real-jiakai
commited on
Commit
•
cc4e08c
1
Parent(s):
cf30567
Update README.md
Browse files
README.md
CHANGED
@@ -71,6 +71,30 @@ model_path = "real-jiakai/roberta-base-uncased-finetuned-swag"
|
|
71 |
tokenizer = AutoTokenizer.from_pretrained(model_path)
|
72 |
model = AutoModelForMultipleChoice.from_pretrained(model_path)
|
73 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
74 |
# Example scenarios
|
75 |
test_examples = [
|
76 |
{
|
@@ -93,29 +117,22 @@ test_examples = [
|
|
93 |
}
|
94 |
]
|
95 |
|
96 |
-
|
97 |
-
|
98 |
-
[context]
|
99 |
-
endings,
|
100 |
-
|
101 |
-
|
102 |
-
padding="max_length",
|
103 |
-
return_tensors="pt"
|
104 |
)
|
105 |
|
106 |
-
|
107 |
-
|
108 |
-
|
109 |
-
|
110 |
-
|
111 |
-
|
112 |
-
|
113 |
-
|
114 |
-
return {
|
115 |
-
'context': context,
|
116 |
-
'predicted_ending': endings[predicted_idx],
|
117 |
-
'probabilities': torch.softmax(logits, dim=1)[0].tolist()
|
118 |
-
}
|
119 |
```
|
120 |
|
121 |
## Limitations and Biases
|
|
|
71 |
tokenizer = AutoTokenizer.from_pretrained(model_path)
|
72 |
model = AutoModelForMultipleChoice.from_pretrained(model_path)
|
73 |
|
74 |
+
def predict_swag(context, endings, model, tokenizer):
|
75 |
+
encoding = tokenizer(
|
76 |
+
[context] * 4,
|
77 |
+
endings,
|
78 |
+
truncation=True,
|
79 |
+
max_length=128,
|
80 |
+
padding="max_length",
|
81 |
+
return_tensors="pt"
|
82 |
+
)
|
83 |
+
|
84 |
+
input_ids = encoding['input_ids'].unsqueeze(0)
|
85 |
+
attention_mask = encoding['attention_mask'].unsqueeze(0)
|
86 |
+
|
87 |
+
outputs = model(input_ids=input_ids, attention_mask=attention_mask)
|
88 |
+
logits = outputs.logits
|
89 |
+
|
90 |
+
predicted_idx = torch.argmax(logits).item()
|
91 |
+
|
92 |
+
return {
|
93 |
+
'context': context,
|
94 |
+
'predicted_ending': endings[predicted_idx],
|
95 |
+
'probabilities': torch.softmax(logits, dim=1)[0].tolist()
|
96 |
+
}
|
97 |
+
|
98 |
# Example scenarios
|
99 |
test_examples = [
|
100 |
{
|
|
|
117 |
}
|
118 |
]
|
119 |
|
120 |
+
for i, example in enumerate(test_examples, 1):
|
121 |
+
result = predict_swag(
|
122 |
+
example['context'],
|
123 |
+
example['endings'],
|
124 |
+
model,
|
125 |
+
tokenizer
|
|
|
|
|
126 |
)
|
127 |
|
128 |
+
print(f"\n=== Test Scenario {i} ===")
|
129 |
+
print(f"Initial Context: {result['context']}")
|
130 |
+
print(f"\nPredicted Most Likely Ending: {result['predicted_ending']}")
|
131 |
+
print("\nProbabilities for All Options:")
|
132 |
+
for idx, (ending, prob) in enumerate(zip(result['all_endings'], result['probabilities'])):
|
133 |
+
print(f"Option {idx}: {ending}")
|
134 |
+
print(f"Probability: {prob:.3f}")
|
135 |
+
print("\n" + "="*50)
|
|
|
|
|
|
|
|
|
|
|
136 |
```
|
137 |
|
138 |
## Limitations and Biases
|