          curl -o main_setup.py https://raw.githubusercontent.com/huggingface/diffusers/refs/heads/main/setup.py
-          curl -o main_check_doc_toc.py https://raw.githubusercontent.com/huggingface/diffusers/refs/heads/main/utils/check_doc_toc.py
-        
-      - name: Compare the files and raise error if needed
-        run: |
-          diff_failed=0
-
-          if ! diff -q main_Makefile Makefile; then
-            echo "Error: The Makefile has changed. Please ensure it matches the main branch."
-            diff_failed=1
-          fi
-
-          if ! diff -q main_setup.py setup.py; then
-            echo "Error: The setup.py has changed. Please ensure it matches the main branch."
-            diff_failed=1
-          fi
-
-          if ! diff -q main_check_doc_toc.py utils/check_doc_toc.py; then
-            echo "Error: The utils/check_doc_toc.py has changed. Please ensure it matches the main branch."
-            diff_failed=1
-          fi
-
-          if [ $diff_failed -eq 1 ]; then
-            echo "❌ Error happened as we detected changes in the files that should not be changed ❌"
-            exit 1
-          fi
-
-          echo "No changes in the files. Proceeding..."
-          rm -rf main_Makefile main_setup.py main_check_doc_toc.py
-
-      - name: Run make style and make quality
-        run: |
-          make style && make quality
-
-      - name: Commit and push changes
-        id: commit_and_push
-        env: 
-          HEADREPOFULLNAME: ${{ steps.pr_info.outputs.headRepoFullName }}
-          HEADREF: ${{ steps.pr_info.outputs.headRef }}
-          PRNUMBER: ${{ steps.pr_info.outputs.prNumber }}
-          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-        run: |
-          echo "HEADREPOFULLNAME: $HEADREPOFULLNAME, HEADREF: $HEADREF"
-          # Configure git with the Actions bot user
-          git config user.name "github-actions[bot]"
-          git config user.email "github-actions[bot]@users.noreply.github.com"
-
-          # Make sure your 'origin' remote is set to the contributor's fork
-          git remote set-url origin "https://x-access-token:${GITHUB_TOKEN}@github.com/$HEADREPOFULLNAME.git"
-
-          # If there are changes after running style/quality, commit them
-          if [ -n "$(git status --porcelain)" ]; then
-            git add .
-            git commit -m "Apply style fixes"
-            # Push to the original contributor's forked branch
-            git push origin HEAD:$HEADREF
-            echo "changes_pushed=true" >> $GITHUB_OUTPUT
-          else
-            echo "No changes to commit."
-            echo "changes_pushed=false" >> $GITHUB_OUTPUT
-          fi
-
-      - name: Comment on PR with workflow run link
-        if: steps.commit_and_push.outputs.changes_pushed == 'true'
-        uses: actions/github-script@v6
-        with:
-          script: |
-            const prNumber = parseInt(process.env.prNumber, 10);
-            const runUrl = `${process.env.GITHUB_SERVER_URL}/${process.env.GITHUB_REPOSITORY}/actions/runs/${process.env.GITHUB_RUN_ID}`
-
-            await github.rest.issues.createComment({
-              owner: context.repo.owner,
-              repo: context.repo.repo,
-              issue_number: prNumber,
-              body: `Style fixes have been applied. [View the workflow run here](${runUrl}).`
-            });
-        env:
-          prNumber: ${{ steps.pr_info.outputs.prNumber }}
+  style:
+    uses: huggingface/huggingface_hub/.github/workflows/style-bot-action.yml@main
+    with:
+      python_quality_dependencies: "[quality]"
+      pre_commit_script_name: "Download and Compare files from the main branch"
+      pre_commit_script: |
+        echo "Downloading the files from the main branch"
+
+        curl -o main_Makefile https://raw.githubusercontent.com/huggingface/diffusers/main/Makefile
+        curl -o main_setup.py https://raw.githubusercontent.com/huggingface/diffusers/refs/heads/main/setup.py
+        curl -o main_check_doc_toc.py https://raw.githubusercontent.com/huggingface/diffusers/refs/heads/main/utils/check_doc_toc.py
+
+        echo "Compare the files and raise error if needed"
+
+        diff_failed=0
+        if ! diff -q main_Makefile Makefile; then
+          echo "Error: The Makefile has changed. Please ensure it matches the main branch."
+          diff_failed=1
+        fi
+
+        if ! diff -q main_setup.py setup.py; then
+          echo "Error: The setup.py has changed. Please ensure it matches the main branch."
+          diff_failed=1
+        fi
+
+        if ! diff -q main_check_doc_toc.py utils/check_doc_toc.py; then
+          echo "Error: The utils/check_doc_toc.py has changed. Please ensure it matches the main branch."
+          diff_failed=1
+        fi
+
+        if [ $diff_failed -eq 1 ]; then
+          echo "❌ Error happened as we detected changes in the files that should not be changed ❌"
+          exit 1
+        fi
+
+        echo "No changes in the files. Proceeding..."
+        rm -rf main_Makefile main_setup.py main_check_doc_toc.py
+      style_command: "make style && make quality"
+    secrets:
+      bot_token: ${{ secrets.GITHUB_TOKEN }}

