add_example_diffusers_test

Author: Cui-yshoho

docs/diffusers/imgs/access_request.png

@@ -29,7 +29,7 @@ cd mindone
 pip install -e .
 # NOTE: transformers requires >=4.46.0
 
-cd examples/cogview
+cd examples/diffusers/cogview
 ```
 
 

docs/diffusers/imgs/diffusers_library.jpg

@@ -0,0 +1,188 @@
+# coding=utf-8
+# Copyright 2025 HuggingFace Inc.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import logging
+import os
+import sys
+import tempfile
+
+sys.path.append("..")
+from examples.diffusers.test_examples_utils import ExamplesTests, run_command  # noqa: E402
+
+ExamplesTests._launch_args = ["python"]
+
+logging.basicConfig(level=logging.DEBUG)
+
+logger = logging.getLogger()
+stream_handler = logging.StreamHandler(sys.stdout)
+logger.addHandler(stream_handler)
+
+
+class ControlNet(ExamplesTests):
+    def test_controlnet_checkpointing_checkpoints_total_limit(self):
+        with tempfile.TemporaryDirectory() as tmpdir:
+            test_args = f"""
+            examples/diffusers/controlnet/train_controlnet.py
+            --pretrained_model_name_or_path=hf-internal-testing/tiny-stable-diffusion-pipe
+            --revision refs/pr/4
+            --dataset_name=hf-internal-testing/fill10
+            --output_dir={tmpdir}
+            --resolution=64
+            --train_batch_size=1
+            --gradient_accumulation_steps=1
+            --max_train_steps=6
+            --checkpoints_total_limit=2
+            --checkpointing_steps=2
+            --controlnet_model_name_or_path=hf-internal-testing/tiny-controlnet
+            """.split()
+
+            run_command(self._launch_args + test_args)
+
+            self.assertEqual(
+                {x for x in os.listdir(tmpdir) if "checkpoint" in x},
+                {"checkpoint-4", "checkpoint-6"},
+            )
+
+    def test_controlnet_checkpointing_checkpoints_total_limit_removes_multiple_checkpoints(self):
+        with tempfile.TemporaryDirectory() as tmpdir:
+            test_args = f"""
+            examples/diffusers/controlnet/train_controlnet.py
+            --pretrained_model_name_or_path=hf-internal-testing/tiny-stable-diffusion-pipe
+            --revision refs/pr/4
+            --dataset_name=hf-internal-testing/fill10
+            --output_dir={tmpdir}
+            --resolution=64
+            --train_batch_size=1
+            --gradient_accumulation_steps=1
+            --controlnet_model_name_or_path=hf-internal-testing/tiny-controlnet
+            --max_train_steps=6
+            --checkpointing_steps=2
+            """.split()
+
+            run_command(self._launch_args + test_args)
+
+            self.assertEqual(
+                {x for x in os.listdir(tmpdir) if "checkpoint" in x},
+                {"checkpoint-2", "checkpoint-4", "checkpoint-6"},
+            )
+
+            resume_run_args = f"""
+            examples/diffusers/controlnet/train_controlnet.py
+            --pretrained_model_name_or_path=hf-internal-testing/tiny-stable-diffusion-pipe
+            --revision refs/pr/4
+            --dataset_name=hf-internal-testing/fill10
+            --output_dir={tmpdir}
+            --resolution=64
+            --train_batch_size=1
+            --gradient_accumulation_steps=1
+            --controlnet_model_name_or_path=hf-internal-testing/tiny-controlnet
+            --max_train_steps=8
+            --checkpointing_steps=2
+            --resume_from_checkpoint=checkpoint-6
+            --checkpoints_total_limit=2
+            """.split()
+
+            run_command(self._launch_args + resume_run_args)
+
+            self.assertEqual({x for x in os.listdir(tmpdir) if "checkpoint" in x}, {"checkpoint-6", "checkpoint-8"})
+
+
+class ControlNetSDXL(ExamplesTests):
+    def test_controlnet_sdxl(self):
+        with tempfile.TemporaryDirectory() as tmpdir:
+            test_args = f"""
+            examples/diffusers/controlnet/train_controlnet_sdxl.py
+            --pretrained_model_name_or_path=hf-internal-testing/tiny-stable-diffusion-xl-pipe
+            --revision refs/pr/2
+            --dataset_name=hf-internal-testing/fill10
+            --output_dir={tmpdir}
+            --resolution=64
+            --train_batch_size=1
+            --gradient_accumulation_steps=1
+            --controlnet_model_name_or_path=hf-internal-testing/tiny-controlnet-sdxl
+            --max_train_steps=4
+            --checkpointing_steps=2
+            """.split()
+
+            run_command(self._launch_args + test_args)
+
+            self.assertTrue(os.path.isfile(os.path.join(tmpdir, "diffusion_pytorch_model.safetensors")))
+
+
+class ControlNetSD3(ExamplesTests):
+    def test_controlnet_sd3(self):
+        with tempfile.TemporaryDirectory() as tmpdir:
+            test_args = f"""
+            examples/diffusers/controlnet/train_controlnet_sd3.py
+            --pretrained_model_name_or_path=DavyMorgan/tiny-sd3-pipe
+            --dataset_name=hf-internal-testing/fill10
+            --output_dir={tmpdir}
+            --resolution=64
+            --train_batch_size=1
+            --gradient_accumulation_steps=1
