Spaces:
Runtime error
Runtime error
File size: 6,620 Bytes
a48f2db 59a7eba a48f2db a57c1e5 59a7eba a48f2db 59a7eba a48f2db 59a7eba a48f2db 59a7eba a48f2db a57c1e5 a48f2db 9111b95 a48f2db 59a7eba a57c1e5 a48f2db 9811800 a48f2db 59a7eba a48f2db 59a7eba a48f2db 59a7eba a48f2db 59a7eba |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 |
import torch
from onnxruntime.quantization import quantize_dynamic,QuantType
import os
import subprocess
import numpy as np
import pandas as pd
import transformers
from pathlib import Path
import streamlit as st
import yaml
def read_yaml(file_path):
with open(file_path, "r") as f:
return yaml.safe_load(f)
config = read_yaml('config.yaml')
zs_chkpt=config['ZEROSHOT_CLF']['zs_chkpt']
zs_mdl_dir=config['ZEROSHOT_CLF']['zs_mdl_dir']
zs_onnx_mdl_dir=config['ZEROSHOT_CLF']['zs_onnx_mdl_dir']
zs_onnx_mdl_name=config['ZEROSHOT_CLF']['zs_onnx_mdl_name']
zs_onnx_quant_mdl_name=config['ZEROSHOT_CLF']['zs_onnx_quant_mdl_name']
zs_mlm_chkpt=config['ZEROSHOT_MLM']['zs_mlm_chkpt']
zs_mlm_mdl_dir=config['ZEROSHOT_MLM']['zs_mlm_mdl_dir']
zs_mlm_onnx_mdl_dir=config['ZEROSHOT_MLM']['zs_mlm_onnx_mdl_dir']
zs_mlm_onnx_mdl_name=config['ZEROSHOT_MLM']['zs_mlm_onnx_mdl_name']
def zero_shot_classification(premise: str, labels: str, model, tokenizer):
"""
Args:
premise:
labels:
model:
tokenizer:
Returns:
"""
try:
labels=labels.split(',')
labels=[l.lower() for l in labels]
except:
raise Exception("please pass atleast 2 labels to classify")
premise=premise.lower()
labels_prob=[]
for l in labels:
hypothesis= f'this is an example of {l}'
input = tokenizer.encode(premise,hypothesis,
return_tensors='pt',
truncation_strategy='only_first')
output = model(input)
entail_contra_prob = output['logits'][:,[0,2]].softmax(dim=1)[:,1].item() #only normalizing entail & contradict probabilties
labels_prob.append(entail_contra_prob)
labels_prob_norm=[np.round(100*c/np.sum(labels_prob),1) for c in labels_prob]
df=pd.DataFrame({'labels':labels,
'Probability':labels_prob_norm})
return df
##example
# zero_shot_classification(premise='Tiny worms and breath analyzers could screen for disease while it’s early and treatable',
# labels='science, sports, museum')
def create_onnx_model_zs_nli(zs_onnx_mdl_dir=zs_onnx_mdl_dir):
"""
Args:
zs_onnx_mdl_dir:
Returns:
"""
# create onnx model using
if not os.path.exists(zs_onnx_mdl_dir):
try:
subprocess.run(['python3', '-m', 'transformers.onnx',
'--model=valhalla/distilbart-mnli-12-1',
'--feature=sequence-classification',
'--atol=1e-3',
zs_onnx_mdl_dir])
except Exception as e:
print(e)
# #create quanitzed model from vanila onnx
# quantize_dynamic(f"{zs_onnx_mdl_dir}/{zs_onnx_mdl_name}",
# f"{zs_onnx_mdl_dir}/{zs_onnx_quant_mdl_name}",
# weight_type=QuantType.QUInt8)
else:
pass
def zero_shot_classification_nli_onnx(premise,labels,_session,_tokenizer,hypothesis="This is an example of"):
"""
Args:
premise:
labels:
_session:
_tokenizer:
hypothesis:
Returns:
"""
try:
labels=labels.split(',')
labels=[l.lower() for l in labels]
except:
raise Exception("please pass atleast 2 labels to classify")
premise=premise.lower()
labels_prob=[]
for l in labels:
hypothesis= f"{hypothesis} {l}"
inputs = _tokenizer(premise,hypothesis,
return_tensors='pt',
truncation_strategy='only_first')
input_feed = {
"input_ids": np.array(inputs['input_ids']),
"attention_mask": np.array((inputs['attention_mask']))
}
output = _session.run(output_names=["logits"],input_feed=dict(input_feed))[0] #returns logits as array
output=torch.from_numpy(output)
entail_contra_prob = output[:,[0,2]].softmax(dim=1)[:,1].item() #only normalizing entail & contradict probabilties
labels_prob.append(entail_contra_prob)
labels_prob_norm=[np.round(100*c/np.sum(labels_prob),1) for c in labels_prob]
df=pd.DataFrame({'labels':labels,
'Probability':labels_prob_norm})
return df
def create_onnx_model_zs_mlm(_model, _tokenizer,zs_mlm_onnx_mdl_dir=zs_mlm_onnx_mdl_dir):
"""
Args:
_model:
_tokenizer:
zs_mlm_onnx_mdl_dir:
Returns:
"""
if not os.path.exists(zs_mlm_onnx_mdl_dir):
try:
subprocess.run(['python3', '-m', 'transformers.onnx',
f'--model={zs_mlm_chkpt}',
'--feature=masked-lm',
zs_mlm_onnx_mdl_dir])
except:
pass
else:
pass
def zero_shot_classification_fillmask_onnx(premise,hypothesis,labels,_session,_tokenizer):
"""
Args:
premise:
hypothesis:
labels:
_session:
_tokenizer:
Returns:
"""
try:
labels=labels.split(',')
labels=[l.lower().rstrip().lstrip() for l in labels]
except:
raise Exception("please pass atleast 2 labels to classify")
premise=premise.lower()
hypothesis=hypothesis.lower()
final_input= f"{premise}.{hypothesis} [MASK]" #this can change depending on chkpt, this is for bert-base-uncased chkpt
_inputs=_tokenizer(final_input,padding=True, truncation=True,
return_tensors="pt")
input_feed={
'input_ids': np.array(_inputs['input_ids']),
'token_type_ids': np.array(_inputs['token_type_ids']),
'attention_mask': np.array(_inputs['attention_mask'])
}
output=_session.run(output_names=['logits'],input_feed=dict(input_feed))[0]
mask_token_index = np.argwhere(_inputs["input_ids"] == _tokenizer.mask_token_id)[1,0]
mask_token_logits=output[0,mask_token_index,:]
#seacrh for logits of input labels
#encode the labels and get the label id -
labels_logits=[]
for l in labels:
encoded_label=_tokenizer.encode(l)[1]
labels_logits.append(mask_token_logits[encoded_label])
#do a softmax on the logits
labels_logits=np.array(labels_logits)
labels_logits=torch.from_numpy(labels_logits)
labels_logits=labels_logits.softmax(dim=0)
output= {'Labels':labels,
'Probability':labels_logits}
df_output = pd.DataFrame(output)
df_output['Probability'] = df_output['Probability'].apply(lambda x: np.round(100*x, 1))
return df_output
|