Integrated updates from diffusers

5 files changed, 59 insertions, 39 deletions

diff --git a/train_dreambooth.py b/train_dreambooth.py
index 325fe90..202d52c 100644
--- a/train_dreambooth.py
+++ b/train_dreambooth.py
@@ -25,7 +25,7 @@ from pipelines.stable_diffusion.vlpn_stable_diffusion import VlpnStableDiffusion
 from data.csv import CSVDataModule
 from training.optimization import get_one_cycle_schedule
 from training.ti import patch_trainable_embeddings
-from training.util import AverageMeter, CheckpointerBase, freeze_params, save_args
+from training.util import AverageMeter, CheckpointerBase, save_args
 from models.clip.prompt import PromptProcessor
 
 logger = get_logger(__name__)
@@ -580,12 +580,10 @@ def main():
 
         patch_trainable_embeddings(text_encoder, placeholder_token_id)
 
-        freeze_params(itertools.chain(
-            text_encoder.text_model.encoder.parameters(),
-            text_encoder.text_model.final_layer_norm.parameters(),
-            text_encoder.text_model.embeddings.position_embedding.parameters(),
-            text_encoder.text_model.embeddings.token_embedding.parameters(),
-        ))
+        text_encoder.text_model.encoder.requires_grad_(False)
+        text_encoder.text_model.final_layer_norm.requires_grad_(False)
+        text_encoder.text_model.embeddings.position_embedding.requires_grad_(False)
+        text_encoder.text_model.embeddings.token_embedding.requires_grad_(False)
 
     prompt_processor = PromptProcessor(tokenizer, text_encoder)
 
@@ -905,9 +903,7 @@ def main():
             if epoch < args.train_text_encoder_epochs:
                 text_encoder.train()
             elif epoch == args.train_text_encoder_epochs:
-                freeze_params(text_encoder.parameters())
-
-            sample_checkpoint = False
+                text_encoder.requires_grad_(False)
 
             for step, batch in enumerate(train_dataloader):
                 with accelerator.accumulate(unet):
diff --git a/train_lora.py b/train_lora.py
index ffca304..9a42cae 100644
--- a/train_lora.py
+++ b/train_lora.py
@@ -25,7 +25,7 @@ from pipelines.stable_diffusion.vlpn_stable_diffusion import VlpnStableDiffusion
 from data.csv import CSVDataModule
 from training.lora import LoraAttnProcessor
 from training.optimization import get_one_cycle_schedule
-from training.util import AverageMeter, CheckpointerBase, freeze_params, save_args
+from training.util import AverageMeter, CheckpointerBase, save_args
 from models.clip.prompt import PromptProcessor
 
 logger = get_logger(__name__)
@@ -513,11 +513,9 @@ def main():
 
     print(f"Training added text embeddings")
 
-    freeze_params(itertools.chain(
-        text_encoder.text_model.encoder.parameters(),
-        text_encoder.text_model.final_layer_norm.parameters(),
-        text_encoder.text_model.embeddings.position_embedding.parameters(),
-    ))
+    text_encoder.text_model.encoder.requires_grad_(False)
+    text_encoder.text_model.final_layer_norm.requires_grad_(False)
+    text_encoder.text_model.embeddings.position_embedding.requires_grad_(False)
 
     index_fixed_tokens = torch.arange(len(tokenizer))
     index_fixed_tokens = index_fixed_tokens[~torch.isin(index_fixed_tokens, torch.tensor(placeholder_token_id))]
diff --git a/train_ti.py b/train_ti.py
index 870b2ba..d7696e5 100644
--- a/train_ti.py
+++ b/train_ti.py
@@ -25,7 +25,7 @@ from data.csv import CSVDataModule, CSVDataItem
 from training.optimization import get_one_cycle_schedule
 from training.lr import LRFinder
 from training.ti import patch_trainable_embeddings
-from training.util import AverageMeter, CheckpointerBase, save_args, freeze_params
+from training.util import AverageMeter, CheckpointerBase, save_args
 from models.clip.prompt import PromptProcessor
 
 logger = get_logger(__name__)
@@ -533,12 +533,10 @@ def main():
 
     patch_trainable_embeddings(text_encoder, placeholder_token_id)
 
-    freeze_params(itertools.chain(
-        text_encoder.text_model.encoder.parameters(),
-        text_encoder.text_model.final_layer_norm.parameters(),
-        text_encoder.text_model.embeddings.position_embedding.parameters(),
-        text_encoder.text_model.embeddings.token_embedding.parameters(),
-    ))
+    text_encoder.text_model.encoder.requires_grad_(False)
+    text_encoder.text_model.final_layer_norm.requires_grad_(False)
+    text_encoder.text_model.embeddings.position_embedding.requires_grad_(False)
+    text_encoder.text_model.embeddings.token_embedding.requires_grad_(False)
 
     prompt_processor = PromptProcessor(tokenizer, text_encoder)
 
@@ -548,6 +546,9 @@ def main():
             args.train_batch_size * accelerator.num_processes
         )
 
