Update

5 files changed, 37 insertions, 27 deletions

diff --git a/pipelines/stable_diffusion/vlpn_stable_diffusion.py b/pipelines/stable_diffusion/vlpn_stable_diffusion.py
index f426de1..4505a2a 100644
--- a/pipelines/stable_diffusion/vlpn_stable_diffusion.py
+++ b/pipelines/stable_diffusion/vlpn_stable_diffusion.py
@@ -394,7 +394,7 @@ class VlpnStableDiffusion(DiffusionPipeline):
         sag_scale: float = 0.75,
         eta: float = 0.0,
         generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
-        image: Optional[Union[torch.FloatTensor, PIL.Image.Image, Literal["noise"]]] = "noise",
+        image: Optional[Union[torch.FloatTensor, PIL.Image.Image, Literal["noise"]]] = None,
         output_type: str = "pil",
         return_dict: bool = True,
         callback: Optional[Callable[[int, int, torch.FloatTensor], None]] = None,
diff --git a/train_lora.py b/train_lora.py
index 6e72376..e65e7be 100644
--- a/train_lora.py
+++ b/train_lora.py
@@ -13,7 +13,7 @@ from accelerate.logging import get_logger
 from accelerate.utils import LoggerType, set_seed
 from slugify import slugify
 from diffusers.loaders import AttnProcsLayers
-from diffusers.models.cross_attention import LoRAXFormersCrossAttnProcessor, LoRACrossAttnProcessor
+from diffusers.models.cross_attention import LoRACrossAttnProcessor
 
 from util.files import load_config, load_embeddings_from_dir
 from data.csv import VlpnDataModule, keyword_filter
@@ -292,6 +292,12 @@ def parse_args():
         ),
     )
     parser.add_argument(
+        "--lora_rank",
+        type=int,
+        default=256,
+        help="LoRA rank.",
+    )
+    parser.add_argument(
         "--sample_frequency",
         type=int,
         default=1,
@@ -420,10 +426,6 @@ def main():
     tokenizer, text_encoder, vae, unet, noise_scheduler, sample_scheduler, embeddings = get_models(
         args.pretrained_model_name_or_path)
 
-    vae.enable_slicing()
-    vae.set_use_memory_efficient_attention_xformers(True)
-    unet.enable_xformers_memory_efficient_attention()
-
     unet.to(accelerator.device, dtype=weight_dtype)
     text_encoder.to(accelerator.device, dtype=weight_dtype)
 
@@ -439,11 +441,18 @@ def main():
             block_id = int(name[len("down_blocks.")])
             hidden_size = unet.config.block_out_channels[block_id]
 
-        lora_attn_procs[name] = LoRAXFormersCrossAttnProcessor(
-            hidden_size=hidden_size, cross_attention_dim=cross_attention_dim
+        lora_attn_procs[name] = LoRACrossAttnProcessor(
+            hidden_size=hidden_size,
+            cross_attention_dim=cross_attention_dim,
+            rank=args.lora_rank
         )
 
     unet.set_attn_processor(lora_attn_procs)
+
+    vae.enable_slicing()
+    vae.set_use_memory_efficient_attention_xformers(True)
+    unet.enable_xformers_memory_efficient_attention()
+
     lora_layers = AttnProcsLayers(unet.attn_processors)
 
     if args.embeddings_dir is not None:
diff --git a/train_ti.py b/train_ti.py
index b9d6e56..81938c8 100644
--- a/train_ti.py
+++ b/train_ti.py
@@ -476,13 +476,10 @@ def parse_args():
     if len(args.placeholder_tokens) != len(args.initializer_tokens):
         raise ValueError("--placeholder_tokens and --initializer_tokens must have the same number of items")
 
-    if args.num_vectors is None:
-        args.num_vectors = 1
-
     if isinstance(args.num_vectors, int):
         args.num_vectors = [args.num_vectors] * len(args.placeholder_tokens)
 
