2 files changed, 282 insertions, 0 deletions

diff --git a/training/functional.py b/training/functional.py
index 1f2ca6d..e54c9c8 100644
--- a/training/functional.py
+++ b/training/functional.py
@@ -2,6 +2,8 @@ import math
 from contextlib import _GeneratorContextManager, nullcontext
 from typing import Callable, Any, Tuple, Union, Optional
 from functools import partial
+from pathlib import Path
+import itertools
 
 import torch
 import torch.nn.functional as F
@@ -26,6 +28,14 @@ def const(result=None):
     return fn
 
 
+def make_grid(images, rows, cols):
+    w, h = images[0].size
+    grid = Image.new('RGB', size=(cols*w, rows*h))
+    for i, image in enumerate(images):
+        grid.paste(image, box=(i % cols*w, i//cols*h))
+    return grid
+
+
 def get_models(pretrained_model_name_or_path: str):
     tokenizer = MultiCLIPTokenizer.from_pretrained(pretrained_model_name_or_path, subfolder='tokenizer')
     text_encoder = CLIPTextModel.from_pretrained(pretrained_model_name_or_path, subfolder='text_encoder')
@@ -40,6 +50,107 @@ def get_models(pretrained_model_name_or_path: str):
     return tokenizer, text_encoder, vae, unet, noise_scheduler, sample_scheduler, embeddings
 
 
+def save_samples(
+    accelerator: Accelerator,
+    unet: UNet2DConditionModel,
+    text_encoder: CLIPTextModel,
+    tokenizer: MultiCLIPTokenizer,
+    vae: AutoencoderKL,
+    sample_scheduler: DPMSolverMultistepScheduler,
+    train_dataloader: DataLoader,
+    val_dataloader: DataLoader,
+    dtype: torch.dtype,
+    output_dir: Path,
+    seed: int,
+    step: int,
+    batch_size: int = 1,
+    num_batches: int = 1,
+    num_steps: int = 20,
+    guidance_scale: float = 7.5,
+    image_size: Optional[int] = None,
+):
+    print(f"Saving samples for step {step}...")
+
+    samples_path = output_dir.joinpath("samples")
+
+    grid_cols = min(batch_size, 4)
+    grid_rows = (num_batches * batch_size) // grid_cols
+
+    unet = accelerator.unwrap_model(unet)
+    text_encoder = accelerator.unwrap_model(text_encoder)
+
+    orig_unet_dtype = unet.dtype
+    orig_text_encoder_dtype = text_encoder.dtype
+
+    unet.to(dtype=dtype)
+    text_encoder.to(dtype=dtype)
+
+    pipeline = VlpnStableDiffusion(
+        text_encoder=text_encoder,
+        vae=vae,
+        unet=unet,
+        tokenizer=tokenizer,
+        scheduler=sample_scheduler,
+    ).to(accelerator.device)
+    pipeline.set_progress_bar_config(dynamic_ncols=True)
+
+    generator = torch.Generator(device=accelerator.device).manual_seed(seed)
+
+    for pool, data, gen in [
+        ("stable", val_dataloader, generator),
+        ("val", val_dataloader, None),
+        ("train", train_dataloader, None)
+    ]:
+        all_samples = []
+        file_path = samples_path.joinpath(pool, f"step_{step}.jpg")
+        file_path.parent.mkdir(parents=True, exist_ok=True)
+
+        batches = list(itertools.islice(itertools.cycle(data), batch_size * num_batches))
+        prompt_ids = [
+            prompt
+            for batch in batches
+            for prompt in batch["prompt_ids"]
+        ]
+        nprompt_ids = [
+            prompt
+            for batch in batches
+            for prompt in batch["nprompt_ids"]
+        ]
+
+        for i in range(num_batches):
+            start = i * batch_size
+            end = (i + 1) * batch_size
+            prompt = prompt_ids[start:end]
+            nprompt = nprompt_ids[start:end]
+
+            samples = pipeline(
+                prompt=prompt,
+                negative_prompt=nprompt,
+                height=image_size,
+                width=image_size,
+                generator=gen,
+                guidance_scale=guidance_scale,
+                num_inference_steps=num_steps,
+                output_type='pil'
+            ).images
+
+            all_samples += samples
+
+        image_grid = make_grid(all_samples, grid_rows, grid_cols)
+        image_grid.save(file_path, quality=85)
+
+    unet.to(dtype=orig_unet_dtype)
+    text_encoder.to(dtype=orig_text_encoder_dtype)
+
+    del unet
+    del text_encoder
+    del generator
+    del pipeline
+
+    if torch.cuda.is_available():
+        torch.cuda.empty_cache()
+
+
 def generate_class_images(
     accelerator: Accelerator,
     text_encoder: CLIPTextModel,
@@ -109,6 +220,10 @@ def get_models(pretrained_model_name_or_path: str):
 
     embeddings = patch_managed_embeddings(text_encoder)
 
+    vae.requires_grad_(False)
+    unet.requires_grad_(False)
+    text_encoder.requires_grad_(False)
+
     return tokenizer, text_encoder, vae, unet, noise_scheduler, sample_scheduler, embeddings
 
 
@@ -427,6 +542,9 @@ def train(
         seed,
     )
 