+    if args.find_lr:
+        args.learning_rate = 1e2
+
     # Use 8-bit Adam for lower memory usage or to fine-tune the model in 16GB GPUs
     if args.use_8bit_adam:
         try:
@@ -715,7 +716,11 @@ def main():
 
     # Keep vae and unet in eval mode as we don't train these
     vae.eval()
-    unet.eval()
+
+    if args.gradient_checkpointing:
+        unet.train()
+    else:
+        unet.eval()
 
     # 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)
diff --git a/training/lr.py b/training/lr.py
index 8e558e1..c1fa3a0 100644
--- a/training/lr.py
+++ b/training/lr.py
@@ -22,10 +22,13 @@ class LRFinder():
         self.model_state = copy.deepcopy(model.state_dict())
         self.optimizer_state = copy.deepcopy(optimizer.state_dict())
 
-    def run(self, min_lr, num_epochs=100, num_train_batches=1, num_val_batches=math.inf, smooth_f=0.05, diverge_th=5):
+    def run(self, min_lr, skip_start=10, skip_end=5, num_epochs=100, num_train_batches=1, num_val_batches=math.inf, smooth_f=0.05, diverge_th=5):
         best_loss = None
+        best_acc = None
+
         lrs = []
         losses = []
+        accs = []
 
         lr_scheduler = get_exponential_schedule(self.optimizer, min_lr, num_epochs)
 
@@ -44,6 +47,7 @@ class LRFinder():
             progress_bar.set_description(f"Epoch {epoch + 1} / {num_epochs}")
 
             avg_loss = AverageMeter()
+            avg_acc = AverageMeter()
 
             self.model.train()
 
@@ -71,28 +75,37 @@ class LRFinder():
 
                     loss, acc, bsz = self.loss_fn(batch)
                     avg_loss.update(loss.detach_(), bsz)
+                    avg_acc.update(acc.detach_(), bsz)
 
                     progress_bar.update(1)
 
             lr_scheduler.step()
 
             loss = avg_loss.avg.item()
+            acc = avg_acc.avg.item()
+
             if epoch == 0:
                 best_loss = loss
+                best_acc = acc
             else:
                 if smooth_f > 0:
                     loss = smooth_f * loss + (1 - smooth_f) * losses[-1]
                 if loss < best_loss:
                     best_loss = loss
+                if acc > best_acc:
+                    best_acc = acc
 
             lr = lr_scheduler.get_last_lr()[0]
 
             lrs.append(lr)
             losses.append(loss)
+            accs.append(acc)
 
             progress_bar.set_postfix({
                 "loss": loss,
-                "best": best_loss,
+                "loss/best": best_loss,
+                "acc": acc,
+                "acc/best": best_acc,
                 "lr": lr,
             })
 
@@ -103,20 +116,37 @@ class LRFinder():
                 print("Stopping early, the loss has diverged")
                 break
 
-        fig, ax = plt.subplots()
-        ax.plot(lrs, losses)
+        if skip_end == 0:
+            lrs = lrs[skip_start:]
+            losses = losses[skip_start:]
+            accs = accs[skip_start:]
+        else:
+            lrs = lrs[skip_start:-skip_end]
+            losses = losses[skip_start:-skip_end]
+            accs = accs[skip_start:-skip_end]
+
+        fig, ax_loss = plt.subplots()
+
+        ax_loss.plot(lrs, losses, color='red', label='Loss')
+        ax_loss.set_xscale("log")
+        ax_loss.set_xlabel("Learning rate")
+
+        # ax_acc = ax_loss.twinx()
+        # ax_acc.plot(lrs, accs, color='blue', label='Accuracy')
 
         print("LR suggestion: steepest gradient")
         min_grad_idx = None
+
         try:
             min_grad_idx = (np.gradient(np.array(losses))).argmin()
         except ValueError:
             print(
                 "Failed to compute the gradients, there might not be enough points."
             )
+
         if min_grad_idx is not None:
             print("Suggested LR: {:.2E}".format(lrs[min_grad_idx]))
-            ax.scatter(
+            ax_loss.scatter(
                 lrs[min_grad_idx],
                 losses[min_grad_idx],
                 s=75,
@@ -125,11 +155,7 @@ class LRFinder():
                 zorder=3,
                 label="steepest gradient",
             )
-            ax.legend()
-
-        ax.set_xscale("log")
-        ax.set_xlabel("Learning rate")
-        ax.set_ylabel("Loss")
+            ax_loss.legend()
 
 
 def get_exponential_schedule(optimizer, min_lr, num_epochs, last_epoch=-1):
diff --git a/training/util.py b/training/util.py
index a0c15cd..d0f7fcd 100644
--- a/training/util.py
+++ b/training/util.py
@@ -5,11 +5,6 @@ import torch
 from PIL import Image
 
 
-def freeze_params(params):
-    for param in params:
-        param.requires_grad = False
-
-
 def save_args(basepath: Path, args, extra={}):
     info = {"args": vars(args)}
     info["args"].update(extra)