-    if len(args.placeholder_tokens) != len(args.num_vectors):
+    if isinstance(args.num_vectors, list) and len(args.placeholder_tokens) != len(args.num_vectors):
         raise ValueError("--placeholder_tokens and --num_vectors must have the same number of items")
 
     if args.sequential:
@@ -491,6 +488,9 @@ def parse_args():
 
         if len(args.placeholder_tokens) != len(args.train_data_template):
             raise ValueError("--placeholder_tokens and --train_data_template must have the same number of items")
+
+        if args.num_vectors is None:
+            args.num_vectors = [None] * len(args.placeholder_tokens)
     else:
         if isinstance(args.train_data_template, list):
             raise ValueError("--train_data_template can't be a list in simultaneous mode")
diff --git a/training/functional.py b/training/functional.py
index 27a43c2..4565612 100644
--- a/training/functional.py
+++ b/training/functional.py
@@ -231,12 +231,16 @@ def add_placeholder_tokens(
     embeddings: ManagedCLIPTextEmbeddings,
     placeholder_tokens: list[str],
     initializer_tokens: list[str],
-    num_vectors: Union[list[int], int]
+    num_vectors: Optional[Union[list[int], int]] = None,
 ):
     initializer_token_ids = [
         tokenizer.encode(token, add_special_tokens=False)
         for token in initializer_tokens
     ]
+
+    if num_vectors is None:
+        num_vectors = [len(ids) for ids in initializer_token_ids]
+
     placeholder_token_ids = tokenizer.add_multi_tokens(placeholder_tokens, num_vectors)
 
     embeddings.resize(len(tokenizer))
diff --git a/training/strategy/lora.py b/training/strategy/lora.py
index ccec215..cab5e4c 100644
--- a/training/strategy/lora.py
+++ b/training/strategy/lora.py
@@ -11,10 +11,7 @@ from transformers import CLIPTextModel
 from diffusers import AutoencoderKL, UNet2DConditionModel, DPMSolverMultistepScheduler
 from diffusers.loaders import AttnProcsLayers
 
-from slugify import slugify
-
 from models.clip.tokenizer import MultiCLIPTokenizer
-from training.util import EMAModel
 from training.functional import TrainingStrategy, TrainingCallbacks, save_samples
 
 
@@ -41,16 +38,9 @@ def lora_strategy_callbacks(
     sample_output_dir.mkdir(parents=True, exist_ok=True)
     checkpoint_output_dir.mkdir(parents=True, exist_ok=True)
 
-    weight_dtype = torch.float32
-    if accelerator.state.mixed_precision == "fp16":
-        weight_dtype = torch.float16
-    elif accelerator.state.mixed_precision == "bf16":
-        weight_dtype = torch.bfloat16
-
     save_samples_ = partial(
         save_samples,
         accelerator=accelerator,
-        unet=unet,
         text_encoder=text_encoder,
         tokenizer=tokenizer,
         vae=vae,
@@ -83,20 +73,27 @@ def lora_strategy_callbacks(
         yield
 
     def on_before_optimize(lr: float, epoch: int):
-        if accelerator.sync_gradients:
-            accelerator.clip_grad_norm_(lora_layers.parameters(), max_grad_norm)
+        accelerator.clip_grad_norm_(lora_layers.parameters(), max_grad_norm)
 
     @torch.no_grad()
     def on_checkpoint(step, postfix):
         print(f"Saving checkpoint for step {step}...")
 
         unet_ = accelerator.unwrap_model(unet, False)
-        unet_.save_attn_procs(checkpoint_output_dir / f"{step}_{postfix}")
+        unet_.save_attn_procs(
+            checkpoint_output_dir / f"{step}_{postfix}",
+            safe_serialization=True
+        )
         del unet_
 
     @torch.no_grad()
     def on_sample(step):
-        save_samples_(step=step)
+        unet_ = accelerator.unwrap_model(unet, False)
+        save_samples_(step=step, unet=unet_)
+        del unet_
+
+        if torch.cuda.is_available():
+            torch.cuda.empty_cache()
 
     return TrainingCallbacks(
         on_prepare=on_prepare,