adding syntheziser

app.py

@@ -12,9 +12,9 @@ DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
 
 # --- Basic Agent Definition ---
 # # ----- THIS IS WERE YOU CAN BUILD WHAT YOU WANT ------
-# class BasicAgent:
+# class GaiaAgent:
 #     def __init__(self):
-#         print("BasicAgent initialized.")
+#         print("GaiaAgent initialized.")
 #     def __call__(self, question: str) -> str:
 #         print(f"Agent received question (first 50 chars): {question[:50]}...")
 #         fixed_answer = "This is a default answer."
@@ -22,7 +22,7 @@ DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
 #         return fixed_answer
 
 
-class BasicAgent:
+class GaiaAgent:
     def __init__(self):
         print("Graph-based agent initialized.")
 
@@ -38,7 +38,7 @@ class BasicAgent:
 
 def run_and_submit_all( profile: gr.OAuthProfile | None):
     """
-    Fetches all questions, runs the BasicAgent on them, submits all answers,
+    Fetches all questions, runs the GaiaAgent on them, submits all answers,
     and displays the results.
     """
     # --- Determine HF Space Runtime URL and Repo URL ---
@@ -57,7 +57,7 @@ def run_and_submit_all( profile: gr.OAuthProfile | None):
 
     # 1. Instantiate Agent ( modify this part to create your agent)
     try:
-        agent = BasicAgent()
+        agent = GaiaAgent()
     except Exception as e:
         print(f"Error instantiating agent: {e}")
         return f"Error initializing agent: {e}", None
@@ -91,9 +91,9 @@ def run_and_submit_all( profile: gr.OAuthProfile | None):
     answers_payload = []
     print(f"Running agent on {len(questions_data)} questions...")
 
-    MAX_QUESTIONS = 3
-    questions_data = questions_data[:MAX_QUESTIONS]
-    print(f"Limiting to first {MAX_QUESTIONS} questions.")
+    # MAX_QUESTIONS = 5
+    # questions_data = questions_data[:MAX_QUESTIONS]
+    # print(f"Limiting to first {MAX_QUESTIONS} questions.")
 
     for item in questions_data:
         task_id = item.get("task_id")

qa_graph.py

@@ -1,43 +1,71 @@
-# simple_qa_graph.py
+# qa_graph.py
 
 from typing import TypedDict
+import re
 from langgraph.graph import StateGraph, START, END
 from tools.calculator_tool import calculator_tool
 from tools.search_tool import search_tool
-import re
+from transformers import pipeline
 
 # 1) Define the shape of our state
 class QAState(TypedDict):
     question: str   # incoming question
-    answer: str     # place to store the tool’s output
+    answer: str     # to store tool output or synthesized answer
+
+# 2) Use text2text-generation for T5 models like flan-t5
+synthesizer = pipeline(
+    "text2text-generation",
+    model="google/flan-t5-small",
+    device=-1,               # CPU; change to 0 for GPU
+    max_new_tokens=100,
+    do_sample=True,
+    top_p=0.95,
+    temperature=0.7
+)
 
-# 2) Define our single agent node
+# 3) Tool agent: calculator for math, search for other
 def QAAgent(state: QAState) -> QAState:
     q = state["question"].strip()
-    # if it looks like math, use the calculator, else do web search:
     if re.fullmatch(r"[0-9\s\+\-\*\/\.\(\)]+", q):
-        state["answer"] = calculator_tool(q)
+        state["answer"] = calculator_tool.invoke(q)
     else:
-        state["answer"] = search_tool(q)
+        state["answer"] = search_tool.invoke(q)  # update to `.invoke(q)` only if search_tool is a LangChain tool
     return state
 
-# 3) Wire it up in a graph: START → QAAgent → END
+# 4) Synthesis agent to generate final response
+def SynthesisAgent(state: QAState) -> QAState:
+    question = state["question"]
+    tool_out = state["answer"]
+    prompt = (
+        f"Question: {question}\n"
+        f"Tool output: {tool_out}\n"
+        "Answer in a comma-separated list (no extra text):"
+    )
+    outputs = synthesizer(prompt)
+    completion = outputs[0]["generated_text"].strip()
+    state["answer"] = completion
+    return state
+
+# 5) Build the graph: START -> QAAgent -> SynthesisAgent -> END
 builder = StateGraph(QAState)
 builder.set_entry_point("QAAgent")
 builder.add_node("QAAgent", QAAgent)
+builder.add_node("SynthesisAgent", SynthesisAgent)
+
 builder.add_edge(START, "QAAgent")
-builder.add_edge("QAAgent", END)
+builder.add_edge("QAAgent", "SynthesisAgent")
+builder.add_edge("SynthesisAgent", END)
 
 graph = builder.compile()
 
-# 4) Run it locally
+# 6) Local testing
 if __name__ == "__main__":
-    # try a math question
+    # Math example
     s1: QAState = {"question": "2 + 2", "answer": ""}
-    out1 = graph.invoke(s1)
-    print("Q:", s1["question"], "→ A:", out1["answer"])
+    o1 = graph.invoke(s1)
+    print("Q:", s1["question"], "-> A:", o1["answer"])
 
-    # try a search question
+    # Search + synthesis example
     s2: QAState = {"question": "What is the capital of France?", "answer": ""}
-    out2 = graph.invoke(s2)
-    print("Q:", s2["question"], "→ A:", out2["answer"])
+    o2 = graph.invoke(s2)
+    print("Q:", s2["question"], "-> A:", o2["answer"])

requirements.txt

@@ -10,3 +10,5 @@ huggingface_hub
 transformers
 langchain-huggingface
 IPython
+numpy<2.0
+

test_gaia.py

@@ -3,6 +3,6 @@ import requests
 import pandas as pd
 
 QUESTIONS = requests.get("https://agents-course-unit4-scoring.hf.space/questions").json()
-for q in QUESTIONS[:3]:
-    ans = graph.invoke({"question":q["question"],"answer":""})["answer"]
-    print(q["task_id"], ans)
+for q in QUESTIONS[:5]:
+    out = graph.invoke({"question": q["question"], "answer": ""})
+    print(q["task_id"], out["answer"])