dc/eval/model_vqa_loader.py

import argparse
import torch
import os
import json
from tqdm import tqdm
import shortuuid

from dc.constants import IMAGE_TOKEN_INDEX, DEFAULT_IMAGE_TOKEN, DEFAULT_IM_START_TOKEN, DEFAULT_IM_END_TOKEN
from dc.conversation import conv_templates, SeparatorStyle
from dc.model.builder import load_pretrained_model
from dc.utils import disable_torch_init
from dc.mm_utils import tokenizer_image_token, process_images, get_model_name_from_path
from torch.utils.data import Dataset, DataLoader

from PIL import Image
import math


def split_list(lst, n):
    """Split a list into n (roughly) equal-sized chunks"""
    chunk_size = math.ceil(len(lst) / n)  # integer division
    return [lst[i:i+chunk_size] for i in range(0, len(lst), chunk_size)]


def get_chunk(lst, n, k):
    chunks = split_list(lst, n)
    return chunks[k]


# Custom dataset class
class CustomDataset(Dataset):
    def __init__(self, questions, image_folder, tokenizer, image_processor, model_config):
        self.questions = questions
        self.image_folder = image_folder
        self.tokenizer = tokenizer
        self.image_processor = image_processor
        self.model_config = model_config

    def __getitem__(self, index):
        line = self.questions[index]
        image_file = line["image"]
        qs = line["text"]
        
        # self.model_config.mm_use_im_start_end=False
        if self.model_config.mm_use_im_start_end:
            qs = DEFAULT_IM_START_TOKEN + DEFAULT_IMAGE_TOKEN + DEFAULT_IM_END_TOKEN + '\n' + qs
        else:
            qs = DEFAULT_IMAGE_TOKEN + '\n' + qs

        conv = conv_templates[args.conv_mode].copy()
        conv.append_message(conv.roles[0], qs)
        conv.append_message(conv.roles[1], None)
        prompt = conv.get_prompt()
        
        # llama3 !!!!!
        # prompt += '<|start_header_id|>assistant<|end_header_id|>\n\n'
        # print([prompt])

        image = Image.open(os.path.join(self.image_folder, image_file)).convert('RGB')
        image_tensor = process_images([image], self.image_processor, self.model_config)[0]

        input_ids = tokenizer_image_token(prompt, self.tokenizer, IMAGE_TOKEN_INDEX, return_tensors='pt')

        return input_ids, image_tensor

    def __len__(self):
        return len(self.questions)


# DataLoader
def create_data_loader(questions, image_folder, tokenizer, image_processor, model_config, batch_size=1, num_workers=4):
    assert batch_size == 1, "batch_size must be 1"
    dataset = CustomDataset(questions, image_folder, tokenizer, image_processor, model_config)
    data_loader = DataLoader(dataset, batch_size=batch_size, num_workers=num_workers, shuffle=False)
    return data_loader


def eval_model(args):
    # Model
    disable_torch_init()
    model_path = os.path.expanduser(args.model_path)
    model_name = get_model_name_from_path(model_path)
    tokenizer, model, image_processor, context_len = load_pretrained_model(model_path, args.model_base, model_name, load_8bit=args.load_8bit)

    terminators = [
            tokenizer.eos_token_id
        ]
    if args.conv_mode == 'llama_3':
        if tokenizer.unk_token is None:
            tokenizer.unk_token = "<|reserved_special_token_0|>"
        tokenizer.pad_token = tokenizer.unk_token
        terminators = [
            tokenizer.eos_token_id,
            tokenizer.convert_tokens_to_ids("<|eot_id|>")
        ]

    if args.projection_path is not None and os.path.exists(args.projection_path):
        print(f"Loading weights from {args.projection_path}")
        status = model.load_state_dict(torch.load(args.projection_path, map_location='cpu'), strict=False)
        if status.unexpected_keys:
            print(f"Unexpected Keys: {status.unexpected_keys}.\nThe Video-ChatGPT weights are not loaded correctly.")
        print(f"Weights loaded from {args.projection_path}")

