v-prediction training support (huggingface#1455)

* add get_velocity

* add v prediction for training

* fix saving

* add revision arg

* fix saving

* save checkpoints dreambooth

* fix saving embeds

* add instruction in readme

* quality

* noise_pred -> model_pred

examples/dreambooth/README.md

@@ -39,6 +39,8 @@ Now let's get our dataset. Download images from [here](https://drive.google.com/
 
 And launch the training using
 
+**___Note: Change the `resolution` to 768 if you are using the [stable-diffusion-2](https://huggingface.co/stabilityai/stable-diffusion-2) 768x768 model.___**
+
 ```bash
 export MODEL_NAME="CompVis/stable-diffusion-v1-4"
 export INSTANCE_DIR="path-to-instance-images"

examples/dreambooth/train_dreambooth.py

@@ -124,6 +124,7 @@ def parse_args(input_args=None):
         default=None,
         help="Total number of training steps to perform.  If provided, overrides num_train_epochs.",
     )
+    parser.add_argument("--save_steps", type=int, default=500, help="Save checkpoint every X updates steps.")
     parser.add_argument(
         "--gradient_accumulation_steps",
         type=int,
@@ -603,23 +604,31 @@ def collate_fn(examples):
                 encoder_hidden_states = text_encoder(batch["input_ids"])[0]
 
                 # Predict the noise residual
-                noise_pred = unet(noisy_latents, timesteps, encoder_hidden_states).sample
+                model_pred = unet(noisy_latents, timesteps, encoder_hidden_states).sample
+
+                # Get the target for loss depending on the prediction type
+                if noise_scheduler.config.prediction_type == "epsilon":
+                    target = noise
+                elif noise_scheduler.config.prediction_type == "v_prediction":
+                    target = noise_scheduler.get_velocity(latents, noise, timesteps)
+                else:
+                    raise ValueError(f"Unknown prediction type {noise_scheduler.config.prediction_type}")
 
                 if args.with_prior_preservation:
-                    # Chunk the noise and noise_pred into two parts and compute the loss on each part separately.
-                    noise_pred, noise_pred_prior = torch.chunk(noise_pred, 2, dim=0)
-                    noise, noise_prior = torch.chunk(noise, 2, dim=0)
+                    # Chunk the noise and model_pred into two parts and compute the loss on each part separately.
+                    model_pred, model_pred_prior = torch.chunk(model_pred, 2, dim=0)
+                    target, target_prior = torch.chunk(target, 2, dim=0)
 
                     # Compute instance loss
-                    loss = F.mse_loss(noise_pred.float(), noise.float(), reduction="none").mean([1, 2, 3]).mean()
+                    loss = F.mse_loss(model_pred.float(), target.float(), reduction="none").mean([1, 2, 3]).mean()
 
                     # Compute prior loss
-                    prior_loss = F.mse_loss(noise_pred_prior.float(), noise_prior.float(), reduction="mean")
+                    prior_loss = F.mse_loss(model_pred_prior.float(), target_prior.float(), reduction="mean")
 
                     # Add the prior loss to the instance loss.
                     loss = loss + args.prior_loss_weight * prior_loss
                 else:
-                    loss = F.mse_loss(noise_pred.float(), noise.float(), reduction="mean")
+                    loss = F.mse_loss(model_pred.float(), target.float(), reduction="mean")
 
                 accelerator.backward(loss)
                 if accelerator.sync_gradients:
@@ -638,6 +647,17 @@ def collate_fn(examples):
                 progress_bar.update(1)
                 global_step += 1
 
+                if global_step % args.save_steps == 0:
+                    if accelerator.is_main_process:
+                        pipeline = StableDiffusionPipeline.from_pretrained(
+                            args.pretrained_model_name_or_path,
+                            unet=accelerator.unwrap_model(unet),
+                            text_encoder=accelerator.unwrap_model(text_encoder),
+                            revision=args.revision,
+                        )
+                        save_path = os.path.join(args.output_dir, f"checkpoint-{global_step}")
+                        pipeline.save_pretrained(save_path)
+
             logs = {"loss": loss.detach().item(), "lr": lr_scheduler.get_last_lr()[0]}
             progress_bar.set_postfix(**logs)
             accelerator.log(logs, step=global_step)

examples/text_to_image/README.md

@@ -42,6 +42,8 @@ If you have already cloned the repo, then you won't need to go through these ste
 #### Hardware
 With `gradient_checkpointing` and `mixed_precision` it should be possible to fine tune the model on a single 24GB GPU. For higher `batch_size` and faster training it's better to use GPUs with >30GB memory.
 
+**___Note: Change the `resolution` to 768 if you are using the [stable-diffusion-2](https://huggingface.co/stabilityai/stable-diffusion-2) 768x768 model.___**
+
 ```bash
 export MODEL_NAME="CompVis/stable-diffusion-v1-4"
 export dataset_name="lambdalabs/pokemon-blip-captions"

