Update

1 files changed, 25 insertions, 1 deletions

diff --git a/training/strategy/dreambooth.py b/training/strategy/dreambooth.py
index dc19ba3..0f64747 100644
--- a/training/strategy/dreambooth.py
+++ b/training/strategy/dreambooth.py
@@ -30,8 +30,13 @@ def dreambooth_strategy_callbacks(
     sample_output_dir: Path,
     checkpoint_output_dir: Path,
     seed: int,
+    placeholder_tokens: list[str],
+    placeholder_token_ids: list[list[int]],
     train_text_encoder_cycles: int,
     text_encoder_unfreeze_last_n_layers: int = 2,
+    use_emb_decay: bool = False,
+    emb_decay_target: float = 0.4,
+    emb_decay: float = 1e-2,
     max_grad_norm: float = 1.0,
     use_ema: bool = False,
     ema_inv_gamma: float = 1.0,
@@ -112,11 +117,29 @@ def dreambooth_strategy_callbacks(
             params_to_clip.append(text_encoder.parameters())
         accelerator.clip_grad_norm_(itertools.chain(*params_to_clip), max_grad_norm)
 
+        if len(placeholder_tokens) != 0 and use_emb_decay:
+            params = [
+                p
+                for p in text_encoder.text_model.embeddings.parameters()
+                if p.grad is not None
+            ]
+            return torch.stack(params) if len(params) != 0 else None
+
     @torch.no_grad()
-    def on_after_optimize(_, lrs: dict[str, float]):
+    def on_after_optimize(w, lrs: dict[str, float]):
         if ema_unet is not None:
             ema_unet.step(unet.parameters())
 
+        if w is not None and "emb" in lrs:
+            lr = lrs["emb"]
+            lambda_ = emb_decay * lr
+
+            if lambda_ != 0:
+                norm = w[:, :].norm(dim=-1, keepdim=True)
+                w[:].add_(
+                    (w[:] / norm.clamp_min(1e-12)) * lambda_ * (emb_decay_target - norm)
+                )
+
     def on_log():
         if ema_unet is not None:
             return {"ema_decay": ema_unet.decay}
@@ -212,6 +235,7 @@ def dreambooth_prepare(
         ]:
             layer.requires_grad_(False)
 
+    text_encoder.text_model.embeddings.position_embedding.requires_grad_(False)
     # text_encoder.text_model.embeddings.requires_grad_(False)
 
     return text_encoder, unet, optimizer, train_dataloader, val_dataloader, lr_scheduler