.github/workflows/pr_tests.yml

@@ -3,7 +3,6 @@ name: Fast tests for PRs
 on:
   pull_request:
     branches: [main]
-    types: [synchronize]
     paths:
       - "src/diffusers/**.py"
       - "benchmarks/**.py"

examples/advanced_diffusion_training/train_dreambooth_lora_flux_advanced.py

@@ -880,9 +880,7 @@ def save_embeddings(self, file_path: str):
         idx_to_text_encoder_name = {0: "clip_l", 1: "t5"}
         for idx, text_encoder in enumerate(self.text_encoders):
             train_ids = self.train_ids if idx == 0 else self.train_ids_t5
-            embeds = (
-                text_encoder.text_model.embeddings.token_embedding if idx == 0 else text_encoder.encoder.embed_tokens
-            )
+            embeds = text_encoder.text_model.embeddings.token_embedding if idx == 0 else text_encoder.shared
             assert embeds.weight.data.shape[0] == len(self.tokenizers[idx]), "Tokenizers should be the same."
             new_token_embeddings = embeds.weight.data[train_ids]
 
@@ -904,9 +902,7 @@ def device(self):
     @torch.no_grad()
     def retract_embeddings(self):
         for idx, text_encoder in enumerate(self.text_encoders):
-            embeds = (
-                text_encoder.text_model.embeddings.token_embedding if idx == 0 else text_encoder.encoder.embed_tokens
-            )
+            embeds = text_encoder.text_model.embeddings.token_embedding if idx == 0 else text_encoder.shared
             index_no_updates = self.embeddings_settings[f"index_no_updates_{idx}"]
             embeds.weight.data[index_no_updates] = (
                 self.embeddings_settings[f"original_embeddings_{idx}"][index_no_updates]
@@ -1749,7 +1745,7 @@ def load_model_hook(models, input_dir):
         if args.enable_t5_ti:  # whether to do pivotal tuning/textual inversion for T5 as well
             text_lora_parameters_two = []
             for name, param in text_encoder_two.named_parameters():
-                if "token_embedding" in name:
+                if "shared" in name:
                     # ensure that dtype is float32, even if rest of the model that isn't trained is loaded in fp16
                     param.data = param.to(dtype=torch.float32)
                     param.requires_grad = True

examples/dreambooth/train_dreambooth_lora.py

@@ -1119,17 +1119,22 @@ def compute_text_embeddings(prompt):
     )
 
