初始化项目，由ModelHub XC社区提供模型

Model: iic/ERank-4B
Source: Original Platform

examples/utils.py

@@ -0,0 +1,44 @@
+import numpy as np
+
+prompt_template = """Given a query and a document, please give a relevance score of 0~10.
+The goal or relevance definition is: {instruction}
+
+Here is the query:
+{query}
+
+Here is the document:
+{doc}
+
+After thinking, directly choose a relevance score from [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10].
+- 0 represents completely not related
+- 10 means perfectly related.
+
+Desired output format:
+<think>put your thinking here</think><answer>Only allows an integer here</answer>
+
+Your output:"""
+
+
+def truncate(tokenizer, text, length):
+    if length == None or text == None:
+        return text
+    return tokenizer.convert_tokens_to_string(tokenizer.tokenize(text)[:length])
+
+
+def hybrid_scores(results, alpha):
+    first_stage_scores = [each["first_stage_score"] for each in results]
+    rank_scores = [each["rank_score"] for each in results]
+    first_stage_mean, first_stage_std = np.mean(first_stage_scores), np.std(first_stage_scores)
+    rank_mean, rank_std = np.mean(rank_scores), np.std(rank_scores)
+    
+    hybrid_results = []
+    for result in results:
+        normalized_first_stage_score = (result["first_stage_score"] - first_stage_mean) / first_stage_std
+        normalized_rank_score = (result["rank_score"] - rank_mean) / rank_std
+        hybrid_results.append({
+            **result,
+            "hybrid_score": float(alpha * normalized_first_stage_score + (1-alpha) * normalized_rank_score)
+        }) 
+    hybrid_results.sort(key=lambda x:x['hybrid_score'], reverse=True)
+
+    return hybrid_results