Updated Dreambooth training

3 files changed, 76 insertions, 37 deletions

diff --git a/dreambooth.py b/dreambooth.py
index 9786e0f..d1cf535 100644
--- a/dreambooth.py
+++ b/dreambooth.py
@@ -1,4 +1,5 @@
 import argparse
+import itertools
 import math
 import os
 import datetime
@@ -113,7 +114,7 @@ def parse_args():
     parser.add_argument(
         "--max_train_steps",
         type=int,
-        default=1200,
+        default=3600,
         help="Total number of training steps to perform.  If provided, overrides num_train_epochs.",
     )
     parser.add_argument(
@@ -128,9 +129,15 @@ def parse_args():
         help="Whether or not to use gradient checkpointing to save memory at the expense of slower backward pass.",
     )
     parser.add_argument(
-        "--learning_rate",
+        "--learning_rate_unet",
         type=float,
-        default=5e-5,
+        default=3e-6,
+        help="Initial learning rate (after the potential warmup period) to use.",
+    )
+    parser.add_argument(
+        "--learning_rate_text",
+        type=float,
+        default=3e-6,
         help="Initial learning rate (after the potential warmup period) to use.",
     )
     parser.add_argument(
@@ -358,12 +365,14 @@ class Checkpointer:
     def save_model(self):
         print("Saving model...")
 
-        unwrapped = self.accelerator.unwrap_model(
+        unwrapped_unet = self.accelerator.unwrap_model(
             self.ema_unet.averaged_model if self.ema_unet is not None else self.unet)
+        unwrapped_text_encoder = self.accelerator.unwrap_model(self.text_encoder)
+
         pipeline = VlpnStableDiffusion(
-            text_encoder=self.text_encoder,
+            text_encoder=unwrapped_text_encoder,
             vae=self.vae,
-            unet=unwrapped,
+            unet=unwrapped_unet,
             tokenizer=self.tokenizer,
             scheduler=PNDMScheduler(
                 beta_start=0.00085, beta_end=0.012, beta_schedule="scaled_linear", skip_prk_steps=True
@@ -371,7 +380,8 @@ class Checkpointer:
         )
         pipeline.save_pretrained(self.output_dir.joinpath("model"))
 
-        del unwrapped
+        del unwrapped_unet
+        del unwrapped_text_encoder
         del pipeline
 
         if torch.cuda.is_available():
@@ -381,16 +391,18 @@ class Checkpointer:
     def save_samples(self, step, height, width, guidance_scale, eta, num_inference_steps):
         samples_path = Path(self.output_dir).joinpath("samples")
 
-        unwrapped = self.accelerator.unwrap_model(
+        unwrapped_unet = self.accelerator.unwrap_model(
             self.ema_unet.averaged_model if self.ema_unet is not None else self.unet)
+        unwrapped_text_encoder = self.accelerator.unwrap_model(self.text_encoder)
+
         scheduler = EulerAScheduler(
             beta_start=0.00085, beta_end=0.012, beta_schedule="scaled_linear"
         )
 
         pipeline = VlpnStableDiffusion(
-            text_encoder=self.text_encoder,
+            text_encoder=unwrapped_text_encoder,
             vae=self.vae,
-            unet=unwrapped,
+            unet=unwrapped_unet,
             tokenizer=self.tokenizer,
             scheduler=scheduler,
         ).to(self.accelerator.device)
@@ -416,9 +428,16 @@ class Checkpointer:
 
                 for i in range(self.sample_batches):
                     batches = [batch for j, batch in data_enum if j * data.batch_size < self.sample_batch_size]
-                    prompt = [prompt.format(self.instance_identifier)
-                              for batch in batches for prompt in batch["prompts"]][:self.sample_batch_size]
-                    nprompt = [prompt for batch in batches for prompt in batch["nprompts"]][:self.sample_batch_size]
+                    prompt = [
+                        prompt.format(self.instance_identifier)
+                        for batch in batches
+                        for prompt in batch["prompts"]
+                    ][:self.sample_batch_size]
+                    nprompt = [
+                        prompt
+                        for batch in batches
+                        for prompt in batch["nprompts"]
+                    ][:self.sample_batch_size]
 
                     samples = pipeline(
                         prompt=prompt,
@@ -443,7 +462,8 @@ class Checkpointer:
                 del all_samples
                 del image_grid
 
