Update app.py

app.py

@@ -1,69 +1,37 @@
-import os
+# app.py
 import gradio as gr
 import json
+import os
 import requests
-from agent import load_tasks, solve_task
-from smolagents import Task
-
-# Load metadata
-METADATA_PATH = "metadata.jsonl"
-GAIA_API_URL = "https://huggingface.co/spaces/gaia-benchmark/api/submit"
-
-# Optional auth for GAIA
-HF_USER = os.getenv("HF_USERNAME")
-HF_TOKEN = os.getenv("HF_TOKEN")
-
-def run_and_submit_all(submit: bool = False):
-    tasks = load_tasks(METADATA_PATH)
-    results = []
-    for task in tasks:
-        answer = solve_task(task)
-        results.append({
-            "question_id": task.task_id,
-            "answer": answer,
-        })
-
-    if submit:
-        if not HF_USER or not HF_TOKEN:
-            raise ValueError("Missing HF_USERNAME or HF_TOKEN in env vars.")
-        response = requests.post(
-            GAIA_API_URL,
-            headers={
-                "Authorization": f"Bearer {HF_TOKEN}",
-                "Content-Type": "application/json",
-            },
-            json={
-                "username": HF_USER,
-                "level": "level1",
-                "answers": results,
-            },
-        )
-        return f"Submitted! Response: {response.text}"
-
-    return json.dumps(results, indent=2)
 
-def single_task_run(question_id: str, question_input: str):
-    task = Task(task_id=question_id, input=question_input)
-    output = solve_task(task)
-    return output
+from mistral_hf_wrapper import MistralInference
+from agent import solve_task, load_tasks
 
-# Gradio UI
-with gr.Blocks() as demo:
-    gr.Markdown("# GAIA Benchmark Agent")
+API_URL = os.getenv("HF_MISTRAL_URL")
+API_TOKEN = os.getenv("HF_TOKEN")
+USERNAME = os.getenv("HF_USERNAME")
+CODE_LINK = os.getenv("HF_CODE_LINK")  # e.g., https://huggingface.co/spaces/yourname/yourspace/tree/main
 
-    with gr.Tab("Run Single Task"):
-        with gr.Row():
-            qid = gr.Textbox(label="Question ID", placeholder="e.g., q123")
-            qtext = gr.Textbox(label="Question Text", lines=4)
-        out = gr.Textbox(label="Generated Answer")
-        btn = gr.Button("Run Agent")
-        btn.click(fn=single_task_run, inputs=[qid, qtext], outputs=out)
+def run_and_submit_all():
+    model = MistralInference(api_url=API_URL, api_token=API_TOKEN)
+    tasks = load_tasks()
+    answers = []
 
-    with gr.Tab(" Run All + Submit"):
-        submit_toggle = gr.Checkbox(label="Submit to GAIA?", value=False)
-        submit_btn = gr.Button("Run All Tasks")
-        batch_out = gr.Textbox(label="Batch Output", lines=20)
-        submit_btn.click(fn=run_and_submit_all, inputs=[submit_toggle], outputs=batch_out)
-
-if __name__ == "__main__":
-    demo.launch()
+    for task in tasks:
+        try:
+            answer = solve_task(task, model)
+            answers.append({"task_id": task["task_id"], "submitted_answer": answer.strip()})
+        except Exception as e:
+            answers.append({"task_id": task["task_id"], "submitted_answer": f"ERROR: {str(e)}"})
+
+    res = requests.post(
+        "https://agents-course-unit4-scoring.hf.space/submit",
+        headers={"Content-Type": "application/json"},
+        json={"username": USERNAME, "agent_code": CODE_LINK, "answers": answers},
+    )
+
+    if res.ok:
+        return json.dumps(res.json(), indent=2)
+    return f"Error submitting: {res.status_code} - {res.text}"
+
+gr.Interface(run_and_submit_all, inputs=[], outputs="textbox", title="GAIA Benchmark Agent Submission").launch()