Update app.py

app.py

@@ -4,14 +4,14 @@ import gradio as gr
 import requests
 import pandas as pd
 
-from smolagents import InferenceClientModel, ToolCallingAgent
+from transformers import BartForConditionalGeneration, BartTokenizer
+from smolagents import ToolCallingAgent
 from audio_transcriber import AudioTranscriptionTool
 from image_analyzer import ImageAnalysisTool
 from wikipedia_searcher import WikipediaSearcher
 
 DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
 
-# Zephyr-compatible system prompt to prepend manually
 SYSTEM_PROMPT = (
     "You are an agent solving the GAIA benchmark and must provide exact answers.\n"
     "Rules:\n"
@@ -30,14 +30,30 @@ SYSTEM_PROMPT = (
     "Never say 'the answer is...'. Only return the answer.\n"
 )
 
+class LocalBartModel:
+    def __init__(self, model_name="facebook/bart-base", device=None):
+        import torch
+        self.device = device if device else ("cuda" if torch.cuda.is_available() else "cpu")
+        self.tokenizer = BartTokenizer.from_pretrained(model_name)
+        self.model = BartForConditionalGeneration.from_pretrained(model_name).to(self.device)
+
+    def __call__(self, prompt: str) -> str:
+        import torch
+
+        inputs = self.tokenizer(prompt, return_tensors="pt", truncation=True, max_length=512).to(self.device)
+        outputs = self.model.generate(
+            **inputs,
+            max_length=128,
+            num_beams=5,
+            early_stopping=True
+        )
+        decoded = self.tokenizer.decode(outputs[0], skip_special_tokens=True)
+        return decoded.strip()
+
 class GaiaAgent:
     def __init__(self):
         print("Gaia Agent Initialized")
-
-        self.model = InferenceClientModel(
-            model_id="HuggingFaceH4/zephyr-7b-beta",
-            token=os.getenv("HF_API_TOKEN", "").strip()
-        )
+        self.model = LocalBartModel()
 
         self.tools = [
             AudioTranscriptionTool(),
@@ -164,7 +180,6 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
     except Exception as e:
         return f"An unexpected error occurred during submission: {e}", pd.DataFrame(results_log)
 
-# Gradio UI
 with gr.Blocks() as demo:
     gr.Markdown("# Basic Agent Evaluation Runner")
     gr.Markdown("""