-        del unwrapped
+        del unwrapped_unet
+        del unwrapped_text_encoder
         del scheduler
         del pipeline
         del generator
@@ -482,8 +502,7 @@ def main():
         tokenizer = CLIPTokenizer.from_pretrained(args.pretrained_model_name_or_path, subfolder='tokenizer')
 
     # Load models and create wrapper for stable diffusion
-    text_encoder = CLIPTextModel.from_pretrained(
-        args.pretrained_model_name_or_path, subfolder='text_encoder')
+    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')
 
@@ -499,17 +518,21 @@ def main():
 
     if args.gradient_checkpointing:
         unet.enable_gradient_checkpointing()
+        text_encoder.gradient_checkpointing_enable()
 
     # slice_size = unet.config.attention_head_dim // 2
     # unet.set_attention_slice(slice_size)
 
     # Freeze text_encoder and vae
     freeze_params(vae.parameters())
-    freeze_params(text_encoder.parameters())
 
     if args.scale_lr:
-        args.learning_rate = (
-            args.learning_rate * args.gradient_accumulation_steps *
+        args.learning_rate_unet = (
+            args.learning_rate_unet * args.gradient_accumulation_steps *
+            args.train_batch_size * accelerator.num_processes
+        )
+        args.learning_rate_text = (
+            args.learning_rate_text * args.gradient_accumulation_steps *
             args.train_batch_size * accelerator.num_processes
         )
 
@@ -526,8 +549,16 @@ def main():
 
     # Initialize the optimizer
     optimizer = optimizer_class(
-        unet.parameters(),  # only optimize unet
-        lr=args.learning_rate,
+        [
+            {
+                'params': unet.parameters(),
+                'lr': args.learning_rate_unet,
+            },
+            {
+                'params': text_encoder.parameters(),
+                'lr': args.learning_rate_text,
+            }
+        ],
         betas=(args.adam_beta1, args.adam_beta2),
         weight_decay=args.adam_weight_decay,
         eps=args.adam_epsilon,
@@ -592,8 +623,10 @@ def main():
         missing_data = [item for item in datamodule.data_train if not item.class_image_path.exists()]
 
         if len(missing_data) != 0:
-            batched_data = [missing_data[i:i+args.sample_batch_size]
-                            for i in range(0, len(missing_data), args.sample_batch_size)]
+            batched_data = [
+                missing_data[i:i+args.sample_batch_size]
+                for i in range(0, len(missing_data), args.sample_batch_size)
+            ]
 
             scheduler = EulerAScheduler(
                 beta_start=0.00085, beta_end=0.012, beta_schedule="scaled_linear"
@@ -610,9 +643,9 @@ def main():
 
             with torch.inference_mode():
                 for batch in batched_data:
-                    image_name = [p.class_image_path for p in batch]
-                    prompt = [p.prompt.format(args.class_identifier) for p in batch]
-                    nprompt = [p.nprompt for p in batch]
+                    image_name = [item.class_image_path for item in batch]
+                    prompt = [item.prompt.format(args.class_identifier) for item in batch]
+                    nprompt = [item.nprompt for item in batch]
 
                     images = pipeline(
                         prompt=prompt,
@@ -655,16 +688,14 @@ def main():
             num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
         )
 
-    unet, optimizer, train_dataloader, val_dataloader, lr_scheduler = accelerator.prepare(
-        unet, optimizer, train_dataloader, val_dataloader, lr_scheduler
+    unet, text_encoder, optimizer, train_dataloader, val_dataloader, lr_scheduler = accelerator.prepare(
+        unet, text_encoder, optimizer, train_dataloader, val_dataloader, lr_scheduler
     )
 
