Spaces:
Sleeping
Sleeping
File size: 6,865 Bytes
622a0b7 |
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 |
# models/nli_classifier.py
from transformers import pipeline
import torch
from collections import Counter
class NLIClassifier:
_instance = None
_initialized = False
def __new__(cls):
if cls._instance is None:
cls._instance = super().__new__(cls)
return cls._instance
def __init__(self):
if self._initialized:
return
try:
print("Loading NLI models (this may take a moment)...")
device = 0 if torch.cuda.is_available() else -1
self.models = []
# Model 1: RoBERTa-large-MNLI (most accurate)
try:
self.models.append({
'name': 'roberta-large-mnli',
'pipeline': pipeline(
"text-classification",
model="roberta-large-mnli",
device=device
),
'weight': 0.5
})
print("✓ Loaded RoBERTa-large-MNLI")
except Exception as e:
print(f"⚠ Failed to load RoBERTa-large-MNLI: {e}")
# Model 2: DeBERTa-v3-large MNLI fine-tuned (use pre-trained version)
try:
self.models.append({
'name': 'deberta-v3-large-mnli',
'pipeline': pipeline(
"text-classification",
model="MoritzLaurer/DeBERTa-v3-large-mnli-fever-anli-ling-wanli",
device=device
),
'weight': 0.5
})
print("✓ Loaded DeBERTa-v3-large-MNLI")
except Exception as e:
print(f"⚠ Failed to load DeBERTa-v3-large-MNLI: {e}")
if not self.models:
# Fallback to BART if both fail
try:
self.models.append({
'name': 'bart-large-mnli',
'pipeline': pipeline(
"zero-shot-classification",
model="facebook/bart-large-mnli",
device=device
),
'weight': 1.0
})
print("✓ Loaded BART-large-MNLI (fallback)")
except Exception as e:
print(f"✗ Failed to load any NLI model: {e}")
raise Exception("No NLI models loaded successfully")
# Normalize weights
total_weight = sum(m['weight'] for m in self.models)
for model in self.models:
model['weight'] /= total_weight
self._initialized = True
print(f"✓ Successfully loaded {len(self.models)} NLI model(s)")
except Exception as e:
print(f"Error loading NLI models: {e}")
self.models = []
self._initialized = False
def classify(self, claim, evidence):
"""Classify relationship between claim and evidence using ensemble"""
if not self.models:
return {
'label': 'NEUTRAL',
'confidence': 0.5,
'model_votes': {}
}
try:
results = []
model_votes = {}
for model_info in self.models:
try:
pipeline_obj = model_info['pipeline']
model_name = model_info['name']
# Handle different pipeline types
if 'bart' in model_name:
result = pipeline_obj(
evidence,
candidate_labels=["entailment", "contradiction", "neutral"],
hypothesis_template="This example is {}."
)
label = result['labels'][0]
confidence = result['scores'][0]
else:
# Standard NLI: premise [SEP] hypothesis
input_text = f"{evidence} [SEP] {claim}"
result = pipeline_obj(input_text)[0]
label = result['label']
confidence = result['score']
# Map labels
label_mapping = {
'ENTAILMENT': 'ENTAILMENT',
'CONTRADICTION': 'CONTRADICTION',
'NEUTRAL': 'NEUTRAL',
'entailment': 'ENTAILMENT',
'contradiction': 'CONTRADICTION',
'neutral': 'NEUTRAL',
'LABEL_0': 'CONTRADICTION',
'LABEL_1': 'NEUTRAL',
'LABEL_2': 'ENTAILMENT'
}
mapped_label = label_mapping.get(label, 'NEUTRAL')
results.append({
'label': mapped_label,
'confidence': confidence,
'weight': model_info['weight']
})
model_votes[model_name] = mapped_label
except Exception as e:
print(f"Error with model {model_info['name']}: {e}")
continue
if not results:
return {
'label': 'NEUTRAL',
'confidence': 0.5,
'model_votes': {}
}
# Weighted voting
weighted_scores = {
'ENTAILMENT': 0.0,
'CONTRADICTION': 0.0,
'NEUTRAL': 0.0
}
for result in results:
weighted_scores[result['label']] += result['confidence'] * result['weight']
# Get final label and confidence
final_label = max(weighted_scores, key=weighted_scores.get)
total_score = sum(weighted_scores.values())
final_confidence = weighted_scores[final_label] / total_score if total_score > 0 else 0.5
return {
'label': final_label,
'confidence': final_confidence,
'model_votes': model_votes,
'weighted_scores': weighted_scores
}
except Exception as e:
print(f"NLI classification error: {e}")
return {
'label': 'NEUTRAL',
'confidence': 0.5,
'model_votes': {}
}
|