[LoRA training] update metadata use for lora alpha + README (#11723)

* lora alpha

* Apply style fixes

* Update examples/advanced_diffusion_training/README_flux.md
Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>

* fix readme format

---------
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>

examples/advanced_diffusion_training/README_flux.md

@@ -76,6 +76,24 @@ This command will prompt you for a token. Copy-paste yours from your [settings/t
 > `pip install wandb`
 > Alternatively, you can use other tools / train without reporting by modifying the flag  `--report_to="wandb"`.
 
+### LoRA Rank and Alpha
+Two key LoRA hyperparameters are LoRA rank and LoRA alpha. 
+- `--rank`: Defines the dimension of the trainable LoRA matrices. A higher rank means more expressiveness and capacity to learn (and more parameters).
+- `--lora_alpha`: A scaling factor for the LoRA's output. The LoRA update is scaled by lora_alpha / lora_rank.
+- lora_alpha vs. rank:
+This ratio dictates the LoRA's effective strength:
+lora_alpha == rank: Scaling factor is 1. The LoRA is applied with its learned strength. (e.g., alpha=16, rank=16)
+lora_alpha < rank: Scaling factor < 1. Reduces the LoRA's impact. Useful for subtle changes or to prevent overpowering the base model. (e.g., alpha=8, rank=16)
+lora_alpha > rank: Scaling factor > 1. Amplifies the LoRA's impact. Allows a lower rank LoRA to have a stronger effect. (e.g., alpha=32, rank=16)
+
+> [!TIP]
+> A common starting point is to set `lora_alpha` equal to `rank`. 
+> Some also set `lora_alpha` to be twice the `rank` (e.g., lora_alpha=32 for lora_rank=16) 
+> to give the LoRA updates more influence without increasing parameter count. 
+> If you find your LoRA is "overcooking" or learning too aggressively, consider setting `lora_alpha` to half of `rank` 
+> (e.g., lora_alpha=8 for rank=16). Experimentation is often key to finding the optimal balance for your use case.
+
+
 ### Target Modules
 When LoRA was first adapted from language models to diffusion models, it was applied to the cross-attention layers in the Unet that relate the image representations with the prompts that describe them. 
 More recently, SOTA text-to-image diffusion models replaced the Unet with a diffusion Transformer(DiT). With this change, we may also want to explore 

examples/advanced_diffusion_training/test_dreambooth_lora_flux_advanced.py

@@ -13,6 +13,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import json
 import logging
 import os
 import sys
@@ -20,6 +21,8 @@ import tempfile
 
 import safetensors
 
+from diffusers.loaders.lora_base import LORA_ADAPTER_METADATA_KEY
+
 
 sys.path.append("..")
 from test_examples_utils import ExamplesTestsAccelerate, run_command  # noqa: E402
@@ -281,3 +284,45 @@ class DreamBoothLoRAFluxAdvanced(ExamplesTestsAccelerate):
             run_command(self._launch_args + resume_run_args)
 
             self.assertEqual({x for x in os.listdir(tmpdir) if "checkpoint" in x}, {"checkpoint-6", "checkpoint-8"})
+
+    def test_dreambooth_lora_with_metadata(self):
+        # Use a `lora_alpha` that is different from `rank`.
+        lora_alpha = 8
+        rank = 4
+        with tempfile.TemporaryDirectory() as tmpdir:
+            test_args = f"""
+                {self.script_path}
+                --pretrained_model_name_or_path {self.pretrained_model_name_or_path}
+                --instance_data_dir {self.instance_data_dir}
+                --instance_prompt {self.instance_prompt}
+                --resolution 64
+                --train_batch_size 1
+                --gradient_accumulation_steps 1
+                --max_train_steps 2
+                --lora_alpha={lora_alpha}
+                --rank={rank}
+                --learning_rate 5.0e-04
+                --scale_lr
+                --lr_scheduler constant
+                --lr_warmup_steps 0
+                --output_dir {tmpdir}
+                """.split()
+
+            run_command(self._launch_args + test_args)
+            # save_pretrained smoke test
+            state_dict_file = os.path.join(tmpdir, "pytorch_lora_weights.safetensors")
+            self.assertTrue(os.path.isfile(state_dict_file))
+
+            # Check if the metadata was properly serialized.
+            with safetensors.torch.safe_open(state_dict_file, framework="pt", device="cpu") as f:
+                metadata = f.metadata() or {}
+
+            metadata.pop("format", None)
+            raw = metadata.get(LORA_ADAPTER_METADATA_KEY)
+            if raw:
+                raw = json.loads(raw)
+
+            loaded_lora_alpha = raw["transformer.lora_alpha"]
+            self.assertTrue(loaded_lora_alpha == lora_alpha)
+            loaded_lora_rank = raw["transformer.r"]
+            self.assertTrue(loaded_lora_rank == rank)

examples/advanced_diffusion_training/train_dreambooth_lora_flux_advanced.py

@@ -55,6 +55,7 @@ from diffusers import (
 )
 from diffusers.optimization import get_scheduler
 from diffusers.training_utils import (
+    _collate_lora_metadata,
     _set_state_dict_into_text_encoder,
     cast_training_params,
     compute_density_for_timestep_sampling,
@@ -431,6 +432,13 @@ def parse_args(input_args=None):
         help=("The dimension of the LoRA update matrices."),
     )
 