     # Move text_encoder and vae to device
-    text_encoder.to(accelerator.device, dtype=weight_dtype)
     vae.to(accelerator.device, dtype=weight_dtype)
 
     # Keep text_encoder and vae in eval mode as we don't train these
-    text_encoder.eval()
     vae.eval()
 
     # We need to recalculate our total training steps as the size of the training dataloader may have changed.
@@ -736,12 +767,13 @@ def main():
             local_progress_bar.reset()
 
             unet.train()
+            text_encoder.train()
             train_loss = 0.0
 
             sample_checkpoint = False
 
             for step, batch in enumerate(train_dataloader):
-                with accelerator.accumulate(unet):
+                with accelerator.accumulate(itertools.chain(unet, text_encoder)):
                     # Convert images to latent space
                     latents = vae.encode(batch["pixel_values"]).latent_dist.sample()
                     latents = latents * 0.18215
@@ -782,7 +814,8 @@ def main():
 
                     accelerator.backward(loss)
                     if accelerator.sync_gradients:
-                        accelerator.clip_grad_norm_(unet.parameters(), args.max_grad_norm)
+                        accelerator.clip_grad_norm_(itertools.chain(
+                            unet.parameters(), text_encoder.parameters()), args.max_grad_norm)
                     optimizer.step()
                     if not accelerator.optimizer_step_was_skipped:
                         lr_scheduler.step()
@@ -804,7 +837,11 @@ def main():
                     if global_step % args.sample_frequency == 0:
                         sample_checkpoint = True
 
-                logs = {"train/loss": loss, "lr": lr_scheduler.get_last_lr()[0]}
+                logs = {
+                    "train/loss": loss,
+                    "lr/unet": lr_scheduler.get_last_lr()[0],
+                    "lr/text": lr_scheduler.get_last_lr()[1]
+                }
                 if args.use_ema:
                     logs["ema_decay"] = ema_unet.decay
 
@@ -820,6 +857,7 @@ def main():
             accelerator.wait_for_everyone()
 
             unet.eval()
+            text_encoder.eval()
             val_loss = 0.0
 
             for step, batch in enumerate(val_dataloader):
diff --git a/dreambooth_plus.py b/dreambooth_plus.py
index 06ff45b..413abe3 100644
--- a/dreambooth_plus.py
+++ b/dreambooth_plus.py
@@ -125,7 +125,7 @@ def parse_args():
     parser.add_argument(
         "--max_train_steps",
         type=int,
-        default=2400,
+        default=4700,
         help="Total number of training steps to perform.  If provided, overrides num_train_epochs.",
     )
     parser.add_argument(
@@ -142,13 +142,13 @@ def parse_args():
     parser.add_argument(
         "--learning_rate_unet",
         type=float,
-        default=5e-6,
+        default=2e-6,
         help="Initial learning rate (after the potential warmup period) to use.",
     )
     parser.add_argument(
         "--learning_rate_text",
         type=float,
-        default=5e-6,
+        default=2e-6,
         help="Initial learning rate (after the potential warmup period) to use.",
     )
     parser.add_argument(
@@ -578,6 +578,7 @@ def main():
 
     if args.gradient_checkpointing:
         unet.enable_gradient_checkpointing()
+        text_encoder.gradient_checkpointing_enable()
 
     # slice_size = unet.config.attention_head_dim // 2
     # unet.set_attention_slice(slice_size)
diff --git a/models/clip/prompt.py b/models/clip/prompt.py
index c1e3340..259ac44 100644
--- a/models/clip/prompt.py
+++ b/models/clip/prompt.py
@@ -1,4 +1,4 @@
-from typing import List, Optional, Union
+from typing import List, Union
 
 import torch