lora merger

scripts/train.py

@@ -0,0 +1,207 @@
+import torch
+import pandas as pd
+from PIL import Image
+import lightning as pl
+from diffsynth import ModelManager, FluxImagePipeline, download_models, load_state_dict
+from diffsynth.models.lora import LoRAFromCivitai, FluxLoRAConverter
+from diffsynth.data.video import crop_and_resize
+from diffsynth.pipelines.flux_image import lets_dance_flux
+from torchvision.transforms import v2
+
+
+
+class LoraMerger(torch.nn.Module):
+    def __init__(self, dim):
+        super().__init__()
+        self.weight_base = torch.nn.Parameter(torch.randn((dim,)))
+        self.weight_lora = torch.nn.Parameter(torch.randn((dim,)))
+        self.weight_cross = torch.nn.Parameter(torch.randn((dim,)))
+        self.weight_out = torch.nn.Parameter(torch.ones((dim,)))
+        self.bias = torch.nn.Parameter(torch.randn((dim,)))
+        self.activation = torch.nn.Sigmoid()
+        self.norm_base = torch.nn.LayerNorm(dim, eps=1e-5)
+        self.norm_lora = torch.nn.LayerNorm(dim, eps=1e-5)
+        
+    def forward(self, base_output, lora_outputs):
+        norm_base_output = self.norm_base(base_output)
+        norm_lora_outputs = self.norm_lora(lora_outputs)
+        gate = self.activation(
+            norm_base_output * self.weight_base \
+            + norm_lora_outputs * self.weight_lora \
+            + norm_base_output * norm_lora_outputs * self.weight_cross + self.bias
+        )
+        output = base_output + (self.weight_out * gate * lora_outputs).sum(dim=0)
+        return output
+
+
+
+class LoraPatcher(torch.nn.Module):
+    def __init__(self, lora_patterns=None):
+        super().__init__()
+        if lora_patterns is None:
+            lora_patterns = self.default_lora_patterns()
+        model_dict = {}
+        for lora_pattern in lora_patterns:
+            name, dim = lora_pattern["name"], lora_pattern["dim"]
+            model_dict[name.replace(".", "___")] = LoraMerger(dim)
+        self.model_dict = torch.nn.ModuleDict(model_dict)
+        
+    def default_lora_patterns(self):
+        lora_patterns = []
+        lora_dict = {
+            "attn.a_to_qkv": 9216, "attn.a_to_out": 3072, "ff_a.0": 12288, "ff_a.2": 3072, "norm1_a.linear": 18432,
+            "attn.b_to_qkv": 9216, "attn.b_to_out": 3072, "ff_b.0": 12288, "ff_b.2": 3072, "norm1_b.linear": 18432,
+        }
+        for i in range(19):
+            for suffix in lora_dict:
+                lora_patterns.append({
+                    "name": f"blocks.{i}.{suffix}",
+                    "dim": lora_dict[suffix]
+                })
+        lora_dict = {"to_qkv_mlp": 21504, "proj_out": 3072, "norm.linear": 9216}
+        for i in range(38):
+            for suffix in lora_dict:
+                lora_patterns.append({
+                    "name": f"single_blocks.{i}.{suffix}",
+                    "dim": lora_dict[suffix]
+                })
+        return lora_patterns
+        
+    def forward(self, base_output, lora_outputs, name):
+        return self.model_dict[name.replace(".", "___")](base_output, lora_outputs)
+
+
+
+class LoraDataset(torch.utils.data.Dataset):
+    def __init__(self, metadata_path, steps_per_epoch=1000):
+        data_df = pd.read_csv(metadata_path)
+        self.model_file = data_df["model_file"].tolist()
+        self.image_file = data_df["image_file"].tolist()
+        self.text = data_df["text"].tolist()
+        self.max_resolution = 1920 * 1080
+        self.steps_per_epoch = steps_per_epoch
+        
+        
+    def read_image(self, image_file):
+        image = Image.open(image_file)
+        width, height = image.size
+        if width * height > self.max_resolution:
+            scale = (width * height / self.max_resolution) ** 0.5
+            image = image.resize((int(width / scale), int(height / scale)))
+            width, height = image.size
+        if width % 16 != 0 or height % 16 != 0:
+            image = crop_and_resize(image, height // 16 * 16, width // 16 * 16)
+        image = v2.functional.to_image(image)
+        image = v2.functional.to_dtype(image, dtype=torch.float32, scale=True)
+        image = v2.functional.normalize(image, [0.5], [0.5])
+        return image
+
+
+    def __getitem__(self, index):
+        data_id = torch.randint(0, len(self.model_file), (1,))[0]
+        data_id = (data_id + index) % len(self.model_file) # For fixed seed.
