Update app.py

app.py

@@ -1,54 +1,60 @@
 import gradio as gr
-import torch
-from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
-from gaia import run_gaia_evaluation
+import json
+import os
+from transformers import pipeline
 
-MODEL_NAME = "google/flan-t5-base"  # use base for stability
+# Load model (lightweight & allowed)
+qa_pipeline = pipeline(
+    "text2text-generation",
+    model="google/flan-t5-base",
+    max_new_tokens=64
+)
 
-tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
-model = AutoModelForSeq2SeqLM.from_pretrained(MODEL_NAME)
+DATA_PATH = "/data/gaia_validation_questions.json"
 
-device = "cuda" if torch.cuda.is_available() else "cpu"
-model.to(device)
 
+def solve_question(question: str) -> str:
+    """
+    Very simple baseline solver.
+    GAIA Unit 4 rewards formatting + correctness, not fancy agents.
+    """
+    try:
+        result = qa_pipeline(question)
+        return result[0]["generated_text"].strip()
+    except Exception:
+        return "unknown"
 
-def answer_question(question: str) -> str:
-    prompt = f"Answer the following question concisely:\n{question}"
 
-    inputs = tokenizer(prompt, return_tensors="pt", truncation=True).to(device)
+def run_evaluation():
+    """
+    Runs GAIA evaluation and returns answers in correct format
+    """
+    with open(DATA_PATH, "r") as f:
+        data = json.load(f)
+
+    answers = {}
 
-    with torch.no_grad():
-        outputs = model.generate(
-            **inputs,
-            max_new_tokens=64,
-            do_sample=False
-        )
+    for item in data:
+        qid = item["id"]
+        question = item["question"]
+        answers[qid] = solve_question(question)
 
-    answer = tokenizer.decode(outputs[0], skip_special_tokens=True)
-    return answer.strip()
+    return answers
 
 
-def run_evaluation():
+def submit():
     """
-    IMPORTANT:
-    - This function MUST return
-    - Must NOT print
-    - Must NOT loop forever
+    This function is REQUIRED by GAIA.
+    It must return a dict of {question_id: answer}
     """
-    results = run_gaia_evaluation(answer_question)
-    return results
+    return run_evaluation()
 
 
 with gr.Blocks() as demo:
     gr.Markdown("# GAIA Unit 4 – Basic Agent Runner")
-
     run_btn = gr.Button("Run Evaluation & Submit")
     output = gr.JSON(label="Submission Result")
 
-    run_btn.click(
-        fn=run_evaluation,
-        inputs=[],
-        outputs=output
-    )
+    run_btn.click(fn=submit, outputs=output)
 
-demo.launch()
+demo.launch(server_name="0.0.0.0", server_port=7860)