Update app.py

app.py

@@ -34,7 +34,6 @@ def rate_limited_groq(api_key, prompt, system="", max_tokens=128):
     return resp.json()["choices"][0]["message"]["content"].strip()
 
 def download_task_file(task_id, hf_token=None):
-    """Download with HF OAuth token for authentication."""
     url = f"{DEFAULT_API_URL}/files/{task_id}"
     headers = {}
     if hf_token:
@@ -42,7 +41,7 @@ def download_task_file(task_id, hf_token=None):
     try:
         resp = requests.get(url, headers=headers, timeout=30)
         print(f"  File [{task_id[:8]}]: HTTP {resp.status_code}, "
-              f"size={len(resp.content)}, ct={resp.headers.get('content-type','?')[:40]}")
+              f"size={len(resp.content)}, ct={resp.headers.get('content-type','?')[:50]}")
         if resp.status_code != 200 or len(resp.content) == 0:
             return None, None
         cd = resp.headers.get("content-disposition", "")
@@ -62,7 +61,7 @@ def download_task_file(task_id, hf_token=None):
         tmp = tempfile.NamedTemporaryFile(delete=False, suffix=ext, prefix="gaia_")
         tmp.write(resp.content)
         tmp.close()
-        print(f"  Saved: {fname} ({len(resp.content)} bytes)")
+        print(f"  Saved: {fname} ({len(resp.content)} bytes) -> {tmp.name}")
         return tmp.name, fname
     except Exception as e:
         print(f"  Download error: {e}")
@@ -131,9 +130,9 @@ def test_api():
     except Exception as e:
         return f"❌ {e}"
 
-SYSTEM = """You are a GAIA benchmark agent. Exact match grading — your answer must match exactly.
+SYSTEM = """You are a GAIA benchmark agent. Exact match grading is used.
 Reply with ONLY the final answer. No explanation. No prefix. No "The answer is".
-Give the bare answer: a name, number, word, or short phrase only."""
+Give only: a name, number, word, or short phrase."""
 
 class BasicAgent:
     def __init__(self, hf_token=None):
@@ -141,7 +140,7 @@ class BasicAgent:
         if not self.key:
             raise RuntimeError("GROQ_API_KEY not set!")
         self.hf_token = hf_token
-        print(f"Agent ready. Groq key: {self.key[:8]}... | HF token: {'YES' if hf_token else 'NO'}")
+        print(f"Agent ready. Groq: {self.key[:8]}... | HF token: {'YES ✅' if hf_token else 'NO ❌'}")
 
     def ask(self, prompt, max_tokens=128):
         return clean_answer(rate_limited_groq(self.key, prompt, SYSTEM, max_tokens))
@@ -157,7 +156,7 @@ class BasicAgent:
         file_ctx = ""
         is_py = False
 
-        # Download file with HF token
+        # Download file using HF OAuth token
         if task_id:
             lp, fn = download_task_file(task_id, self.hf_token)
             if lp and fn:
@@ -183,7 +182,7 @@ class BasicAgent:
             if results and "error" not in results.lower():
                 search_ctx = f"\n[Search]\n{results[:3500]}\n"
 
-        # Format hints
+        # Format hints per question type
         q = question.lower()
         fmt = ""
         if "studio album" in q:
@@ -196,20 +195,20 @@ class BasicAgent:
             fmt = "\nSingle integer only."
         elif "how many" in q:
             fmt = "\nSingle integer only."
-        elif "ioc" in q or ("country" in q and "olympic" in q):
-            fmt = "\nIOC country code only (3 letters). If tied, alphabetically first country."
+        elif "ioc" in q:
+            fmt = "\nIOC country code only (3 letters e.g. USA). If tied on athlete count, alphabetically first."
         elif "excel" in q or ("sale" in q and "food" in q):
-            fmt = "\nNumber with exactly two decimal places. No $ sign, no commas (e.g. 89.50)."
+            fmt = "\nUSD with two decimal places, no $ sign, no commas (e.g. 8945.50)."
         elif "chess" in q:
-            fmt = "\nChess move in algebraic notation only (e.g. Qd8, e5)."
+            fmt = "\nChess move in algebraic notation only (e.g. Qd8)."
         elif "pitcher" in q and "number" in q:
-            fmt = "\nTwo last names, comma-separated. Pitcher with lower jersey number first."
+            fmt = "\nTwo last names comma-separated. Lower jersey number pitcher first."
         elif "wikipedia" in q and "nominat" in q:
             fmt = "\nWikipedia username only."
         elif "grocery" in q or ("shopping" in q and "list" in q):
             fmt = "\nComma-separated list, alphabetical order."
         elif "youtube" in q or "video" in q:
-            fmt = "\nExact short answer — quote, number, or brief phrase only."
+            fmt = "\nExact short answer only."
 
         prompt = (
             f"Question: {question}"
@@ -232,15 +231,20 @@ class BasicAgent:
             print(f"  Error: {e}")
             return ""
 
-def run_and_submit_all(profile: gr.OAuthProfile | None):
+def run_and_submit_all(profile: gr.OAuthProfile | None,
+                       oauth_token: gr.OAuthToken | None):
+    """
+    IMPORTANT: oauth_token gives us the actual HF bearer token
+    needed to download task files from the scoring API.
+    """
     space_id = os.getenv("SPACE_ID")
     if not profile:
         return "Please Login to Hugging Face.", None
 
     username = profile.username
-    # Get the HF OAuth token from the login profile
-    hf_token = getattr(profile, "token", None)
-    print(f"User: {username}, HF token present: {bool(hf_token)}")
+    # Extract the actual token string
+    hf_token = oauth_token.token if oauth_token else None
+    print(f"User: {username} | HF token present: {'YES ✅' if hf_token else 'NO ❌'}")
 
     try:
         agent = BasicAgent(hf_token=hf_token)
@@ -280,7 +284,8 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
 
     try:
         resp = requests.post(f"{DEFAULT_API_URL}/submit",
-            json={"username": username.strip(), "agent_code": agent_code, "answers": answers_payload},
+            json={"username": username.strip(), "agent_code": agent_code,
+                  "answers": answers_payload},
             timeout=60)
         resp.raise_for_status()
         r = resp.json()
@@ -292,7 +297,10 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
 
 with gr.Blocks() as demo:
     gr.Markdown("# Basic Agent Evaluation Runner")
-    gr.Markdown("**Setup:** `GROQ_API_KEY` in Space Settings → Secrets. Free at [console.groq.com](https://console.groq.com)")
+    gr.Markdown(
+        "**Setup:** Add `GROQ_API_KEY` in Space Settings → Secrets. "
+        "Free key at [console.groq.com](https://console.groq.com)"
+    )
     gr.LoginButton()
     with gr.Row():
         test_btn = gr.Button("🔬 Test Groq API", variant="secondary")
@@ -302,7 +310,12 @@ with gr.Blocks() as demo:
     run_button = gr.Button("🚀 Run Evaluation & Submit All Answers", variant="primary")
     status_output = gr.Textbox(label="Run Status / Submission Result", lines=5, interactive=False)
     results_table = gr.DataFrame(label="Questions and Agent Answers", wrap=True)
-    run_button.click(fn=run_and_submit_all, outputs=[status_output, results_table])
+    # Pass BOTH profile AND oauth_token so we can use the HF bearer token
+    run_button.click(
+        fn=run_and_submit_all,
+
+        outputs=[status_output, results_table]
+    )
 
 if __name__ == "__main__":
     key = os.getenv("GROQ_API_KEY", "")