app.py

import torch
import torch.amp.autocast_mode
import os
import sys
import logging
import warnings
import argparse
from PIL import Image
from pathlib import Path
from tqdm import tqdm
from torch import nn
from transformers import AutoModel, AutoProcessor, AutoTokenizer, PreTrainedTokenizer, PreTrainedTokenizerFast, AutoModelForCausalLM
from typing import List, Union
import torchvision.transforms.functional as TVF
from peft import PeftModel
import gc
import sys
IS_COLAB = 'google.colab' in sys.modules

# Constants
HF_TOKEN = os.environ.get("HF_TOKEN", None)
BASE_DIR = Path(__file__).resolve().parent # Define the base directory
CLIP_PATH = "google/siglip-so400m-patch14-384"
DEFAULT_MODEL_PATH = "unsloth/Meta-Llama-3.1-8B-Instruct-bnb-4bit"
#DEFAULT_MODEL_PATH = "Orenguteng/Llama-3.1-8B-Lexi-Uncensored-V2" # Works better but full weight.
CHECKPOINT_PATH = BASE_DIR / Path("9em124t2-499968")
LORA_PATH = CHECKPOINT_PATH / "text_model"
CAPTION_TYPE_MAP = {
    ("descriptive", "formal", False, False): ["Write a descriptive caption for this image in a formal tone."],
    ("descriptive", "formal", False, True): ["Write a descriptive caption for this image in a formal tone within {word_count} words."],
    ("descriptive", "formal", True, False): ["Write a {length} descriptive caption for this image in a formal tone."],
    ("descriptive", "informal", False, False): ["Write a descriptive caption for this image in a casual tone."],
    ("descriptive", "informal", False, True): ["Write a descriptive caption for this image in a casual tone within {word_count} words."],
    ("descriptive", "informal", True, False): ["Write a {length} descriptive caption for this image in a casual tone."],

    ("training_prompt", "formal", False, False): ["Write a stable diffusion prompt for this image."],
    ("training_prompt", "formal", False, True): ["Write a stable diffusion prompt for this image within {word_count} words."],
    ("training_prompt", "formal", True, False): ["Write a {length} stable diffusion prompt for this image."],

    ("rng-tags", "formal", False, False): ["Write a list of Booru tags for this image."],
    ("rng-tags", "formal", False, True): ["Write a list of Booru tags for this image within {word_count} words."],
    ("rng-tags", "formal", True, False): ["Write a {length} list of Booru tags for this image."],
}
IMAGE_EXTENSIONS = ('.jpg', '.jpeg', '.png', '.bmp', '.webp')

# Global Variables
IS_NF4 = True
IS_LORA = True
MODEL_PATH = DEFAULT_MODEL_PATH
device = "cuda" if torch.cuda.is_available() else "cpu"
print(f"Running on {device}")

warnings.filterwarnings("ignore", category=UserWarning)
logging.getLogger("transformers").setLevel(logging.ERROR)

class ImageAdapter(nn.Module):
    def __init__(self, input_features: int, output_features: int, ln1: bool, pos_emb: bool, num_image_tokens: int, deep_extract: bool):
        super().__init__()
        self.deep_extract = deep_extract

        if self.deep_extract:
            input_features = input_features * 5

        self.linear1 = nn.Linear(input_features, output_features)
        self.activation = nn.GELU()
        self.linear2 = nn.Linear(output_features, output_features)
        self.ln1 = nn.Identity() if not ln1 else nn.LayerNorm(input_features)
        self.pos_emb = None if not pos_emb else nn.Parameter(torch.zeros(num_image_tokens, input_features))

        # Mode token
        #self.mode_token = nn.Embedding(n_modes, output_features)
        #self.mode_token.weight.data.normal_(mean=0.0, std=0.02)   # Matches HF's implementation of llama3