examples/text_to_image/train_text_to_image.py

@@ -15,13 +15,12 @@
 from accelerate.logging import get_logger
 from accelerate.utils import set_seed
 from datasets import load_dataset
-from diffusers import AutoencoderKL, DDPMScheduler, PNDMScheduler, StableDiffusionPipeline, UNet2DConditionModel
+from diffusers import AutoencoderKL, DDPMScheduler, StableDiffusionPipeline, UNet2DConditionModel
 from diffusers.optimization import get_scheduler
-from diffusers.pipelines.stable_diffusion import StableDiffusionSafetyChecker
 from huggingface_hub import HfFolder, Repository, whoami
 from torchvision import transforms
 from tqdm.auto import tqdm
-from transformers import CLIPFeatureExtractor, CLIPTextModel, CLIPTokenizer
+from transformers import CLIPTextModel, CLIPTokenizer
 
 
 logger = get_logger(__name__)
@@ -36,6 +35,13 @@ def parse_args():
         required=True,
         help="Path to pretrained model or model identifier from huggingface.co/models.",
     )
+    parser.add_argument(
+        "--revision",
+        type=str,
+        default=None,
+        required=False,
+        help="Revision of pretrained model identifier from huggingface.co/models.",
+    )
     parser.add_argument(
         "--dataset_name",
         type=str,
@@ -335,10 +341,24 @@ def main():
             os.makedirs(args.output_dir, exist_ok=True)
 
     # Load models and create wrapper for stable diffusion
-    tokenizer = CLIPTokenizer.from_pretrained(args.pretrained_model_name_or_path, subfolder="tokenizer")
-    text_encoder = CLIPTextModel.from_pretrained(args.pretrained_model_name_or_path, subfolder="text_encoder")
-    vae = AutoencoderKL.from_pretrained(args.pretrained_model_name_or_path, subfolder="vae")
-    unet = UNet2DConditionModel.from_pretrained(args.pretrained_model_name_or_path, subfolder="unet")
+    tokenizer = CLIPTokenizer.from_pretrained(
+        args.pretrained_model_name_or_path, subfolder="tokenizer", revision=args.revision
+    )
+    text_encoder = CLIPTextModel.from_pretrained(
+        args.pretrained_model_name_or_path,
+        subfolder="text_encoder",
+        revision=args.revision,
+    )
+    vae = AutoencoderKL.from_pretrained(
+        args.pretrained_model_name_or_path,
+        subfolder="vae",
+        revision=args.revision,
+    )
+    unet = UNet2DConditionModel.from_pretrained(
+        args.pretrained_model_name_or_path,
+        subfolder="unet",
+        revision=args.revision,
+    )
 
     # Freeze vae and text_encoder
     vae.requires_grad_(False)
@@ -562,9 +582,17 @@ def collate_fn(examples):
                 # Get the text embedding for conditioning
                 encoder_hidden_states = text_encoder(batch["input_ids"])[0]
 
