Switch to Groq LLM for faster inference, simplify system prompt

agent.py

@@ -25,11 +25,8 @@ from langchain_community.tools import DuckDuckGoSearchRun
 from langchain_community.document_loaders import WikipediaLoader
 from langchain_community.document_loaders import ArxivLoader
 from langgraph.prebuilt import ToolNode, tools_condition
-from langchain_huggingface import (
-    ChatHuggingFace,
-    HuggingFaceEndpoint,
-    HuggingFaceEmbeddings,
-)
+from langchain_huggingface import HuggingFaceEmbeddings
+from langchain_groq import ChatGroq
 from langchain_community.vectorstores import SupabaseVectorStore
 from langchain_core.messages import SystemMessage, HumanMessage
 from langchain_core.tools import tool, Tool
@@ -649,21 +646,12 @@ tools = [
 def build_graph():
     """Build the graph"""
     
-    # Use HuggingFaceEndpoint with Qwen
-    hf_token = os.environ.get("HF_TOKEN")
-    
-    # Configure LLM
-    llm = ChatHuggingFace(
-        llm=HuggingFaceEndpoint(
-            repo_id="Qwen/Qwen2.5-Coder-32B-Instruct",
-            task="text-generation",
-            max_new_tokens=4096,
-            do_sample=False,
-            repetition_penalty=1.03,
-            temperature=0.01, # Low temperature for factual tasks
-            huggingfacehub_api_token=hf_token
-        ),
-        verbose=True,
+    # Use Groq (fast, reliable, free tier)
+    # Model: qwen/qwen3-32b (same as reference implementation)
+    llm = ChatGroq(
+        model="qwen/qwen3-32b",
+        temperature=0,
+        api_key=os.environ.get("GROQ_API_KEY")
     )
 
     # Bind tools to LLM

main.py

@@ -45,26 +45,32 @@ def download_file(task_id):
     return None
 
 def clean_answer(raw_response):
-    """Strip 'Final Answer:' and other conversational fluff."""
+    """Strip everything except the final answer."""
     if not raw_response:
         return ""
     clean = str(raw_response)
     
     # 1. Look for explicit "FINAL ANSWER:" marker (case insensitive)
-    markers = ["FINAL ANSWER:", "Final Answer:", "Final answer:"]
-    for marker in markers:
-        if marker in clean:
-            # Take everything AFTER the last occurrence of the marker
-            clean = clean.split(marker)[-1].strip()
-            return clean
+    import re
+    # Match "FINAL ANSWER:" with any casing
+    match = re.search(r'FINAL\s*ANSWER\s*:\s*(.+)', clean, re.IGNORECASE | re.DOTALL)
+    if match:
+        answer = match.group(1).strip()
+        # Remove trailing punctuation that might be added
+        answer = answer.rstrip('.')
+        # If multiline, take just the first line (the actual answer)
+        if '\n' in answer:
+            answer = answer.split('\n')[0].strip()
+        return answer
             
-    # 2. If no marker, fallback to standard cleanup
-    for prefix in ["Answer:", "Assistant:"]:
-        if clean.startswith(prefix):
-            clean = clean[len(prefix):]
+    # 2. If no marker, the model didn't follow instructions - return empty or last line
+    # This signals an error to the evaluation
+    lines = [l.strip() for l in clean.strip().split('\n') if l.strip()]
+    if lines:
+        # Return the last non-empty line as a fallback
+        return lines[-1]
     
-    clean = clean.strip()
-    return clean
+    return clean.strip()
 
 def run_agent_on_task(prompt):
     """Run the graph agent on a single prompt."""

system_prompt.txt

@@ -1,7 +1,5 @@
 You are a helpful assistant tasked with answering questions using a set of tools. 
-First, ALWAYS use the `search_known_qa` tool to see if the question has already been answered. If it returns a good match, use that answer.
-If not, use your other tools to solve the problem.
-Report your thoughts, and finish your answer with the following template: 
+Now, I will ask you a question. Report your thoughts, and finish your answer with the following template: 
 FINAL ANSWER: [YOUR FINAL ANSWER]. 
 YOUR FINAL ANSWER should be a number OR as few words as possible OR a comma separated list of numbers and/or strings. If you are asked for a number, don't use comma to write your number neither use units such as $ or percent sign unless specified otherwise. If you are asked for a string, don't use articles, neither abbreviations (e.g. for cities), and write the digits in plain text unless specified otherwise. If you are asked for a comma separated list, Apply the rules above for each element (number or string), ensure there is exactly one space after each comma.
 Your answer should only start with "FINAL ANSWER: ", then follows with the answer.