        # Other tokens (<|image_start|>, <|image_end|>, <|eot_id|>)
        self.other_tokens = nn.Embedding(3, output_features)
        self.other_tokens.weight.data.normal_(mean=0.0, std=0.02)   # Matches HF's implementation of llama3

    def forward(self, vision_outputs: torch.Tensor):
        if self.deep_extract:
            x = torch.concat((
                vision_outputs[-2],
                vision_outputs[3],
                vision_outputs[7],
                vision_outputs[13],
                vision_outputs[20],
            ), dim=-1)
            assert len(x.shape) == 3, f"Expected 3, got {len(x.shape)}"  # batch, tokens, features
            assert x.shape[-1] == vision_outputs[-2].shape[-1] * 5, f"Expected {vision_outputs[-2].shape[-1] * 5}, got {x.shape[-1]}"
        else:
            x = vision_outputs[-2]

        x = self.ln1(x)

        if self.pos_emb is not None:
            assert x.shape[-2:] == self.pos_emb.shape, f"Expected {self.pos_emb.shape}, got {x.shape[-2:]}"
            x = x + self.pos_emb

        x = self.linear1(x)
        x = self.activation(x)
        x = self.linear2(x)

        # Mode token
        #mode_token = self.mode_token(mode)
        #assert mode_token.shape == (x.shape[0], mode_token.shape[1], x.shape[2]), f"Expected {(x.shape[0], 1, x.shape[2])}, got {mode_token.shape}"
        #x = torch.cat((x, mode_token), dim=1)

        # <|image_start|>, IMAGE, <|image_end|>
        other_tokens = self.other_tokens(torch.tensor([0, 1], device=self.other_tokens.weight.device).expand(x.shape[0], -1))
        assert other_tokens.shape == (x.shape[0], 2, x.shape[2]), f"Expected {(x.shape[0], 2, x.shape[2])}, got {other_tokens.shape}"
        x = torch.cat((other_tokens[:, 0:1], x, other_tokens[:, 1:2]), dim=1)

        return x

    def get_eot_embedding(self):
        return self.other_tokens(torch.tensor([2], device=self.other_tokens.weight.device)).squeeze(0)

def load_models():
    global MODEL_PATH, IS_NF4, IS_LORA
    try:
        if IS_NF4:
            from transformers import BitsAndBytesConfig
            nf4_config = BitsAndBytesConfig(load_in_4bit=True, bnb_4bit_quant_type="nf4",
                                            bnb_4bit_use_double_quant=True, bnb_4bit_compute_dtype=torch.bfloat16)
            print("Loading in NF4")
            print("Loading CLIP 📎")
            clip_processor = AutoProcessor.from_pretrained(CLIP_PATH)
            clip_model = AutoModel.from_pretrained(CLIP_PATH).vision_model
            if (CHECKPOINT_PATH / "clip_model.pt").exists():
                print("Loading VLM's custom vision model 📎")
                checkpoint = torch.load(CHECKPOINT_PATH / "clip_model.pt", map_location='cpu', weights_only=False)
                checkpoint = {k.replace("_orig_mod.module.", ""): v for k, v in checkpoint.items()}
                clip_model.load_state_dict(checkpoint)
                del checkpoint
            clip_model.eval().requires_grad_(False).to(device)

            print("Loading tokenizer 🪙")
            tokenizer = AutoTokenizer.from_pretrained(MODEL_PATH, use_fast=False)
            assert isinstance(tokenizer, (PreTrainedTokenizer, PreTrainedTokenizerFast)), f"Tokenizer is of type {type(tokenizer)}"

            print(f"Loading LLM: {MODEL_PATH} 🤖")
            text_model = AutoModelForCausalLM.from_pretrained(MODEL_PATH, quantization_config=nf4_config, device_map=device, torch_dtype=torch.bfloat16).eval()

            if False and IS_LORA and LORA_PATH.exists(): # omitted
                print("Loading VLM's custom text model 🤖")
                text_model = PeftModel.from_pretrained(model=text_model, model_id=LORA_PATH, device_map=device, quantization_config=nf4_config)
                text_model = text_model.merge_and_unload(safe_merge=True) # to avoid PEFT bug https://github.com/huggingface/transformers/issues/28515
            else: print("VLM's custom text model isn't loaded 🤖")

