Update app.py

app.py

@@ -1,10 +1,13 @@
 import os
+import mimetypes
 import io
 import base64
 import time
+
 import requests
 import pandas as pd
 import gradio as gr
+import pypdf
 
 from typing import TypedDict, Annotated
 from langgraph.graph import StateGraph
@@ -20,9 +23,11 @@ from groq import Groq
 
 DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
 
-GROQ_TEXT_MODEL = "llama-3.3-70b-versatile" # для текстовых запросов
-GROQ_VISION_MODEL = "meta-llama/llama-4-scout-17b-16e-instruct" # для картинок
-GROQ_AUDIO_MODEL = "whisper-large-v3-turbo" # для аудио
+GROQ_TEXT_MODEL   = os.getenv("GROQ_TEXT_MODEL",   "llama-3.3-70b-versatile")
+GROQ_VISION_MODEL = os.getenv("GROQ_VISION_MODEL", "meta-llama/llama-4-scout-17b-16e-instruct")
+GROQ_AUDIO_MODEL  = os.getenv("GROQ_AUDIO_MODEL",  "whisper-large-v3-turbo")
+
+GAIA_DIR = "./data/gaia"
 
 def get_groq_client() -> Groq:
     key = os.getenv("GROQ_API_KEY")
@@ -30,15 +35,35 @@ def get_groq_client() -> Groq:
         raise ValueError("GROQ_API_KEY secret not set!")
     return Groq(api_key=key)
 
+def _get_task_file_map() -> dict[str, str]:
+    result = {}
+    validation_dir = os.path.join(GAIA_DIR, "2023", "validation")
+    for file_name in os.listdir(validation_dir):
+        if file_name.endswith(".parquet"):
+            continue
+        task_id = os.path.splitext(file_name)[0]
+        result[task_id] = os.path.join(validation_dir, file_name)
+    return result
+
+def get_task_file(task_id: str) -> str | None:
+    return _get_task_file_map().get(task_id)
 
 def _fetch_task_bytes(task_id: str) -> tuple[bytes, str]:
-    """Returns (raw_bytes, content_type)."""
-    url = f"{DEFAULT_API_URL}/files/{task_id}"
-    resp = requests.get(url, timeout=20)
-    resp.raise_for_status()
-    content_type = resp.headers.get("Content-Type", "").lower()
-    return resp.content, content_type
-
+    local_path = get_task_file(task_id)
+    if not local_path:
+        raise FileNotFoundError(f"No local file mapped for task_id: {task_id}")
+    
+    if not os.path.exists(local_path):
+        raise FileNotFoundError(f"File not found at path: {local_path}")
+        
+    with open(local_path, "rb") as f:
+        data = f.read()
+        
+    content_type, _ = mimetypes.guess_type(local_path)
+    if not content_type:
+        content_type = "application/octet-stream"
+        
+    return data, content_type
 
 def _is_image(ct: str, data: bytes) -> bool:
     image_types = ("image/", "png", "jpeg", "jpg", "gif", "webp")
@@ -147,24 +172,41 @@ def transcribe_audio(task_id: str) -> str:
 @tool
 def read_text_file(task_id: str) -> str:
     """
-    Download a text-based file (txt, csv, json, py, xlsx summary, etc.)
-    for a GAIA task and return its content (up to 4000 characters).
+    Read a text-based, spreadsheet (Excel), or PDF file 
+    for a GAIA task and return its content or summary (up to 4000 characters).
     """
     try:
+        local_path = get_task_file(task_id)
+        if not local_path:
+            return f"No file attached for task {task_id}."
+        
+        if local_path.endswith((".xlsx", ".xls")):
+            df = pd.read_excel(local_path)
+            summary = f"Excel file columns: {list(df.columns)}\nShape: {df.shape}\n\n"
+            summary += df.head(15).to_string()
+            return summary[:4000]
+            
+        elif local_path.endswith(".pdf"):
+            try:
+                reader = pypdf.PdfReader(local_path)
+                text = ""
+                for page in reader.pages[:5]:
+                    text += page.extract_text() + "\n"
+                return text[:4000]
+            except ImportError:
+                return "PDF file found, but 'pypdf' library is not installed in the environment."
+            
         data, ct = _fetch_task_bytes(task_id)
-    except Exception as e:
-        return f"Could not fetch file for task {task_id}: {e}"
-
-    if _is_image(ct, data) or _is_audio(ct, data):
-        return (
-            f"File for task {task_id} is binary ({ct}). "
-            "Use analyze_image or transcribe_audio instead."
-        )
-
-    try:
+        if _is_image(ct, data) or _is_audio(ct, data):
+            return (
+                f"File for task {task_id} is binary image/audio. "
+                "Use analyze_image or transcribe_audio instead."
+            )
+            
         return data.decode("utf-8", errors="replace")[:4000]
+        
     except Exception as e:
-        return f"Text decode error: {e}"
+        return f"Error reading file for task {task_id}: {e}"
 
 
 TOOLS = [web_search_tool, wikipedia_tool, analyze_image, transcribe_audio, read_text_file]