3 files changed, 25 insertions, 6 deletions

diff --git a/training/strategy/dreambooth.py b/training/strategy/dreambooth.py
index 4ae28b7..e6fcc89 100644
--- a/training/strategy/dreambooth.py
+++ b/training/strategy/dreambooth.py
@@ -148,7 +148,7 @@ def dreambooth_strategy_callbacks(
             torch.cuda.empty_cache()
 
     @torch.no_grad()
-    def on_sample(step):
+    def on_sample(cycle, step):
         unet_ = accelerator.unwrap_model(unet, keep_fp32_wrapper=True)
         text_encoder_ = accelerator.unwrap_model(text_encoder, keep_fp32_wrapper=True)
 
@@ -158,7 +158,7 @@ def dreambooth_strategy_callbacks(
         unet_.to(dtype=weight_dtype)
         text_encoder_.to(dtype=weight_dtype)
 
-        save_samples_(step=step, unet=unet_, text_encoder=text_encoder_)
+        save_samples_(cycle=cycle, step=step, unet=unet_, text_encoder=text_encoder_)
 
         unet_.to(dtype=orig_unet_dtype)
         text_encoder_.to(dtype=orig_text_encoder_dtype)
diff --git a/training/strategy/lora.py b/training/strategy/lora.py
index 48236fb..5c3012e 100644
--- a/training/strategy/lora.py
+++ b/training/strategy/lora.py
@@ -146,11 +146,11 @@ def lora_strategy_callbacks(
             torch.cuda.empty_cache()
 
     @torch.no_grad()
-    def on_sample(step):
+    def on_sample(cycle, step):
         unet_ = accelerator.unwrap_model(unet, keep_fp32_wrapper=True)
         text_encoder_ = accelerator.unwrap_model(text_encoder, keep_fp32_wrapper=True)
 
-        save_samples_(step=step, unet=unet_, text_encoder=text_encoder_)
+        save_samples_(cycle=cycle, step=step, unet=unet_, text_encoder=text_encoder_)
 
         del unet_, text_encoder_
 
diff --git a/training/strategy/ti.py b/training/strategy/ti.py
index f0b84b5..6bbff64 100644
--- a/training/strategy/ti.py
+++ b/training/strategy/ti.py
@@ -104,10 +104,28 @@ def textual_inversion_strategy_callbacks(
             yield
 
     @torch.no_grad()
+    def on_before_optimize(epoch: int):
+        if use_emb_decay:
+            params = [
+                p
+                for p in text_encoder.text_model.embeddings.token_embedding.parameters()
+                if p.grad is not None
+            ]
+            return torch.stack(params) if len(params) != 0 else None
+
+    @torch.no_grad()
     def on_after_optimize(w, lrs: dict[str, float]):
         if ema_embeddings is not None:
             ema_embeddings.step(text_encoder.text_model.embeddings.token_embedding.parameters())
 
+        if use_emb_decay and w is not None:
+            lr = lrs["emb"] or lrs["0"]
+            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_embeddings is not None:
             return {"ema_decay": ema_embeddings.decay}
@@ -125,7 +143,7 @@ def textual_inversion_strategy_callbacks(
                 )
 
     @torch.no_grad()
-    def on_sample(step):
+    def on_sample(cycle, step):
         unet_ = accelerator.unwrap_model(unet, keep_fp32_wrapper=True)
         text_encoder_ = accelerator.unwrap_model(text_encoder, keep_fp32_wrapper=True)
 
@@ -135,7 +153,7 @@ def textual_inversion_strategy_callbacks(
         unet_.to(dtype=weight_dtype)
         text_encoder_.to(dtype=weight_dtype)
 
-        save_samples_(step=step, unet=unet_, text_encoder=text_encoder_)
+        save_samples_(cycle=cycle, step=step, unet=unet_, text_encoder=text_encoder_)
 
         unet_.to(dtype=orig_unet_dtype)
         text_encoder_.to(dtype=orig_text_encoder_dtype)
@@ -148,6 +166,7 @@ def textual_inversion_strategy_callbacks(
     return TrainingCallbacks(
         on_train=on_train,
         on_eval=on_eval,
+        on_before_optimize=on_before_optimize,
         on_after_optimize=on_after_optimize,
         on_log=on_log,
         on_checkpoint=on_checkpoint,