            print("Loading image adapter 🖼️")
            image_adapter = ImageAdapter(clip_model.config.hidden_size, text_model.config.hidden_size, False, False, 38, False).eval().to("cpu")
            image_adapter.load_state_dict(torch.load(CHECKPOINT_PATH / "image_adapter.pt", map_location="cpu", weights_only=False))
            image_adapter.eval().to(device)
        else:
            print("Loading in bfloat16")
            print("Loading CLIP 📎")
            clip_processor = AutoProcessor.from_pretrained(CLIP_PATH)
            clip_model = AutoModel.from_pretrained(CLIP_PATH).vision_model
            if (CHECKPOINT_PATH / "clip_model.pt").exists():
                print("Loading VLM's custom vision model 📎")
                checkpoint = torch.load(CHECKPOINT_PATH / "clip_model.pt", map_location='cpu', weights_only=False)
                checkpoint = {k.replace("_orig_mod.module.", ""): v for k, v in checkpoint.items()}
                clip_model.load_state_dict(checkpoint)
                del checkpoint
            clip_model.eval().requires_grad_(False).to(device)

            print("Loading tokenizer 🪙")
            tokenizer = AutoTokenizer.from_pretrained(MODEL_PATH, use_fast=False)
            assert isinstance(tokenizer, (PreTrainedTokenizer, PreTrainedTokenizerFast)), f"Tokenizer is of type {type(tokenizer)}"

            print(f"Loading LLM: {MODEL_PATH} 🤖")
            text_model = AutoModelForCausalLM.from_pretrained(MODEL_PATH, device_map="auto", torch_dtype=torch.bfloat16).eval() # device_map="auto" may cause LoRA issue

            if IS_LORA and LORA_PATH.exists():
                print("Loading VLM's custom text model 🤖")
                text_model = PeftModel.from_pretrained(model=text_model, model_id=LORA_PATH, device_map=device)
                text_model = text_model.merge_and_unload(safe_merge=True) # to avoid PEFT bug https://github.com/huggingface/transformers/issues/28515
            else: print("VLM's custom text model isn't loaded 🤖")

            print("Loading image adapter 🖼️")
            image_adapter = ImageAdapter(clip_model.config.hidden_size, text_model.config.hidden_size, False, False, 38, False).eval().to("cpu")
            image_adapter.load_state_dict(torch.load(CHECKPOINT_PATH / "image_adapter.pt", map_location="cpu", weights_only=False))
    except Exception as e:
        print(f"Error loading models: {e}")
        sys.exit(1)
    finally:
        torch.cuda.empty_cache()
        gc.collect()
    return clip_processor, clip_model, tokenizer, text_model, image_adapter

@torch.inference_mode()
def stream_chat(input_images: List[Image.Image], caption_type: str, caption_tone: str, caption_length: Union[str, int],

                max_new_tokens: int, top_p: float, temperature: float, batch_size: int, pbar: tqdm, models: tuple) -> List[str]:
    global MODEL_PATH
    clip_processor, clip_model, tokenizer, text_model, image_adapter = models
    torch.cuda.empty_cache()
    all_captions = []

    # 'any' means no length specified
    length = None if caption_length == "any" else caption_length

    if isinstance(length, str):
        try:
            length = int(length)
        except ValueError:
            pass

    # 'rng-tags' and 'training_prompt' don't have formal/informal tones
    if caption_type == "rng-tags" or caption_type == "training_prompt":
        caption_tone = "formal"

    # Build prompt
    prompt_key = (caption_type, caption_tone, isinstance(length, str), isinstance(length, int))
    if prompt_key not in CAPTION_TYPE_MAP:
        raise ValueError(f"Invalid caption type: {prompt_key}")

    prompt_str = CAPTION_TYPE_MAP[prompt_key][0].format(length=length, word_count=length)
    print(f"Prompt: {prompt_str}")

    for i in range(0, len(input_images), batch_size):
        batch = input_images[i:i+batch_size]
        # Preprocess image
        for input_image in input_images:
            try:
                image = input_image.resize((384, 384), Image.LANCZOS)
                pixel_values = TVF.pil_to_tensor(image).unsqueeze(0) / 255.0
                pixel_values = TVF.normalize(pixel_values, [0.5], [0.5])
                pixel_values = pixel_values.to(device)
            except ValueError as e:
                print(f"Error processing image: {e}")
                print("Skipping this image and continuing...")
                continue

