upgraded highlighting, added bitnet

utils/models.py

@@ -18,6 +18,7 @@ models = {
     "Phi-4-mini-instruct": "microsoft/phi-4-mini-instruct",
     #"Cogito-v1-preview-llama-3b": "deepcogito/cogito-v1-preview-llama-3b",
     "IBM Granite-3.3-2b-instruct": "ibm-granite/granite-3.3-2b-instruct",
+    "Bitnet-b1.58-2B4T": "microsoft/bitnet-b1.58-2B-4T"
 
 }
 

utils/prompts.py

@@ -1,34 +1,33 @@
 def format_rag_prompt( query: str, context: str, accepts_sys: bool) -> str:
     system_prompt = """
-    You are a helpful assistant that provides answers to queries based on the provided context.
+You are a helpful assistant that provides answers to queries based on the provided context.
 
-    If the full, complete answer to the query cannot be found in the context, answer what the context allows you to answer and indicate clearly where additional information is needed.
-    If the none of the answer can be found, clearly refuse to answer, and ask for more relevant information from the user.
-
-    The output should not contain your judgment on answerability, only your answer OR your refusal + clarifications.
-
-    Stay within the bounds of the provided context and avoid making assumptions.
+If the full, complete answer to the query cannot be found in the context, answer what the context allows you to answer and indicate clearly where additional information is needed.
+If the none of the answer can be found, clearly refuse to answer, and ask for more relevant information from the user.
 
+The output should not contain your judgment on answerability, only your answer OR your refusal + clarifications.
 
+Stay within the bounds of the provided context and avoid making assumptions.
     """
-    user_prompt = f"""
+    user_prompt = f"""# Role and Task Description
+Judge if the following query is answerable from ONLY the provided context.
+If so, provide a complete, grounded answer to the query, and do not mention your judgement.
+Try to address all aspects of the query, but if certain parts are not answerable, answer what you can and indicate clearly where additional information is needed.
 
-    # Role and Task Description
-    Judge if the following query is answerable from ONLY the provided context.
-    If so, provide a complete, grounded answer to the query, and do not mention your judgement.
-    Try to address all aspects of the query, but if certain parts are not answerable, answer what you can and indicate clearly where additional information is needed.
+If none of the query's answer can be found in the context, clearly refuse to answer, and ask for more relevant information from the user.
+You should give a concise explanation of why you cannot answer the query based on the context, and ask for more relevant information from the user.
 
-    If none of the query's answer can be found in the context, clearly refuse to answer, and ask for more relevant information from the user.
-    You should give a concise explanation of why you cannot answer the query based on the context, and ask for more relevant information from the user.
+# Task
+Given the following query and context, please provide your response:
+## Query: 
+{query}
 
-    # Task
-    Given the following query and context, please provide your response:
-    Query: {query}
+## Context: 
 
-    Context: {context}
+{context}
 
-    WITHOUT mentioning your judgement either your grounded answer, OR refusal and clarifications:
-    """
+WITHOUT mentioning your judgement either your grounded answer, OR refusal and clarifications:
+"""
 
     messages = (
         [

utils/vote_logger.py

@@ -18,7 +18,8 @@ hub_votes = hf_hub_download(
     repo_id="aizip-dev/Arena-Metadata",
     filename="vote_details.csv",
     repo_type="dataset",
-    revision="main",)
+    revision="main",
+    )
 
 df = pd.read_csv(hub_votes)
 print(f"Successfully loaded leaderboard from the Hub. {len(df)} votes.")
@@ -66,6 +67,7 @@ def save_vote_details(example, model_a, model_b, winner, feedback, summary_a, su
                 
                 # Write the vote record
                 writer.writerow(vote_record)
+                
             
             print(f"Vote details saved to {csv_path}")
     except Exception as e: