Spaces:
Running
Running
File size: 6,690 Bytes
569f484 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 |
import torch
from PIL import Image
from transformers import AutoModel, AutoTokenizer
import random
import math
import numpy as np
Image.MAX_IMAGE_PIXELS = 1000000000
max_token = {
'docVQA': 100,
'textVQA': 100,
"docVQATest": 100
}
class MiniCPM_V:
def __init__(self, model_path, ckpt, device=None)->None:
self.model_path = model_path
self.ckpt = ckpt
self.model = AutoModel.from_pretrained(self.model_path, trust_remote_code=True).eval()
if self.ckpt is not None:
self.ckpt = ckpt
self.state_dict = torch.load(self.ckpt, map_location=torch.device('cpu'))
self.model.load_state_dict(self.state_dict)
self.model = self.model.to(dtype=torch.float16)
self.model.to(device)
self.tokenizer = AutoTokenizer.from_pretrained(self.model_path, trust_remote_code=True)
torch.cuda.empty_cache()
def generate(self, images, questions, datasetname):
image = Image.open(images[0]).convert('RGB')
try:
max_new_tokens = max_token[datasetname]
except:
max_new_tokens = 1024
if (datasetname == 'docVQA') or (datasetname == "docVQATest") :
prompt = "Answer the question directly with single word." + "\n" + questions[0]
elif (datasetname == 'textVQA') :
prompt = "Answer the question directly with single word." + '\n'+ questions[0]
msgs = [{'role': 'user', 'content': prompt}]
default_kwargs = dict(
max_new_tokens=max_new_tokens,
sampling=False,
num_beams=3
)
res = self.model.chat(
image=image,
msgs=msgs,
context=None,
tokenizer=self.tokenizer,
**default_kwargs
)
return [res]
def generate_with_interleaved(self, images, questions, datasetname):
try:
max_new_tokens = max_token[datasetname]
except:
max_new_tokens = 1024
prompt = "Answer the question directly with single word."
default_kwargs = dict(
max_new_tokens=max_new_tokens,
sampling=False,
num_beams=3
)
content = []
message = [
{'type': 'text', 'value': prompt},
{'type': 'image', 'value': images[0]},
{'type': 'text', 'value': questions[0]}
]
for x in message:
if x['type'] == 'text':
content.append(x['value'])
elif x['type'] == 'image':
image = Image.open(x['value']).convert('RGB')
content.append(image)
msgs = [{'role': 'user', 'content': content}]
res = self.model.chat(
image=None,
msgs=msgs,
context=None,
tokenizer=self.tokenizer,
**default_kwargs
)
if isinstance(res, tuple) and len(res) > 0:
res = res[0]
print(f"Q: {content}, \nA: {res}")
return [res]
class MiniCPM_V_2_6:
def __init__(self, model_path, ckpt, device=None)->None:
seed = 0
random.seed(seed)
np.random.seed(seed)
torch.manual_seed(seed)
torch.cuda.manual_seed_all(seed)
self.model_path = model_path
self.ckpt = ckpt
self.model = AutoModel.from_pretrained(self.model_path, trust_remote_code=True).eval()
if self.ckpt is not None:
self.ckpt = ckpt
self.state_dict = torch.load(self.ckpt, map_location=torch.device('cpu'))
self.model.load_state_dict(self.state_dict)
self.model = self.model.to(dtype=torch.bfloat16)
self.model.to(device)
self.tokenizer = AutoTokenizer.from_pretrained(self.model_path, trust_remote_code=True)
torch.cuda.empty_cache()
def generate(self, images, questions, datasetname):
image = Image.open(images[0]).convert('RGB')
try:
max_new_tokens = max_token[datasetname]
except:
max_new_tokens = 1024
if (datasetname == 'docVQA') or (datasetname == "docVQATest") :
prompt = "Answer the question directly with single word." + "\n" + questions[0]
elif (datasetname == 'textVQA') :
prompt = "Answer the question directly with single word." + '\n'+ questions[0]
msgs = [{'role': 'user', 'content': prompt}]
default_kwargs = dict(
max_new_tokens=max_new_tokens,
sampling=False,
num_beams=3
)
res = self.model.chat(
image=image,
msgs=msgs,
context=None,
tokenizer=self.tokenizer,
**default_kwargs
)
return [res]
def generate_with_interleaved(self, images, questions, datasetname):
try:
max_new_tokens = max_token[datasetname]
except:
max_new_tokens = 1024
prompt = "Answer the question directly with single word."
default_kwargs = dict(
max_new_tokens=max_new_tokens,
sampling=False,
num_beams=3
)
content = []
message = [
{'type': 'text', 'value': prompt},
{'type': 'image', 'value': images[0]},
{'type': 'text', 'value': questions[0]}
]
for x in message:
if x['type'] == 'text':
content.append(x['value'])
elif x['type'] == 'image':
image = Image.open(x['value']).convert('RGB')
img_width, img_height = image.width, image.height
if (img_width * img_height) >= (1344 * 1344):
content.append(image)
else:
ratio = math.sqrt((1344 * 1344) / (img_width * img_height))
max_img_width = int(img_width * ratio)
new_img_width = random.randint(img_width, max_img_width)
new_img_height = int(new_img_width / img_width * img_height)
resized_image = image.resize((new_img_width, new_img_height))
content.append(resized_image)
msgs = [{'role': 'user', 'content': content}]
res = self.model.chat(
image=None,
msgs=msgs,
context=None,
tokenizer=self.tokenizer,
**default_kwargs
)
if isinstance(res, tuple) and len(res) > 0:
res = res[0]
print(f"Q: {content}, \nA: {res}")
return [res]
|