            # Embed image
            with torch.amp.autocast_mode.autocast(device, enabled=True):
                vision_outputs = clip_model(pixel_values=pixel_values, output_hidden_states=True)
                image_features = vision_outputs.hidden_states
                embedded_images = image_adapter(image_features).to(device)

            # Tokenize the prompt
            prompt = tokenizer.encode(prompt_str, return_tensors='pt', padding=False, truncation=False, add_special_tokens=False)

            # Embed prompt
            prompt_embeds = text_model.model.embed_tokens(prompt.to(device))
            assert prompt_embeds.shape == (1, prompt.shape[1], text_model.config.hidden_size), f"Prompt shape is {prompt_embeds.shape}, expected {(1, prompt.shape[1], text_model.config.hidden_size)}"
            embedded_bos = text_model.model.embed_tokens(torch.tensor([[tokenizer.bos_token_id]], device=text_model.device, dtype=torch.int64))
            eot_embed = image_adapter.get_eot_embedding().unsqueeze(0).to(dtype=text_model.dtype)

            # Construct prompts
            inputs_embeds = torch.cat([
                embedded_bos.expand(embedded_images.shape[0], -1, -1),
                embedded_images.to(dtype=embedded_bos.dtype),
                prompt_embeds.expand(embedded_images.shape[0], -1, -1),
                eot_embed.expand(embedded_images.shape[0], -1, -1),
            ], dim=1)

            input_ids = torch.cat([
                torch.tensor([[tokenizer.bos_token_id]], dtype=torch.long),
                torch.zeros((1, embedded_images.shape[1]), dtype=torch.long),
                prompt,
                torch.tensor([[tokenizer.convert_tokens_to_ids("<|eot_id|>")]], dtype=torch.long),
            ], dim=1).to(device)
            attention_mask = torch.ones_like(input_ids)

            generate_ids = text_model.generate(input_ids=input_ids, inputs_embeds=inputs_embeds, attention_mask=attention_mask, do_sample=True, 
                                            suppress_tokens=None, max_new_tokens=max_new_tokens, top_p=top_p, temperature=temperature)

            # Trim off the prompt
            generate_ids = generate_ids[:, input_ids.shape[1]:]
            if generate_ids[0][-1] == tokenizer.eos_token_id or generate_ids[0][-1] == tokenizer.convert_tokens_to_ids("<|eot_id|>"):
                generate_ids = generate_ids[:, :-1]

            caption = tokenizer.batch_decode(generate_ids, skip_special_tokens=False, clean_up_tokenization_spaces=False)[0]
            all_captions.append(caption.strip())

        if pbar:
            pbar.update(len(batch))

    return all_captions

def process_directory(input_dir: Path, output_dir: Path, caption_type: str, caption_tone: str, caption_length: Union[str, int],

                      max_new_tokens: int, top_p: float, temperature: float, batch_size: int, models: tuple):
    output_dir.mkdir(parents=True, exist_ok=True)
    image_files = [f for f in input_dir.iterdir() if f.suffix.lower() in IMAGE_EXTENSIONS]
    images_to_process = [f for f in image_files if not (output_dir / f"{f.stem}.txt").exists()]

    if not images_to_process:
        print("No new images to process.")
        return

    with tqdm(total=len(images_to_process), desc="Processing images", unit="image") as pbar:
        for i in range(0, len(images_to_process), batch_size):
            batch_files = images_to_process[i:i+batch_size]
            batch_images = [Image.open(f).convert('RGB') for f in batch_files]

            captions = stream_chat(batch_images, caption_type, caption_tone, caption_length,
                                   max_new_tokens, top_p, temperature, batch_size, pbar, models)
            
            for file, caption in zip(batch_files, captions):
                with open(output_dir / f"{file.stem}.txt", 'w', encoding='utf-8') as f:
                    f.write(caption)

            for img in batch_images:
                img.close()

def parse_arguments():
    parser = argparse.ArgumentParser(description="Process images and generate captions.")
    parser.add_argument("input", nargs='+', help="Input image file or directory (or multiple directories)")
    parser.add_argument("--output", help="Output directory (optional)")
    parser.add_argument("--bs", type=int, default=4, help="Batch size (default: 4)")
    parser.add_argument("--type", type=str, default="descriptive", choices=["descriptive", "training_prompt", "rng-tags"],
                        help='Caption Type (default: "descriptive")')
    parser.add_argument("--tone", type=str, default="formal", choices=["formal", "informal"],
                        help='Caption Tone (default: "formal")')
    parser.add_argument("--len", default="any",
                        choices=["any", "very short", "short", "medium-length", "long", "very long"] + [str(i) for i in range(20, 261, 10)],
                        help='Caption Length (default: "any")')
    parser.add_argument("--model", type=str, default=DEFAULT_MODEL_PATH,
                        help='Huggingface LLM repo (default: "unsloth/Meta-Llama-3.1-8B-bnb-4bit")')
    parser.add_argument("--bf16", action="store_true", default=False, help="Use bfloat16 (default: NF4)")
    parser.add_argument("--nolora", action="store_true", default=False, help="Disable VLM's custom text model (default: Enable)")
    parser.add_argument("--tokens", type=int, default=300, help="Max tokens (default: 300)")
    parser.add_argument("--topp", type=float, default=0.9, help="Top-P (default: 0.9)")
    parser.add_argument("--temp", type=float, default=0.6, help="Temperature (default: 0.6)")
    return parser.parse_args()

def is_valid_repo(repo_id):
    from huggingface_hub import HfApi
    import re
    try:
        if not re.fullmatch(r'^[^/,\s\"\']+/[^/,\s\"\']+$', repo_id): return False
        api = HfApi()
        if api.repo_exists(repo_id=repo_id): return True
        else: return False
    except Exception as e:
        print(f"Failed to connect {repo_id}. {e}")
        return False

def main():
    global MODEL_PATH, IS_NF4, IS_LORA
    args = parse_arguments()
    input_paths = [Path(input_path) for input_path in args.input]
    batch_size = args.bs
    caption_type = args.type
    caption_tone = args.tone
    caption_length = args.len
    max_new_tokens = args.tokens
    top_p = args.topp
    temperature = args.temp
    IS_NF4 = False if args.bf16 else True
    IS_LORA = False if args.nolora else True
    if is_valid_repo(args.model): MODEL_PATH = args.model
    else: sys.exit(1)
    models = load_models()

    for input_path in input_paths:
        if input_path.is_file() and input_path.suffix.lower() in IMAGE_EXTENSIONS:
            output_path = input_path.with_suffix('.txt')
            print(f"Processing single image 🎞️: {input_path.name}")
            with tqdm(total=1, desc="Processing image", unit="image") as pbar:
                captions = stream_chat([Image.open(input_path).convert('RGB')], caption_type, caption_tone, caption_length,
                                       max_new_tokens, top_p, temperature, 1, pbar, models)
                with open(output_path, 'w', encoding='utf-8') as f:
                    f.write(captions[0])
            print(f"Output saved to {output_path}")
        elif input_path.is_dir():
            output_path = Path(args.output) if args.output else input_path
            print(f"Processing directory 📁: {input_path}")
            print(f"Output directory 📦: {output_path}")
            print(f"Batch size 🗄️: {batch_size}")
            process_directory(input_path, output_path, caption_type, caption_tone, caption_length,
                              max_new_tokens, top_p, temperature, batch_size, models)
        else:
            print(f"Invalid input: {input_path}")
            print("Skipping...")

    if not input_paths:
        print("Usage:")
        print("For single image: python app.py [image_file] [--bs batch_size]")
        print("For directory (same input/output): python app.py [directory] [--bs batch_size]")
        print("For directory (separate input/output): python app.py [directory] --output [output_directory] [--bs batch_size]")
        print("For multiple directories: python app.py [directory1] [directory2] ... [--output output_directory] [--bs batch_size]")
        sys.exit(1)

if __name__ == "__main__":
    main()