    for n, p in model.named_parameters():
        if "model.local_enc.0.b1.conv1.conv.weight" in n:
            print(p[:5, :1])
            import time
            time.sleep(20)

    questions = [json.loads(q) for q in open(os.path.expanduser(args.question_file), "r")]
    questions = get_chunk(questions, args.num_chunks, args.chunk_idx)
    answers_file = os.path.expanduser(args.answers_file)
    os.makedirs(os.path.dirname(answers_file), exist_ok=True)
    ans_file = open(answers_file, "w")

    if 'plain' in model_name and 'finetune' not in model_name.lower() and 'mmtag' not in args.conv_mode:
        args.conv_mode = args.conv_mode + '_mmtag'
        print(f'It seems that this is a plain model, but it is not using a mmtag prompt, auto switching to {args.conv_mode}.')

    data_loader = create_data_loader(questions, args.image_folder, tokenizer, image_processor, model.config)

    for (input_ids, image_tensor), line in tqdm(zip(data_loader, questions), total=len(questions)):
        idx = line["question_id"]
        cur_prompt = line["text"]

        input_ids = input_ids.to(device='cuda', non_blocking=True)
        # print(input_ids)

        # terminators = [
        # tokenizer.eos_token_id,
        # tokenizer.convert_tokens_to_ids("<|eot_id|>")
        # ]

        with torch.inference_mode():
            # outputs_ = model.generate(
            #     input_ids,
            #     max_new_tokens=256,
            #     eos_token_id=terminators,
            #     do_sample=True,
            #     temperature=0.6,
            #     top_p=0.9,
            # )
            output_ids = model.generate(
                input_ids,
                images=image_tensor.to(dtype=torch.float16, device='cuda', non_blocking=True),
                do_sample=True if args.temperature > 0 else False,
                temperature=args.temperature,
                eos_token_id=terminators,
                top_p=args.top_p,
                num_beams=args.num_beams,
                max_new_tokens=args.max_new_tokens,
                use_cache=True)

        
        outputs = tokenizer.batch_decode(output_ids, skip_special_tokens=True)[0].strip()
        # print(tokenizer.batch_decode(output_ids))
        # input_token_len = input_ids.shape[1]
        # n_diff_input_output = (input_ids != output_ids[:, :input_token_len]).sum().item()
        # if n_diff_input_output > 0:
        #     print(f'[Warning] {n_diff_input_output} output_ids are not the same as the input_ids')
        # outputs = tokenizer.batch_decode(output_ids[:, input_token_len:], skip_special_tokens=True)[0]
        # outputs = outputs.strip()

        ans_id = shortuuid.uuid()
        ans_file.write(json.dumps({"question_id": idx,
                                   "prompt": cur_prompt,
                                   "text": outputs,
                                   "answer_id": ans_id,
                                   "model_id": model_name,
                                   "metadata": {}}) + "\n")
        # ans_file.flush()
    ans_file.close()

if __name__ == "__main__":
    parser = argparse.ArgumentParser()
    parser.add_argument("--model-path", type=str, default="facebook/opt-350m")
    parser.add_argument("--model-base", type=str, default=None)
    parser.add_argument("--image-folder", type=str, default="")
    parser.add_argument("--question-file", type=str, default="tables/question.jsonl")
    parser.add_argument("--answers-file", type=str, default="answer.jsonl")
    parser.add_argument("--conv-mode", type=str, default="llava_v1")
    parser.add_argument("--num-chunks", type=int, default=1)
    parser.add_argument("--chunk-idx", type=int, default=0)
    parser.add_argument("--temperature", type=float, default=0.2)
    parser.add_argument("--top_p", type=float, default=None)
    parser.add_argument("--num_beams", type=int, default=1)
    parser.add_argument('--load_8bit', type=bool, default=False)

    parser.add_argument("--max_new_tokens", type=int, default=128)
    parser.add_argument("--projection_path", type=str, required=False)
    args = parser.parse_args()

    eval_model(args)