+                # Get the target for loss depending on the prediction type
+                if noise_scheduler.config.prediction_type == "epsilon":
+                    target = noise
+                elif noise_scheduler.config.prediction_type == "v_prediction":
+                    target = noise_scheduler.get_velocity(latents, noise, timesteps)
+                else:
+                    raise ValueError(f"Unknown prediction type {noise_scheduler.config.prediction_type}")
+
                 # Predict the noise residual and compute loss
-                noise_pred = unet(noisy_latents, timesteps, encoder_hidden_states).sample
-                loss = F.mse_loss(noise_pred.float(), noise.float(), reduction="mean")
+                model_pred = unet(noisy_latents, timesteps, encoder_hidden_states).sample
+                loss = F.mse_loss(model_pred.float(), target.float(), reduction="mean")
 
                 # Gather the losses across all processes for logging (if we use distributed training).
                 avg_loss = accelerator.gather(loss.repeat(args.train_batch_size)).mean()
@@ -600,14 +628,12 @@ def collate_fn(examples):
         if args.use_ema:
             ema_unet.copy_to(unet.parameters())
 
-        pipeline = StableDiffusionPipeline(
+        pipeline = StableDiffusionPipeline.from_pretrained(
+            args.pretrained_model_name_or_path,
             text_encoder=text_encoder,
             vae=vae,
             unet=unet,
-            tokenizer=tokenizer,
-            scheduler=PNDMScheduler.from_config(args.pretrained_model_name_or_path, subfolder="scheduler"),
-            safety_checker=StableDiffusionSafetyChecker.from_pretrained("CompVis/stable-diffusion-safety-checker"),
-            feature_extractor=CLIPFeatureExtractor.from_pretrained("openai/clip-vit-base-patch32"),
+            revision=args.revision,
         )
         pipeline.save_pretrained(args.output_dir)
 

examples/textual_inversion/README.md

@@ -47,6 +47,8 @@ Now let's get our dataset.Download 3-4 images from [here](https://drive.google.c
 