+            --controlnet_model_name_or_path=DavyMorgan/tiny-controlnet-sd3
+            --max_train_steps=4
+            --checkpointing_steps=2
+            """.split()
+
+            run_command(self._launch_args + test_args)
+
+            self.assertTrue(os.path.isfile(os.path.join(tmpdir, "diffusion_pytorch_model.safetensors")))
+
+
+class ControlNetSD35(ExamplesTests):
+    def test_controlnet_sd3(self):
+        with tempfile.TemporaryDirectory() as tmpdir:
+            test_args = f"""
+            examples/diffusers/controlnet/train_controlnet_sd3.py
+            --pretrained_model_name_or_path=hf-internal-testing/tiny-sd35-pipe
+            --dataset_name=hf-internal-testing/fill10
+            --output_dir={tmpdir}
+            --resolution=64
+            --train_batch_size=1
+            --gradient_accumulation_steps=1
+            --controlnet_model_name_or_path=DavyMorgan/tiny-controlnet-sd35
+            --max_train_steps=4
+            --checkpointing_steps=2
+            """.split()
+
+            run_command(self._launch_args + test_args)
+
+            self.assertTrue(os.path.isfile(os.path.join(tmpdir, "diffusion_pytorch_model.safetensors")))
+
+
+class ControlNetflux(ExamplesTests):
+    def test_controlnet_flux(self):
+        with tempfile.TemporaryDirectory() as tmpdir:
+            test_args = f"""
+            examples/diffusers/controlnet/train_controlnet_flux.py
+            --pretrained_model_name_or_path=hf-internal-testing/tiny-flux-pipe
+            --output_dir={tmpdir}
+            --dataset_name=hf-internal-testing/fill10
+            --conditioning_image_column=conditioning_image
+            --image_column=image
+            --caption_column=text
+            --resolution=64
+            --train_batch_size=1
+            --gradient_accumulation_steps=1
+            --max_train_steps=4
+            --checkpointing_steps=2
+            --num_double_layers=1
+            --num_single_layers=1
+            """.split()
+
+            run_command(self._launch_args + test_args)
+
+            self.assertTrue(os.path.isfile(os.path.join(tmpdir, "diffusion_pytorch_model.safetensors")))

examples/diffusers/cogview/README.md

@@ -879,8 +879,8 @@ def __len__(self):
             if is_master(args):
                 logger.info(f"Resuming from checkpoint {path}")
             # TODO: load optimizer & grad scaler etc. like accelerator.load_state
-            input_model_file = os.path.join(args.output_dir, path, "pytorch_model.ckpt")
-            ms.load_param_into_net(unet, ms.load_checkpoint(input_model_file), strict_load=True)
+            input_model_file = os.path.join(args.output_dir, path, "unet/diffusion_pytorch_model.safetensors")
+            ms.load_param_into_net(unet, ms.load_checkpoint(input_model_file, format="safetensors"), strict_load=True)
             global_step = int(path.split("-")[1])
 
             initial_global_step = global_step
@@ -939,8 +939,7 @@ def __len__(self):
                         save_path = os.path.join(args.output_dir, f"checkpoint-{global_step}")
                         # TODO: save optimizer & grad scaler etc. like accelerator.save_state
                         os.makedirs(save_path, exist_ok=True)
-                        output_model_file = os.path.join(save_path, "pytorch_model.ckpt")
-                        ms.save_checkpoint(unet, output_model_file)
+                        unet.save_pretrained(os.path.join(save_path, "unet"))
                         logger.info(f"Saved state to {save_path}")
 
                 if args.validation_prompt is not None and global_step % args.validation_steps == 0:

examples/diffusers/controlnet/test_controlnet.py

@@ -990,8 +990,8 @@ def __len__(self):
             if is_master(args):
                 logger.info(f"Resuming from checkpoint {path}")
             # TODO: load optimizer & grad scaler etc. like accelerator.load_state
-            input_model_file = os.path.join(args.output_dir, path, "pytorch_model.ckpt")
-            ms.load_param_into_net(unet, ms.load_checkpoint(input_model_file), strict_load=True)
+            input_model_file = os.path.join(args.output_dir, path, "unet/diffusion_pytorch_model.safetensors")
+            ms.load_param_into_net(unet, ms.load_checkpoint(input_model_file, format="safetensors"), strict_load=True)
             global_step = int(path.split("-")[1])
 
             initial_global_step = global_step
@@ -1050,8 +1050,7 @@ def __len__(self):
                         save_path = os.path.join(args.output_dir, f"checkpoint-{global_step}")
                         # TODO: save optimizer & grad scaler etc. like accelerator.save_state
                         os.makedirs(save_path, exist_ok=True)
-                        output_model_file = os.path.join(save_path, "pytorch_model.ckpt")
-                        ms.save_checkpoint(unet, output_model_file)
+                        unet.save_pretrained(os.path.join(save_path, "unet"))
                         logger.info(f"Saved state to {save_path}")
 
                 if args.validation_prompt is not None and global_step % args.validation_steps == 0: