Update README.md

README.md

@@ -5,8 +5,45 @@ tags: []
 
 # Model Card for Model ID
 
-<!-- Provide a quick summary of what the model is/does. -->
-
+```
+rm = trainer.model
+rm.eval()
+
+def render_pair_for_scoring(source_text, revision_text):
+    messages = [
+        {
+            "role": "user",
+            "content": f"{USER_PREFIX}\n\n{str(source_text).strip()}",
+        },
+        {
+            "role": "assistant",
+            "content": f"{ASSISTANT_PREFIX}\n\n{str(revision_text).strip()}",
+        },
+    ]
+    return apply_qwen_template_clean(messages)
+
+def reward_score(source_text, revision_text):
+    text = render_pair_for_scoring(source_text, revision_text)
+    inputs = tokenizer(
+        text,
+        return_tensors="pt",
+        truncation=True,
+        max_length=MAX_LENGTH,
+    ).to(rm.device)
+
+    with torch.no_grad():
+        logits = rm(**inputs).logits
+
+    return logits.squeeze().float().item()
+
+```
+
+```
+row = split["test"][0]
+
+print("chosen score:  ", reward_score(row["prompt"], row["chosen"]))
+print("rejected score:", reward_score(row["prompt"], row["rejected"]))
+```
 
 
 ## Model Details