+    if accelerator.is_main_process:
+        accelerator.init_trackers("textual_inversion")
+
     train_loop(
         accelerator=accelerator,
         optimizer=optimizer,
diff --git a/training/strategy/ti.py b/training/strategy/ti.py
new file mode 100644
index 0000000..83dc566
--- /dev/null
+++ b/training/strategy/ti.py
@@ -0,0 +1,164 @@
+from contextlib import nullcontext
+from typing import Optional
+from functools import partial
+from contextlib import contextmanager, nullcontext
+from pathlib import Path
+
+import torch
+from torch.utils.data import DataLoader
+
+from accelerate import Accelerator
+from transformers import CLIPTextModel
+from diffusers import AutoencoderKL, UNet2DConditionModel, DPMSolverMultistepScheduler
+
+from slugify import slugify
+
+from models.clip.tokenizer import MultiCLIPTokenizer
+from training.util import EMAModel
+from training.functional import save_samples
+
+
+def textual_inversion_strategy(
+    accelerator: Accelerator,
+    unet: UNet2DConditionModel,
+    text_encoder: CLIPTextModel,
+    tokenizer: MultiCLIPTokenizer,
+    vae: AutoencoderKL,
+    sample_scheduler: DPMSolverMultistepScheduler,
+    train_dataloader: DataLoader,
+    val_dataloader: DataLoader,
+    dtype: torch.dtype,
+    output_dir: Path,
+    seed: int,
+    placeholder_tokens: list[str],
+    placeholder_token_ids: list[list[int]],
+    learning_rate: float,
+    gradient_checkpointing: bool = False,
+    use_emb_decay: bool = False,
+    emb_decay_target: float = 0.4,
+    emb_decay_factor: float = 1,
+    emb_decay_start: float = 1e-4,
+    use_ema: bool = False,
+    ema_inv_gamma: float = 1.0,
+    ema_power: int = 1,
+    ema_max_decay: float = 0.9999,
+    sample_batch_size: int = 1,
+    sample_num_batches: int = 1,
+    sample_num_steps: int = 20,
+    sample_guidance_scale: float = 7.5,
+    sample_image_size: Optional[int] = None,
+):
+    save_samples_ = partial(
+        save_samples,
+        accelerator=accelerator,
+        unet=unet,
+        text_encoder=text_encoder,
+        tokenizer=tokenizer,
+        vae=vae,
+        sample_scheduler=sample_scheduler,
+        train_dataloader=train_dataloader,
+        val_dataloader=val_dataloader,
+        dtype=dtype,
+        output_dir=output_dir,
+        seed=seed,
+        batch_size=sample_batch_size,
+        num_batches=sample_num_batches,
+        num_steps=sample_num_steps,
+        guidance_scale=sample_guidance_scale,
+        image_size=sample_image_size,
+    )
+
+    if use_ema:
+        ema_embeddings = EMAModel(
+            text_encoder.text_model.embeddings.temp_token_embedding.parameters(),
+            inv_gamma=ema_inv_gamma,
+            power=ema_power,
+            max_value=ema_max_decay,
+        )
+    else:
+        ema_embeddings = None
+
+    def on_prepare():
+        text_encoder.text_model.embeddings.temp_token_embedding.requires_grad_(True)
+
+        if use_ema:
+            ema_embeddings.to(accelerator.device)
+
+        if gradient_checkpointing:
+            unet.train()
+
+    @contextmanager
+    def on_train(epoch: int):
+        try:
+            tokenizer.train()
+            yield
+        finally:
+            pass
+
+    @contextmanager
+    def on_eval():
+        try:
+            tokenizer.eval()
+
+            ema_context = ema_embeddings.apply_temporary(
+                text_encoder.text_model.embeddings.temp_token_embedding.parameters()) if use_ema else nullcontext()
+
+            with ema_context:
+                yield
+        finally:
+            pass
+
+    @torch.no_grad()
+    def on_after_optimize(lr: float):
+        if use_emb_decay:
+            text_encoder.text_model.embeddings.normalize(
+                emb_decay_target,
+                min(1.0, max(0.0, emb_decay_factor * ((lr - emb_decay_start) / (learning_rate - emb_decay_start))))
+            )
+
+        if use_ema:
+            ema_embeddings.step(text_encoder.text_model.embeddings.temp_token_embedding.parameters())
+
+    def on_log():
+        if use_ema:
+            return {"ema_decay": ema_embeddings.decay}
+        return {}
+
+    @torch.no_grad()
+    def on_checkpoint(step, postfix):
+        print(f"Saving checkpoint for step {step}...")
+
+        checkpoints_path = output_dir.joinpath("checkpoints")
+        checkpoints_path.mkdir(parents=True, exist_ok=True)
+
+        text_encoder = accelerator.unwrap_model(text_encoder)
+
+        ema_context = ema_embeddings.apply_temporary(
+            text_encoder.text_model.embeddings.temp_token_embedding.parameters()
+        ) if ema_embeddings is not None else nullcontext()
+
+        with ema_context:
+            for (token, ids) in zip(placeholder_tokens, placeholder_token_ids):
+                text_encoder.text_model.embeddings.save_embed(
+                    ids,
+                    checkpoints_path.joinpath(f"{slugify(token)}_{step}_{postfix}.bin")
+                )
+
+    @torch.no_grad()
+    def on_sample(step):
+        ema_context = ema_embeddings.apply_temporary(
+            text_encoder.text_model.embeddings.temp_token_embedding.parameters()
+        ) if ema_embeddings is not None else nullcontext()
+
+        with ema_context:
+            save_samples_(step=step)
+
+    return {
+        "on_prepare": on_prepare,
+        "on_train": on_train,
+        "on_eval": on_eval,
+        "on_after_optimize": on_after_optimize,
+        "on_log": on_log,
+        "on_checkpoint": on_checkpoint,
+        "on_sample": on_sample,
+    }