import argparse import base64 import io import itertools import json import os import random import subprocess import time from functools import partial import torch from internvl.model import load_model_and_tokenizer from internvl.train.dataset import build_transform, dynamic_preprocess from PIL import Image from tqdm import tqdm from transformers import AutoConfig, AutoTokenizer ds_collections = { 'mmhal-bench_with_image': { 'root': 'data/mm-halbench/mmhal-bench_with_image.jsonl', 'max_new_tokens': 1024, 'min_new_tokens': 1, 'split': 'validation' }, } def collate_fn(batches, tokenizer): pixel_values = torch.cat([_['pixel_values'] for _ in batches], dim=0) questions = [_['question'] for _ in batches] question_ids = [_['question_id'] for _ in batches] annotations = [_['annotation'] for _ in batches] lines = [_['line'] for _ in batches] return pixel_values, questions, question_ids, annotations, lines class VQADataset(torch.utils.data.Dataset): def __init__( self, questions_path, prompt, input_size=224, dynamic_image_size=False, use_thumbnail=False, max_num=6, ): self.test = open(questions_path).readlines() 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.test) def __getitem__(self, idx): data = json.loads(self.test[idx].strip()) question = data['question'] question_id = data.get('question_id', idx) annotation = data.get('answer', None) if 'image' in data: image_file = data['image'] image_bytes = base64.b64decode(image_file) image = Image.open(io.BytesIO(image_bytes)).convert('RGB') elif 'image_path' in data: image_path = data['image_path'] image = Image.open(image_path).convert('RGB') else: raise NotImplementedError if self.dynamic_image_size: images = dynamic_preprocess(image, image_size=self.input_size, use_thumbnail=self.use_thumbnail, max_num=self.max_num) else: images = [image] pixel_values = [self.transform(image) for image in images] pixel_values = torch.stack(pixel_values) if len(self.prompt) != 0: question = question + '\n' + self.prompt question = question.strip() return { 'question_id': question_id, 'question': question, 'pixel_values': pixel_values, 'annotation': annotation, 'line': data, } 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 evaluate_chat_model(): random.seed(args.seed) input_prompt = '' if args.cot: cot_prompt = ( 'Please think step by step and ' 'output in the following format:\n\n' 'Rationale:\n{rationale}\n' 'Answer: {answer}\n\n' 'where rationale is the reasoning process of the given question.' ) input_prompt = f'{input_prompt}\n{cot_prompt}' for ds_name in args.datasets: dataset = VQADataset( questions_path=ds_collections[ds_name]['root'], prompt=input_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, question_ids, annotations, lines) 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 args.cot: pred = pred.split('Answer:')[-1].strip() preds = [pred] for question, question_id, pred, line in zip(questions, question_ids, preds, lines): del line['image'] line['model_answer'] = pred outputs.append(line) 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' results_file = os.path.join(args.out_dir, results_file) with open(results_file, 'w') as file: json.dump(merged_outputs, file) print('Results saved to {}'.format(results_file)) cmd = f'python eval/mmhal/eval_gpt_mmhal.py --response {results_file}' print(cmd) os.system(cmd) if __name__ == '__main__': parser = argparse.ArgumentParser() parser.add_argument('--checkpoint', type=str, default='') parser.add_argument('--datasets', type=str, default='mmhal-bench_with_image') 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=1) 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') parser.add_argument('--cot', action='store_true') args = parser.parse_args() if not os.path.exists(args.out_dir): os.makedirs(args.out_dir, exist_ok=True) 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))) model, tokenizer = load_model_and_tokenizer(args) 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()