Added Prodigy optimizer

3 files changed, 26 insertions, 1 deletions

diff --git a/environment.yaml b/environment.yaml
index 2c81a90..4a73688 100644
--- a/environment.yaml
+++ b/environment.yaml
@@ -27,6 +27,7 @@ dependencies:
           - bitsandbytes==0.39.1
           - lion-pytorch==0.0.7
           - peft==0.3.0
+          - prodigyopt==1.0
           - python-slugify>=6.1.2
           - safetensors==0.3.1
           - setuptools==65.6.3
diff --git a/train_dreambooth.py b/train_dreambooth.py
index 84197c8..beb65fc 100644
--- a/train_dreambooth.py
+++ b/train_dreambooth.py
@@ -348,6 +348,7 @@ def parse_args():
             "dadan",
             "dlion",
             "adafactor",
+            "prodigy",
         ],
         help="Optimizer to use",
     )
@@ -828,6 +829,21 @@ def main():
         args.learning_rate_text = 1.0
     elif args.optimizer == "dlion":
         raise ImportError("DLion has not been merged into dadaptation yet")
+    elif args.optimizer == "prodigy":
+        try:
+            import prodigyopt
+        except ImportError:
+            raise ImportError(
+                "To use DAdaptAdan, please install the dadaptation library: `pip install dadaptation`."
+            )
+
+        create_optimizer = partial(
+            prodigyopt.Prodigy,
+            weight_decay=args.adam_weight_decay,
+        )
+
+        args.learning_rate_unet = 1.0
+        args.learning_rate_text = 1.0
     else:
         raise ValueError(f'Unknown --optimizer "{args.optimizer}"')
 
diff --git a/training/functional.py b/training/functional.py
index 34a701b..cc079ef 100644
--- a/training/functional.py
+++ b/training/functional.py
@@ -525,6 +525,7 @@ def train_loop(
     on_checkpoint = callbacks.on_checkpoint
 
     isDadaptation = False
+    isProdigy = False
 
     try:
         import dadaptation
@@ -535,6 +536,13 @@ def train_loop(
     except ImportError:
         pass
 
+    try:
+        import prodigyopt
+
+        isProdigy = isinstance(optimizer.optimizer, prodigyopt.Prodigy)
+    except ImportError:
+        pass
+
     num_training_steps += global_step_offset
     global_step += global_step_offset
 
@@ -582,7 +590,7 @@ def train_loop(
                             lr = lr.item()
                         label = group_labels[i] if i < len(group_labels) else f"{i}"
                         logs[f"lr/{label}"] = lr
-                        if isDadaptation:
+                        if isDadaptation or isProdigy:
                             lr = (
                                 optimizer.param_groups[i]["d"]
                                 * optimizer.param_groups[i]["lr"]