 And launch the training using
 
+**___Note: Change the `resolution` to 768 if you are using the [stable-diffusion-2](https://huggingface.co/stabilityai/stable-diffusion-2) 768x768 model.___**
+
 ```bash
 export MODEL_NAME="runwayml/stable-diffusion-v1-5"
 export DATA_DIR="path-to-dir-containing-images"

examples/textual_inversion/textual_inversion.py

@@ -16,17 +16,16 @@
 from accelerate import Accelerator
 from accelerate.logging import get_logger
 from accelerate.utils import set_seed
-from diffusers import AutoencoderKL, DDPMScheduler, PNDMScheduler, StableDiffusionPipeline, UNet2DConditionModel
+from diffusers import AutoencoderKL, DDPMScheduler, StableDiffusionPipeline, UNet2DConditionModel
 from diffusers.optimization import get_scheduler
-from diffusers.pipelines.stable_diffusion import StableDiffusionSafetyChecker
 from huggingface_hub import HfFolder, Repository, whoami
 
 # TODO: remove and import from diffusers.utils when the new version of diffusers is released
 from packaging import version
 from PIL import Image
 from torchvision import transforms
 from tqdm.auto import tqdm
-from transformers import CLIPFeatureExtractor, CLIPTextModel, CLIPTokenizer
+from transformers import CLIPTextModel, CLIPTokenizer
 
 
 if version.parse(version.parse(PIL.__version__).base_version) >= version.parse("9.1.0"):
@@ -51,11 +50,11 @@
 logger = get_logger(__name__)
 
 
-def save_progress(text_encoder, placeholder_token_id, accelerator, args):
+def save_progress(text_encoder, placeholder_token_id, accelerator, args, save_path):
     logger.info("Saving embeddings")
     learned_embeds = accelerator.unwrap_model(text_encoder).get_input_embeddings().weight[placeholder_token_id]
     learned_embeds_dict = {args.placeholder_token: learned_embeds.detach().cpu()}
-    torch.save(learned_embeds_dict, os.path.join(args.output_dir, "learned_embeds.bin"))
+    torch.save(learned_embeds_dict, save_path)
 
 
 def parse_args():
@@ -73,6 +72,13 @@ def parse_args():
         required=True,
         help="Path to pretrained model or model identifier from huggingface.co/models.",
     )
+    parser.add_argument(
+        "--revision",
+        type=str,
+        default=None,
+        required=False,
+        help="Revision of pretrained model identifier from huggingface.co/models.",
+    )
     parser.add_argument(
         "--tokenizer_name",
         type=str,
@@ -405,9 +411,21 @@ def main():
     placeholder_token_id = tokenizer.convert_tokens_to_ids(args.placeholder_token)
 
     # Load models and create wrapper for stable diffusion
-    text_encoder = CLIPTextModel.from_pretrained(args.pretrained_model_name_or_path, subfolder="text_encoder")
-    vae = AutoencoderKL.from_pretrained(args.pretrained_model_name_or_path, subfolder="vae")
-    unet = UNet2DConditionModel.from_pretrained(args.pretrained_model_name_or_path, subfolder="unet")
+    text_encoder = CLIPTextModel.from_pretrained(
+        args.pretrained_model_name_or_path,
+        subfolder="text_encoder",
+        revision=args.revision,
+    )
+    vae = AutoencoderKL.from_pretrained(
+        args.pretrained_model_name_or_path,
+        subfolder="vae",
+        revision=args.revision,
+    )
+    unet = UNet2DConditionModel.from_pretrained(
+        args.pretrained_model_name_or_path,
+        subfolder="unet",
+        revision=args.revision,
+    )
 
     # Resize the token embeddings as we are adding new special tokens to the tokenizer
     text_encoder.resize_token_embeddings(len(tokenizer))
@@ -532,9 +550,17 @@ def main():
                 encoder_hidden_states = text_encoder(batch["input_ids"])[0]
 
                 # Predict the noise residual
-                noise_pred = unet(noisy_latents, timesteps, encoder_hidden_states).sample
+                model_pred = unet(noisy_latents, timesteps, encoder_hidden_states).sample
+
+                # Get the target for loss depending on the prediction type
+                if noise_scheduler.config.prediction_type == "epsilon":
+                    target = noise
+                elif noise_scheduler.config.prediction_type == "v_prediction":
+                    target = noise_scheduler.get_velocity(latents, noise, timesteps)
+                else:
+                    raise ValueError(f"Unknown prediction type {noise_scheduler.config.prediction_type}")
 
-                loss = F.mse_loss(noise_pred, noise, reduction="none").mean([1, 2, 3]).mean()
+                loss = F.mse_loss(model_pred, target, reduction="none").mean([1, 2, 3]).mean()
                 accelerator.backward(loss)
 
                 # Zero out the gradients for all token embeddings except the newly added
