Create inference.py

inference.py

@@ -0,0 +1,21 @@
+import torch
+from init_model import get_tokenizer, get_base_model
+from model import EvoTransformer
+
+tokenizer = get_tokenizer()
+bert = get_base_model()
+model = EvoTransformer()
+model.eval()
+
+def evo_suggest(question, option1, option2):
+    inputs = [question + " " + option1, question + " " + option2]
+    scores = []
+
+    for text in inputs:
+        encoded = tokenizer(text, return_tensors="pt", padding=True, truncation=True, max_length=128)
+        with torch.no_grad():
+            outputs = bert(**encoded).last_hidden_state[:, 0, :]  # [CLS] token
+            logits = model(outputs)
+            scores.append(logits[0][1].item())  # Confidence for class 1
+
+    return option1 if scores[0] > scores[1] else option2