Changed model and updated prompt

app.py

@@ -25,12 +25,15 @@ image_captioner = pipeline(
 
 
 @tool
-def describe_image(path: str) -> str:
-    """Describes the image. requires image path. return image description.
+def image_tool(directory: str = "/tmp") -> str:
+    """Describes the image. Requires the temporary directory file. Returns image description.
     Args:
-        path: Path to the image file to analyze.
+        directory: Temporary directory where .jpg file are stored as input
     """
-    return image_captioner(path)[0]["generated_text"]
+    files = os.listdir('/tmp')
+    image = [f for f in files if f.endswith('.jpg')]
+    audio_file_path = f"/tmp/{image}"
+    return image_captioner(audio_file_path)[0]["generated_text"]
 
 @tool    
 def audio_to_text_tool(directory: str = "/tmp") -> str:
@@ -39,7 +42,7 @@ def audio_to_text_tool(directory: str = "/tmp") -> str:
         directory: Temporary directory where .mp3 file are stored as input
     """
     files = os.listdir('/tmp')
-    audio_file = [f for f in files if f.endswith('.mp3')][0]
+    audio_file = [f for f in files if f.endswith('.mp3')]
     audio_file_path = f"/tmp/{audio_file}"
     output = SpeechToTextTool()
     return output(audio_file_path) 
@@ -51,7 +54,7 @@ def excel_reader(directory: str = "/tmp") -> str:
         directory: Temporary directory where .xlsx file are stored as input
     """
     files = os.listdir('/tmp')
-    excel_file = [f for f in files if f.endswith('.xlsx')][0]
+    excel_file = [f for f in files if f.endswith('.xlsx')]
     df = pd.read_excel(f"/tmp/{excel_file}")
     return f"Rows: {len(df)}, Columns: {list(df.columns)}, Description: {df.describe()}"
 
@@ -60,13 +63,13 @@ class BasicAgent:
         print("Srj's Agent initialized.")
         self.model = InferenceClientModel(max_tokens=2096,
             temperature=0.5,
-            model_id='meta-llama/Meta-Llama-3-8B-Instruct',
-            provider="novita"
+            model_id='google/gemma-3n-E4B-it',
+            provider="together"
                                 
         )       
         self.agent = CodeAgent(
             model=self.model,
-            tools=[final_answer, web_search,audio_to_text_tool,wikipedia_search,python_code_executor,excel_reader, describe_image], # add your tools here (don't remove final_answer)
+            tools=[final_answer, web_search,audio_to_text_tool,wikipedia_search,python_code_executor,excel_reader, image_tool], # add your tools here (don't remove final_answer)
             max_steps=7,
             verbosity_level=1,          
             planning_interval=None,
@@ -92,12 +95,14 @@ CRITICAL RULES (Follow EXACTLY):
 5. Spell out numbers in words when asked for text
 6. For cities: full name, no abbreviations (Los Angeles, not LA)
 
-Important: Do not hallucinate. Try to keep your answers based on facts as much as you can. You will be asked a question, make sure you only give the answer asked in the question. Do a web search first and then wikipediasearch.
-CRITICAL: 
+Important: Do not hallucinate. Try to keep your answers based on facts as much as you can. You will be asked a question, make sure you only give the answer asked in the question. Do a web search first and then wikipediasearch. You might need to do multiple searches to answer a question. For. eg. Question -> search -> clue -> search again with clue in consideration -> better clue -> result.
+
+When handling with files (Format: .xlxs, .mp3, .jpg) received in the question. You can consider the following steps:
 1. GAIA files are at /tmp/task_*/filename
 2. List files: import os; print(os.listdir('/tmp'))
 3. Use REAL paths from os.listdir()
 4. Paths start with /tmp/task_
+
 Question: {question}
 
 Respond with ONLY the result."""