+    parser.add_argument(
+        "--lora_alpha",
+        type=int,
+        default=4,
+        help="LoRA alpha to be used for additional scaling.",
+    )
+
     parser.add_argument("--lora_dropout", type=float, default=0.0, help="Dropout probability for LoRA layers")
 
     parser.add_argument(
@@ -1556,7 +1564,7 @@ def main(args):
     # now we will add new LoRA weights to the attention layers
     transformer_lora_config = LoraConfig(
         r=args.rank,
-        lora_alpha=args.rank,
+        lora_alpha=args.lora_alpha,
         lora_dropout=args.lora_dropout,
         init_lora_weights="gaussian",
         target_modules=target_modules,
@@ -1565,7 +1573,7 @@ def main(args):
     if args.train_text_encoder:
         text_lora_config = LoraConfig(
             r=args.rank,
-            lora_alpha=args.rank,
+            lora_alpha=args.lora_alpha,
             lora_dropout=args.lora_dropout,
             init_lora_weights="gaussian",
             target_modules=["q_proj", "k_proj", "v_proj", "out_proj"],
@@ -1582,13 +1590,15 @@ def main(args):
         if accelerator.is_main_process:
             transformer_lora_layers_to_save = None
             text_encoder_one_lora_layers_to_save = None
-
+            modules_to_save = {}
             for model in models:
                 if isinstance(model, type(unwrap_model(transformer))):
                     transformer_lora_layers_to_save = get_peft_model_state_dict(model)
+                    modules_to_save["transformer"] = model
                 elif isinstance(model, type(unwrap_model(text_encoder_one))):
                     if args.train_text_encoder:  # when --train_text_encoder_ti we don't save the layers
                         text_encoder_one_lora_layers_to_save = get_peft_model_state_dict(model)
+                        modules_to_save["text_encoder"] = model
                 elif isinstance(model, type(unwrap_model(text_encoder_two))):
                     pass  # when --train_text_encoder_ti and --enable_t5_ti we don't save the layers
                 else:
@@ -1601,6 +1611,7 @@ def main(args):
                 output_dir,
                 transformer_lora_layers=transformer_lora_layers_to_save,
                 text_encoder_lora_layers=text_encoder_one_lora_layers_to_save,
+                **_collate_lora_metadata(modules_to_save),
             )
         if args.train_text_encoder_ti:
             embedding_handler.save_embeddings(f"{args.output_dir}/{Path(args.output_dir).name}_emb.safetensors")
@@ -2359,16 +2370,19 @@ def main(args):
     # Save the lora layers
     accelerator.wait_for_everyone()
     if accelerator.is_main_process:
+        modules_to_save = {}
         transformer = unwrap_model(transformer)
         if args.upcast_before_saving:
             transformer.to(torch.float32)
         else:
             transformer = transformer.to(weight_dtype)
         transformer_lora_layers = get_peft_model_state_dict(transformer)
+        modules_to_save["transformer"] = transformer
 
         if args.train_text_encoder:
             text_encoder_one = unwrap_model(text_encoder_one)
             text_encoder_lora_layers = get_peft_model_state_dict(text_encoder_one.to(torch.float32))
+            modules_to_save["text_encoder"] = text_encoder_one
         else:
             text_encoder_lora_layers = None
 
@@ -2377,6 +2391,7 @@ def main(args):
                 save_directory=args.output_dir,
                 transformer_lora_layers=transformer_lora_layers,
                 text_encoder_lora_layers=text_encoder_lora_layers,
+                **_collate_lora_metadata(modules_to_save),
             )
 
         if args.train_text_encoder_ti:

examples/dreambooth/README_flux.md

@@ -170,6 +170,23 @@ accelerate launch train_dreambooth_lora_flux.py \
   --push_to_hub
 ```
 
+### LoRA Rank and Alpha
+Two key LoRA hyperparameters are LoRA rank and LoRA alpha. 
+- `--rank`: Defines the dimension of the trainable LoRA matrices. A higher rank means more expressiveness and capacity to learn (and more parameters).
+- `--lora_alpha`: A scaling factor for the LoRA's output. The LoRA update is scaled by lora_alpha / lora_rank.
+- lora_alpha vs. rank:
+This ratio dictates the LoRA's effective strength:
+lora_alpha == rank: Scaling factor is 1. The LoRA is applied with its learned strength. (e.g., alpha=16, rank=16)
+lora_alpha < rank: Scaling factor < 1. Reduces the LoRA's impact. Useful for subtle changes or to prevent overpowering the base model. (e.g., alpha=8, rank=16)
+lora_alpha > rank: Scaling factor > 1. Amplifies the LoRA's impact. Allows a lower rank LoRA to have a stronger effect. (e.g., alpha=32, rank=16)
+
+> [!TIP]
+> A common starting point is to set `lora_alpha` equal to `rank`. 
+> Some also set `lora_alpha` to be twice the `rank` (e.g., lora_alpha=32 for lora_rank=16) 
+> to give the LoRA updates more influence without increasing parameter count. 
+> If you find your LoRA is "overcooking" or learning too aggressively, consider setting `lora_alpha` to half of `rank` 
+> (e.g., lora_alpha=8 for rank=16). Experimentation is often key to finding the optimal balance for your use case.
+
 ### Target Modules
 When LoRA was first adapted from language models to diffusion models, it was applied to the cross-attention layers in the Unet that relate the image representations with the prompts that describe them. 
 More recently, SOTA text-to-image diffusion models replaced the Unet with a diffusion Transformer(DiT). With this change, we may also want to explore