MGM / minigemini /eval /model_vqa_loader.py
wcy1122's picture
update code
35153f6
raw
history blame
9.03 kB
import argparse
import torch
import os
import json
from tqdm import tqdm
import shortuuid
from minigemini.constants import IMAGE_TOKEN_INDEX, DEFAULT_IMAGE_TOKEN, DEFAULT_IM_START_TOKEN, DEFAULT_IM_END_TOKEN
from minigemini.conversation import conv_templates, SeparatorStyle
from minigemini.model.builder import load_pretrained_model
from minigemini.utils import disable_torch_init
from minigemini.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"]
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()
image = Image.open(os.path.join(self.image_folder, image_file)).convert('RGB')
if hasattr(self.model_config, 'image_size_aux'):
if not hasattr(self.image_processor, 'image_size_raw'):
self.image_processor.image_size_raw = self.image_processor.crop_size.copy()
self.image_processor.crop_size['height'] = self.model_config.image_size_aux
self.image_processor.crop_size['width'] = self.model_config.image_size_aux
self.image_processor.size['shortest_edge'] = self.model_config.image_size_aux
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')
image_grid = getattr(self.model_config, 'image_grid', 1)
if hasattr(self.model_config, 'image_size_aux'):
raw_shape = [self.image_processor.image_size_raw['height'] * image_grid,
self.image_processor.image_size_raw['width'] * image_grid]
image_tensor_aux = image_tensor
image_tensor = torch.nn.functional.interpolate(image_tensor[None],
size=raw_shape,
mode='bilinear',
align_corners=False)[0]
else:
image_tensor_aux = []
if image_grid >= 2:
raw_image = image_tensor.reshape(3,
image_grid,
self.image_processor.image_size_raw['height'],
image_grid,
self.image_processor.image_size_raw['width'])
raw_image = raw_image.permute(1, 3, 0, 2, 4)
raw_image = raw_image.reshape(-1, 3,
self.image_processor.image_size_raw['height'],
self.image_processor.image_size_raw['width'])
if getattr(self.model_config, 'image_global', False):
global_image = image_tensor
if len(global_image.shape) == 3:
global_image = global_image[None]
global_image = torch.nn.functional.interpolate(global_image,
size=[self.image_processor.image_size_raw['height'],
self.image_processor.image_size_raw['width']],
mode='bilinear',
align_corners=False)
# [image_crops, image_global]
raw_image = torch.cat([raw_image, global_image], dim=0)
image_tensor = raw_image.contiguous()
return input_ids, image_tensor, image_tensor_aux
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)
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 args.conv_mode 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, image_tensor_aux), line in tqdm(zip(data_loader, questions), total=len(questions)):
idx = line["question_id"]
cur_prompt = line["text"]
input_ids = input_ids.to(device=model.device, non_blocking=True)
if hasattr(model, "update_prompt"):
model.update_prompt([[cur_prompt]])
with torch.inference_mode():
output_ids = model.generate(
input_ids,
images=image_tensor.to(dtype=model.dtype, device=model.device, non_blocking=True),
images_aux=image_tensor_aux.to(dtype=model.dtype, device=model.device, non_blocking=True) if len(image_tensor_aux)>0 else None,
do_sample=True if args.temperature > 0 else False,
temperature=args.temperature,
top_p=args.top_p,
num_beams=args.num_beams,
max_new_tokens=args.max_new_tokens,
bos_token_id=tokenizer.bos_token_id, # Begin of sequence token
eos_token_id=tokenizer.eos_token_id, # End of sequence token
pad_token_id=tokenizer.pad_token_id, # Pad token
use_cache=True)
outputs = tokenizer.batch_decode(output_ids, skip_special_tokens=True)[0].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)
args = parser.parse_args()
eval_model(args)