Spaces:
Paused
Paused
File size: 5,174 Bytes
5885496 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 |
import argparse
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch
import os
from llava.conversation import conv_templates, SeparatorStyle
from llava.utils import disable_torch_init
from transformers import CLIPVisionModel, CLIPImageProcessor, StoppingCriteria
from llava.model import *
from llava.model.utils import KeywordsStoppingCriteria
from PIL import Image
import os
import requests
from PIL import Image
from io import BytesIO
DEFAULT_IMAGE_TOKEN = "<image>"
DEFAULT_IMAGE_PATCH_TOKEN = "<im_patch>"
DEFAULT_IM_START_TOKEN = "<im_start>"
DEFAULT_IM_END_TOKEN = "<im_end>"
def load_image(image_file):
if image_file.startswith('http') or image_file.startswith('https'):
response = requests.get(image_file)
image = Image.open(BytesIO(response.content)).convert('RGB')
else:
image = Image.open(image_file).convert('RGB')
return image
def eval_model(args):
# Model
disable_torch_init()
model_name = os.path.expanduser(args.model_name)
tokenizer = AutoTokenizer.from_pretrained(model_name)
if "mpt" in model_name.lower():
model = LlavaMPTForCausalLM.from_pretrained(model_name, low_cpu_mem_usage=True, torch_dtype=torch.float16, use_cache=True).cuda()
else:
# model = LlavaLlamaForCausalLM.from_pretrained(model_name, low_cpu_mem_usage=True, torch_dtype=torch.float16, use_cache=True).cuda()
model = LlavaLlamaForCausalLM.from_pretrained(model_name, torch_dtype=torch.float16, device_map='auto')#.cuda()
image_processor = CLIPImageProcessor.from_pretrained(model.config.mm_vision_tower, torch_dtype=torch.float16)
mm_use_im_start_end = getattr(model.config, "mm_use_im_start_end", False)
tokenizer.add_tokens([DEFAULT_IMAGE_PATCH_TOKEN], special_tokens=True)
if mm_use_im_start_end:
tokenizer.add_tokens([DEFAULT_IM_START_TOKEN, DEFAULT_IM_END_TOKEN], special_tokens=True)
vision_tower = model.get_model().vision_tower[0]
if vision_tower.device.type == 'meta':
vision_tower = CLIPVisionModel.from_pretrained(vision_tower.config._name_or_path, torch_dtype=torch.float16, low_cpu_mem_usage=True).cuda()
model.get_model().vision_tower[0] = vision_tower
else:
vision_tower.to(device='cuda', dtype=torch.float16)
vision_config = vision_tower.config
vision_config.im_patch_token = tokenizer.convert_tokens_to_ids([DEFAULT_IMAGE_PATCH_TOKEN])[0]
vision_config.use_im_start_end = mm_use_im_start_end
if mm_use_im_start_end:
vision_config.im_start_token, vision_config.im_end_token = tokenizer.convert_tokens_to_ids([DEFAULT_IM_START_TOKEN, DEFAULT_IM_END_TOKEN])
image_token_len = (vision_config.image_size // vision_config.patch_size) ** 2
qs = args.query
if mm_use_im_start_end:
qs = qs + '\n' + DEFAULT_IM_START_TOKEN + DEFAULT_IMAGE_PATCH_TOKEN * image_token_len + DEFAULT_IM_END_TOKEN
else:
qs = qs + '\n' + DEFAULT_IMAGE_PATCH_TOKEN * image_token_len
if "v1" in model_name.lower():
conv_mode = "llava_v1"
elif "mpt" in model_name.lower():
conv_mode = "mpt_multimodal"
else:
conv_mode = "multimodal"
if args.conv_mode is not None and conv_mode != args.conv_mode:
print('[WARNING] the auto inferred conversation mode is {}, while `--conv-mode` is {}, using {}'.format(conv_mode, args.conv_mode, args.conv_mode))
else:
args.conv_mode = conv_mode
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()
inputs = tokenizer([prompt])
image = load_image(args.image_file)
image_tensor = image_processor.preprocess(image, return_tensors='pt')['pixel_values'][0]
input_ids = torch.as_tensor(inputs.input_ids).cuda()
stop_str = conv.sep if conv.sep_style != SeparatorStyle.TWO else conv.sep2
keywords = [stop_str]
stopping_criteria = KeywordsStoppingCriteria(keywords, tokenizer, input_ids)
with torch.inference_mode():
output_ids = model.generate(input_ids, images=image_tensor.unsqueeze(0).half().cuda(), do_sample=True, temperature=0.2, max_new_tokens=1024, stopping_criteria=[stopping_criteria])
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()
if outputs.endswith(stop_str):
outputs = outputs[:-len(stop_str)]
outputs = outputs.strip()
print(outputs)
import pdb; pdb.set_trace()
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument("--model-name", type=str, default="facebook/opt-350m")
parser.add_argument("--image-file", type=str, required=True)
parser.add_argument("--query", type=str, required=True)
parser.add_argument("--conv-mode", type=str, default=None)
args = parser.parse_args()
eval_model(args)
|