+        data_id_extra = torch.randint(0, len(self.model_file), (1,))[0]
+        return {
+            "model_file": self.model_file[data_id],
+            "model_file_extra": self.model_file[data_id_extra],
+            "image": self.read_image(self.image_file[data_id]),
+            "text": self.text[data_id]
+        }
+
+
+    def __len__(self):
+        return self.steps_per_epoch
+
+
+
+class LightningModel(pl.LightningModule):
+    def __init__(
+        self,
+        learning_rate=1e-4,
+        use_gradient_checkpointing=True,
+        state_dict_converter=FluxLoRAConverter.align_to_diffsynth_format,
+    ):
+        super().__init__()
+        model_manager = ModelManager(torch_dtype=torch.bfloat16, device=self.device, model_id_list=["FLUX.1-dev"])
+        self.pipe = FluxImagePipeline.from_model_manager(model_manager)
+        self.lora_patcher = LoraPatcher()
+        self.pipe.enable_auto_lora()
+        self.pipe.scheduler.set_timesteps(1000, training=True)
+        self.freeze_parameters()
+        # Set parameters
+        self.learning_rate = learning_rate
+        self.use_gradient_checkpointing = use_gradient_checkpointing
+        self.state_dict_converter = state_dict_converter
+
+
+    def freeze_parameters(self):
+        # Freeze parameters
+        self.pipe.requires_grad_(False)
+        self.pipe.eval()
+        self.pipe.denoising_model().train()
+
+
+    def training_step(self, batch, batch_idx):
+        # Data
+        text, image = batch["text"], batch["image"]
+        lora_state_dicts = [
+            self.state_dict_converter(load_state_dict(batch["model_file"][0], torch_dtype=torch.bfloat16, device=self.device)),
+            self.state_dict_converter(load_state_dict(batch["model_file_extra"][0], torch_dtype=torch.bfloat16, device=self.device)),
+        ]
+        lora_alpahs = [1, 1]
+
+        # Prepare input parameters
+        self.pipe.device = self.device
+        prompt_emb = self.pipe.encode_prompt(text, positive=True)
+        if "latents" in batch:
+            latents = batch["latents"].to(dtype=self.pipe.torch_dtype, device=self.device)
+        else:
+            latents = self.pipe.vae_encoder(image.to(dtype=self.pipe.torch_dtype, device=self.device))
+        noise = torch.randn_like(latents)
+        timestep_id = torch.randint(0, self.pipe.scheduler.num_train_timesteps, (1,))
+        timestep = self.pipe.scheduler.timesteps[timestep_id].to(self.device)
+        extra_input = self.pipe.prepare_extra_input(latents)
+        noisy_latents = self.pipe.scheduler.add_noise(latents, noise, timestep)
+        training_target = self.pipe.scheduler.training_target(latents, noise, timestep)
+
+        # Compute loss
+        noise_pred = lets_dance_flux(
+            self.pipe.dit,
+            hidden_states=noisy_latents, timestep=timestep, **prompt_emb, **extra_input,
+            lora_state_dicts=lora_state_dicts, lora_alpahs=lora_alpahs, lora_patcher=self.lora_patcher,
+            use_gradient_checkpointing=self.use_gradient_checkpointing
+        )
+        loss = torch.nn.functional.mse_loss(noise_pred.float(), training_target.float())
+        loss = loss * self.pipe.scheduler.training_weight(timestep)
+
+        # Record log
+        self.log("train_loss", loss, prog_bar=True)
+        return loss
+
+
+    def configure_optimizers(self):
+        trainable_modules = filter(lambda p: p.requires_grad, self.lora_patcher.parameters())
+        optimizer = torch.optim.AdamW(trainable_modules, lr=self.learning_rate)
+        return optimizer
+    
+    
+    def on_save_checkpoint(self, checkpoint):
+        checkpoint.clear()
+        checkpoint.update(self.lora_patcher.state_dict())
+
+
+if __name__ == '__main__':
+    model = LightningModel(learning_rate=1e-4)
+    dataset = LoraDataset("data/loras.csv", steps_per_epoch=500)
+    train_loader = torch.utils.data.DataLoader(dataset, shuffle=True, batch_size=1, num_workers=1)
+    trainer = pl.Trainer(
+        max_epochs=100000,
+        accelerator="gpu",
+        devices="auto",
+        precision="bf16",
+        strategy="auto",
+        default_root_dir="./models",
+        accumulate_grad_batches=1,
+        callbacks=[pl.pytorch.callbacks.ModelCheckpoint(save_top_k=-1)],
+    )
+    trainer.fit(model=model, train_dataloaders=train_loader)