File size: 3,852 Bytes
256a159 |
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 |
import random
import re
import torch
class InstructBlipMMBenchPostProcessor:
""""Post processor for MiniGPT-4 on MMBench."""
def __init__(self) -> None:
pass
def __call__(self, output_token: torch.tensor, tokenizer) -> str:
# convert output id 0 to 2 (eos_token_id)
output_token[output_token == 0] = 2
output_text = tokenizer.decode(output_token,
add_special_tokens=False) # noqa
output_text = self._extract_key_words(output_text.strip())
return output_text
def _extract_key_words(self, output_text: str) -> str:
output_text = output_text.split('###')[0]
output_text = output_text.split('Assistant:')[-1].strip()
output_text = output_text.strip('</s><s>')
output_text = output_text.strip('</Img>')
output_text = output_text.strip()
pattern = re.compile(r'([A-Z]\.)')
res = pattern.findall(output_text)
if len(res) > 0:
output_text = res[0][:-1]
return output_text
class InstructBlipCOCOCaptionPostProcessor:
""""Post processor for InstructBlip on COCO Caption."""
def __init__(self) -> None:
pass
def __call__(self, output_token: torch.tensor, tokenizer) -> str:
output_token[output_token == 0] = 2
output_text = tokenizer.decode(output_token,
add_special_tokens=False) # noqa
output_text = output_text.split('###')[0]
output_text = output_text.split('Assistant:')[-1].strip()
output_text = output_text.strip('</s><s>')
output_text = output_text.strip('</Img>')
output_text = output_text.strip()
return output_text
class InstructBlipVQAPostProcessor:
""""Post processor for InstructBlip on VQA."""
def __init__(self) -> None:
pass
def __call__(self, output_token: torch.tensor, tokenizer) -> str:
output_token[output_token == 0] = 2
output_text = tokenizer.decode(output_token,
add_special_tokens=False) # noqa
output_text = output_text.split('###')[0]
output_text = output_text.split('Assistant:')[-1].strip()
output_text = output_text.strip('</s><s>')
output_text = output_text.strip('</Img>')
output_text = output_text.strip()
return output_text
class InstructBlipScienceQAPostProcessor:
""""Post processor for InstructBlip on ScienceQA."""
def __init__(self) -> None:
pass
def __call__(self, output_token: torch.tensor, tokenizer) -> str:
output_token[output_token == 0] = 2
output_text = tokenizer.decode(output_token,
add_special_tokens=False) # noqa
output_text = output_text.split('###')[0]
output_text = output_text.split('Assistant:')[-1].strip()
output_text = output_text.strip('</s><s>')
output_text = output_text.strip('</Img>')
output_text = output_text.strip()
pattern = re.compile(r'\(([A-Z])\)')
output_text = pattern.findall(output_text)
if len(output_text) == 0:
output_text = random.choice(['A', 'B', 'C', 'D'])
else:
output_text = output_text[0]
return output_text
class InstructBlipVSRPostProcessor:
""""Post processor for InstructBlip on VSR."""
def __init__(self) -> None:
pass
def __call__(self, output_token: torch.tensor, tokenizer) -> str:
output_token[output_token == 0] = 2
output_text = tokenizer.decode(output_token, add_special_tokens=False)
pattern = r'yes|no|Yes|No'
output_text = re.findall(pattern, output_text)
if len(output_text) > 0:
output_text = output_text[0].lower()
return output_text
|