     # Scheduler and math around the number of training steps.
-    overrode_max_train_steps = False
-    num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
+    # Check the PR https://github.com/huggingface/diffusers/pull/8312 for detailed explanation.
+    num_warmup_steps_for_scheduler = args.lr_warmup_steps * accelerator.num_processes
     if args.max_train_steps is None:
-        args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch
-        overrode_max_train_steps = True
+        len_train_dataloader_after_sharding = math.ceil(len(train_dataloader) / accelerator.num_processes)
+        num_update_steps_per_epoch = math.ceil(len_train_dataloader_after_sharding / args.gradient_accumulation_steps)
+        num_training_steps_for_scheduler = (
+            args.num_train_epochs * accelerator.num_processes * num_update_steps_per_epoch
+        )
+    else:
+        num_training_steps_for_scheduler = args.max_train_steps * accelerator.num_processes
 
     lr_scheduler = get_scheduler(
         args.lr_scheduler,
         optimizer=optimizer,
-        num_warmup_steps=args.lr_warmup_steps * accelerator.num_processes,
-        num_training_steps=args.max_train_steps * accelerator.num_processes,
+        num_warmup_steps=num_warmup_steps_for_scheduler,
+        num_training_steps=num_training_steps_for_scheduler,
         num_cycles=args.lr_num_cycles,
         power=args.lr_power,
     )
@@ -1146,8 +1151,15 @@ def compute_text_embeddings(prompt):
 
     # 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:
+    if args.max_train_steps is None:
         args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch
+        if num_training_steps_for_scheduler != args.max_train_steps:
+            logger.warning(
+                f"The length of the 'train_dataloader' after 'accelerator.prepare' ({len(train_dataloader)}) does not match "
+                f"the expected length ({len_train_dataloader_after_sharding}) when the learning rate scheduler was created. "
+                f"This inconsistency may result in the learning rate scheduler not functioning properly."
+            )
+
     # Afterwards we recalculate our number of training epochs
     args.num_train_epochs = math.ceil(args.max_train_steps / num_update_steps_per_epoch)
 

examples/research_projects/realfill/requirements.txt

@@ -6,4 +6,4 @@ torch==2.2.0
 torchvision>=0.16
 ftfy==6.1.1
 tensorboard==2.14.0
-Jinja2==3.1.5
+Jinja2==3.1.6

src/diffusers/loaders/lora_conversion_utils.py

@@ -654,6 +654,7 @@ def _convert(original_key, diffusers_key, state_dict, new_state_dict):
 
                 _convert(k, diffusers_key, state_dict, new_state_dict)
 
+        remaining_all_unet = False
         if state_dict:
             remaining_all_unet = all(k.startswith("lora_unet_") for k in state_dict)
         if remaining_all_unet:

src/diffusers/pipelines/flux/pipeline_flux.py

@@ -694,7 +694,7 @@ def __call__(
                 Custom sigmas to use for the denoising process with schedulers which support a `sigmas` argument in
                 their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is passed
                 will be used.
-            guidance_scale (`float`, *optional*, defaults to 7.0):
+            guidance_scale (`float`, *optional*, defaults to 3.5):
                 Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://arxiv.org/abs/2207.12598).
                 `guidance_scale` is defined as `w` of equation 2. of [Imagen
                 Paper](https://arxiv.org/pdf/2205.11487.pdf). Guidance scale is enabled by setting `guidance_scale >

src/diffusers/pipelines/flux/pipeline_flux_control.py

@@ -660,7 +660,7 @@ def __call__(
                 Custom sigmas to use for the denoising process with schedulers which support a `sigmas` argument in
                 their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is passed
                 will be used.
-            guidance_scale (`float`, *optional*, defaults to 7.0):
+            guidance_scale (`float`, *optional*, defaults to 3.5):
                 Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://arxiv.org/abs/2207.12598).
                 `guidance_scale` is defined as `w` of equation 2. of [Imagen
                 Paper](https://arxiv.org/pdf/2205.11487.pdf). Guidance scale is enabled by setting `guidance_scale >

src/diffusers/pipelines/flux/pipeline_flux_fill.py

@@ -738,7 +738,7 @@ def __call__(
                 Custom sigmas to use for the denoising process with schedulers which support a `sigmas` argument in
                 their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is passed
                 will be used.
-            guidance_scale (`float`, *optional*, defaults to 7.0):
+            guidance_scale (`float`, *optional*, defaults to 30.0):
                 Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://arxiv.org/abs/2207.12598).
                 `guidance_scale` is defined as `w` of equation 2. of [Imagen
                 Paper](https://arxiv.org/pdf/2205.11487.pdf). Guidance scale is enabled by setting `guidance_scale >