@@ -556,7 +582,8 @@ def main():
                 progress_bar.update(1)
                 global_step += 1
                 if global_step % args.save_steps == 0:
-                    save_progress(text_encoder, placeholder_token_id, accelerator, args)
+                    save_path = os.path.join(args.output_dir, f"learned_embeds-steps-{global_step}.bin")
+                    save_progress(text_encoder, placeholder_token_id, accelerator, args, save_path)
 
             logs = {"loss": loss.detach().item(), "lr": lr_scheduler.get_last_lr()[0]}
             progress_bar.set_postfix(**logs)
@@ -569,18 +596,18 @@ def main():
 
     # Create the pipeline using using the trained modules and save it.
     if accelerator.is_main_process:
-        pipeline = StableDiffusionPipeline(
+        pipeline = StableDiffusionPipeline.from_pretrained(
+            args.pretrained_model_name_or_path,
             text_encoder=accelerator.unwrap_model(text_encoder),
+            tokenizer=tokenizer,
             vae=vae,
             unet=unet,
-            tokenizer=tokenizer,
-            scheduler=PNDMScheduler.from_config(args.pretrained_model_name_or_path, subfolder="scheduler"),
-            safety_checker=StableDiffusionSafetyChecker.from_pretrained("CompVis/stable-diffusion-safety-checker"),
-            feature_extractor=CLIPFeatureExtractor.from_pretrained("openai/clip-vit-base-patch32"),
+            revision=args.revision,
         )
         pipeline.save_pretrained(args.output_dir)
         # Also save the newly trained embeddings
-        save_progress(text_encoder, placeholder_token_id, accelerator, args)
+        save_path = os.path.join(args.output_dir, "learned_embeds.bin")
+        save_progress(text_encoder, placeholder_token_id, accelerator, args, save_path)
 
         if args.push_to_hub:
             repo.push_to_hub(commit_message="End of training", blocking=False, auto_lfs_prune=True)

src/diffusers/schedulers/scheduling_ddim.py

@@ -355,5 +355,25 @@ def add_noise(
         noisy_samples = sqrt_alpha_prod * original_samples + sqrt_one_minus_alpha_prod * noise
         return noisy_samples
 
+    def get_velocity(
+        self, sample: torch.FloatTensor, noise: torch.FloatTensor, timesteps: torch.IntTensor
+    ) -> torch.FloatTensor:
+        # Make sure alphas_cumprod and timestep have same device and dtype as sample
+        self.alphas_cumprod = self.alphas_cumprod.to(device=sample.device, dtype=sample.dtype)
+        timesteps = timesteps.to(sample.device)
+
+        sqrt_alpha_prod = self.alphas_cumprod[timesteps] ** 0.5
+        sqrt_alpha_prod = sqrt_alpha_prod.flatten()
+        while len(sqrt_alpha_prod.shape) < len(sample.shape):
+            sqrt_alpha_prod = sqrt_alpha_prod.unsqueeze(-1)
+
+        sqrt_one_minus_alpha_prod = (1 - self.alphas_cumprod[timesteps]) ** 0.5
+        sqrt_one_minus_alpha_prod = sqrt_one_minus_alpha_prod.flatten()
+        while len(sqrt_one_minus_alpha_prod.shape) < len(sample.shape):
+            sqrt_one_minus_alpha_prod = sqrt_one_minus_alpha_prod.unsqueeze(-1)
+
+        velocity = sqrt_alpha_prod * noise - sqrt_one_minus_alpha_prod * sample
+        return velocity
+
     def __len__(self):
         return self.config.num_train_timesteps

src/diffusers/schedulers/scheduling_ddpm.py

@@ -345,5 +345,25 @@ def add_noise(
         noisy_samples = sqrt_alpha_prod * original_samples + sqrt_one_minus_alpha_prod * noise
         return noisy_samples
 
+    def get_velocity(
+        self, sample: torch.FloatTensor, noise: torch.FloatTensor, timesteps: torch.IntTensor
+    ) -> torch.FloatTensor:
+        # Make sure alphas_cumprod and timestep have same device and dtype as sample
+        self.alphas_cumprod = self.alphas_cumprod.to(device=sample.device, dtype=sample.dtype)
+        timesteps = timesteps.to(sample.device)
+
+        sqrt_alpha_prod = self.alphas_cumprod[timesteps] ** 0.5
+        sqrt_alpha_prod = sqrt_alpha_prod.flatten()
+        while len(sqrt_alpha_prod.shape) < len(sample.shape):
+            sqrt_alpha_prod = sqrt_alpha_prod.unsqueeze(-1)
+
+        sqrt_one_minus_alpha_prod = (1 - self.alphas_cumprod[timesteps]) ** 0.5
+        sqrt_one_minus_alpha_prod = sqrt_one_minus_alpha_prod.flatten()
+        while len(sqrt_one_minus_alpha_prod.shape) < len(sample.shape):
+            sqrt_one_minus_alpha_prod = sqrt_one_minus_alpha_prod.unsqueeze(-1)
+
+        velocity = sqrt_alpha_prod * noise - sqrt_one_minus_alpha_prod * sample
+        return velocity
+
     def __len__(self):
         return self.config.num_train_timesteps