import argparse import itertools import json import os import random import time from functools import partial import torch from data_utils import CAT_SHORT2LONG, process_single_sample from datasets import concatenate_datasets, load_dataset from internvl.model.internvl_chat import InternVLChatModel from internvl.train.dataset import build_transform, dynamic_preprocess from PIL import Image from torch.utils.data import Dataset from tqdm import tqdm from transformers import AutoTokenizer ds_collections = { 'MMMU_validation': { 'root': 'MMMU/MMMU', 'max_new_tokens': 10, 'min_new_tokens': 1, 'split': 'validation' }, 'MMMU_test': { 'root': 'MMMU/MMMU', 'max_new_tokens': 10, 'min_new_tokens': 1, 'split': 'test' }, 'MMMU_dev': { 'root': 'MMMU/MMMU', 'max_new_tokens': 10, 'min_new_tokens': 1, 'split': 'dev' }, } def collate_fn(batches, tokenizer): pixel_values = torch.cat([_['pixel_values'] for _ in batches], dim=0) questions = [_['question'] for _ in batches] answers = [_['answer'] for _ in batches] data_ids = [_['data_id'] for _ in batches] options = [_['option'] for _ in batches] return pixel_values, questions, answers, data_ids, options class MMMUDataset(torch.utils.data.Dataset): def __init__(self, root, split, prompt, input_size=224, dynamic_image_size=False, use_thumbnail=False, max_num=6): # run for each subject sub_dataset_list = [] for subject in tqdm(CAT_SHORT2LONG.values()): sub_dataset = load_dataset(root, subject, split=split, cache_dir=os.path.join(os.getcwd(), 'data/MMMU/')) sub_dataset_list.append(sub_dataset) # merge all dataset self.data = concatenate_datasets(sub_dataset_list) self.prompt = prompt self.input_size = input_size self.dynamic_image_size = dynamic_image_size self.use_thumbnail = use_thumbnail self.max_num = max_num self.transform = build_transform(is_train=False, input_size=input_size) def __len__(self): return len(self.data) def __getitem__(self, idx): data = process_single_sample(self.data[idx]) data_id = data['id'] question = data['question'].strip() pil_images = data['image'] question_type = data['question_type'] choices = eval(data['options']) answer = data['answer'] if 'answer' in data else None choice_list = [] options = {} multiple_choices = ['A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'I', 'J', 'K', 'L', 'M'] for i, c in enumerate(choices): choice_list.append('{}. {}'.format(multiple_choices[i], c.strip())) options[multiple_choices[i]] = c.strip() choice_txt = '\n'.join(choice_list) if self.dynamic_image_size: images = [] for idx, pil_image in enumerate(pil_images): if pil_image is not None: if idx == 0: pil_image = pil_image.resize((pil_image.width * 2, pil_image.height * 2), Image.BILINEAR) pil_image = dynamic_preprocess(pil_image, image_size=self.input_size, use_thumbnail=self.use_thumbnail, max_num=self.max_num) else: pil_image = dynamic_preprocess(pil_image, image_size=self.input_size, use_thumbnail=self.use_thumbnail, max_num=1) images += pil_image else: images = [pil_images[0]] pixel_values = [self.transform(image) for image in images] pixel_values = torch.stack(pixel_values) if len(choice_txt) > 0: question += '\n' + choice_txt question += '\n' + self.prompt[question_type] question = question.strip() return { 'question': question, 'pixel_values': pixel_values, 'answer': answer, 'option': options, 'data_id': data_id } class InferenceSampler(torch.utils.data.sampler.Sampler): def __init__(self, size): self._size = int(size) assert size > 0 self._rank = torch.distributed.get_rank() self._world_size = torch.distributed.get_world_size() self._local_indices = self._get_local_indices(size, self._world_size, self._rank) @staticmethod def _get_local_indices(total_size, world_size, rank): shard_size = total_size // world_size left = total_size % world_size shard_sizes = [shard_size + int(r < left) for r in range(world_size)] begin = sum(shard_sizes[:rank]) end = min(sum(shard_sizes[:rank + 1]), total_size) return range(begin, end) def __iter__(self): yield from self._local_indices def __len__(self): return len(self._local_indices) def post_process(pred, option): pred = pred.strip() option_candidate = list(option.keys()) if len(pred) == 1: return pred elif len(pred) != 1 and pred[0] in option_candidate: return pred[0] elif len(pred) != 1 and pred[0] not in option_candidate: for k, v in option.items(): if v in pred: return k return pred def evaluate_chat_model(): prompt = { 'multiple-choice': "Answer with the option's letter from the given choices directly.", 'open': 'Answer the question using a single word or phrase.' } random.seed(args.seed) for ds_name in args.datasets: dataset = MMMUDataset( root=ds_collections[ds_name]['root'], split=ds_collections[ds_name]['split'], prompt=prompt, input_size=image_size, dynamic_image_size=args.dynamic, use_thumbnail=use_thumbnail, max_num=args.max_num ) dataloader = torch.utils.data.DataLoader( dataset=dataset, sampler=InferenceSampler(len(dataset)), batch_size=args.batch_size, num_workers=args.num_workers, pin_memory=True, drop_last=False, collate_fn=partial(collate_fn, tokenizer=tokenizer), ) outputs = [] for _, (pixel_values, questions, answers, data_ids, options) in tqdm(enumerate(dataloader)): pixel_values = pixel_values.to(torch.bfloat16).cuda() generation_config = dict( num_beams=args.num_beams, max_new_tokens=ds_collections[ds_name]['max_new_tokens'], min_new_tokens=ds_collections[ds_name]['min_new_tokens'], do_sample=True if args.temperature > 0 else False, temperature=args.temperature, ) pred = model.chat( tokenizer=tokenizer, pixel_values=pixel_values, question=questions[0], generation_config=generation_config, verbose=True ) if len(options[0]) == 0: preds = [pred] else: preds = [post_process(pred, options[0])] for question, pred, answer, data_id in zip(questions, preds, answers, data_ids): outputs.append({ 'question': question, 'answer': pred, 'gt_answers': answer, 'data_id': data_id }) torch.distributed.barrier() world_size = torch.distributed.get_world_size() merged_outputs = [None for _ in range(world_size)] torch.distributed.all_gather_object(merged_outputs, json.dumps(outputs)) merged_outputs = [json.loads(_) for _ in merged_outputs] merged_outputs = [_ for _ in itertools.chain.from_iterable(merged_outputs)] if torch.distributed.get_rank() == 0: print(f'Evaluating {ds_name} ...') time_prefix = time.strftime('%y%m%d%H%M%S', time.localtime()) results_file = f'{ds_name}_{time_prefix}.json' output_path = os.path.join(args.out_dir, results_file) outputs = {} for item in merged_outputs: outputs[item['data_id']] = item['answer'] with open(output_path, 'w') as f: json.dump(outputs, f, indent=4) print('Results saved to {}'.format(output_path)) if ds_collections[ds_name]['split'] == 'validation': print('Evaluating ...') cmd = f'python eval/mmmu/main_eval_only.py ' \ f'--output_path {output_path} ' \ f'--answer_path eval/mmmu/answer_dict_val.json' print(cmd) os.system(cmd) time_prefix = time.strftime('%y%m%d%H%M%S', time.localtime()) results_file = f'{ds_name}_{time_prefix}.jsonl' output_path = os.path.join(args.out_dir, results_file) writer = open(output_path, 'w') for item in merged_outputs: writer.write(json.dumps(item) + '\n') writer.close() print('Results saved to {}'.format(output_path)) if __name__ == '__main__': parser = argparse.ArgumentParser() parser.add_argument('--checkpoint', type=str, default='') parser.add_argument('--datasets', type=str, default='MMMU_dev') parser.add_argument('--batch-size', type=int, default=1) parser.add_argument('--num-workers', type=int, default=1) parser.add_argument('--num-beams', type=int, default=5) parser.add_argument('--temperature', type=float, default=0.0) parser.add_argument('--out-dir', type=str, default='results') parser.add_argument('--seed', type=int, default=0) parser.add_argument('--dynamic', action='store_true') parser.add_argument('--max-num', type=int, default=6) parser.add_argument('--load-in-8bit', action='store_true') parser.add_argument('--load-in-4bit', action='store_true') parser.add_argument('--auto', action='store_true') args = parser.parse_args() if not os.path.exists(args.out_dir): os.makedirs(args.out_dir) args.datasets = args.datasets.split(',') print('datasets:', args.datasets) assert args.batch_size == 1, 'Only batch size 1 is supported' torch.distributed.init_process_group( backend='nccl', world_size=int(os.getenv('WORLD_SIZE', '1')), rank=int(os.getenv('RANK', '0')), ) torch.cuda.set_device(int(os.getenv('LOCAL_RANK', 0))) if args.auto: os.environ['CUDA_LAUNCH_BLOCKING'] = '1' kwargs = {'device_map': 'auto'} if args.auto else {} tokenizer = AutoTokenizer.from_pretrained(args.checkpoint, trust_remote_code=True, use_fast=False) model = InternVLChatModel.from_pretrained( args.checkpoint, low_cpu_mem_usage=True, torch_dtype=torch.bfloat16, load_in_8bit=args.load_in_8bit, load_in_4bit=args.load_in_4bit, **kwargs).eval() if not args.load_in_8bit and not args.load_in_4bit and not args.auto: model = model.cuda() image_size = model.config.force_image_size or model.config.vision_config.image_size use_thumbnail = model.config.use_thumbnail total_params = sum(p.numel() for p in model.parameters()) / 1e9 if total_params > 20 or args.dynamic: args.num_beams = 1 print(f'[test] total_params: {total_params}B, use num_beams: {args.num_beams}') else: print(f'[test] total_params: {total_params}B') print(f'[test] image_size: {image_size}') print(f'[test] template: {model.config.template}') print(f'[test] dynamic_image_size: {args.dynamic}') print(f'[test] use_thumbnail: {use_thumbnail}') print(f'[test] max_num: {args.max_num}') evaluate_chat_model()