Fix Ver2

app.py

@@ -5,18 +5,17 @@ import json
 import re
 import numexpr
 import pandas as pd
-import time
-import torch
 import math
 import pdfminer
-from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
 from duckduckgo_search import DDGS
 from pdfminer.high_level import extract_text
 from bs4 import BeautifulSoup
 import html2text
 from typing import Dict, Any, List, Tuple, Callable
 from dotenv import load_dotenv
-
+from transformers import AutoModelForCausalLM, AutoTokenizer
+import torch
+import time
 # --- Load Environment Variables ---
 load_dotenv()
 SERPER_API_KEY = os.getenv("SERPER_API_KEY")
@@ -32,30 +31,23 @@ os.environ["PIP_BREAK_SYSTEM_PACKAGES"] = "1"
 os.environ["HF_HUB_DISABLE_SYMLINKS_WARNING"] = "1"
 os.environ["BITSANDBYTES_NOWELCOME"] = "1"
 
-# --- Load Quantized Model ---
-print("Loading quantized model...")
-start_time = time.time()
 
-# Configure 4-bit quantization
-quant_config = BitsAndBytesConfig(
-    load_in_4bit=True,
-    bnb_4bit_quant_type="nf4",
-    bnb_4bit_use_double_quant=True,
-    bnb_4bit_compute_dtype=torch.bfloat16
-)
+MODEL_NAME = "microsoft/Phi-3-mini-4k-instruct"
+
+print("Loading model (CPU-compatible)...")
+start_time = time.time()
 
-# Load model and tokenizer
 model = AutoModelForCausalLM.from_pretrained(
     MODEL_NAME,
-    device_map="auto",
-    quantization_config=quant_config,
-    trust_remote_code=True
+    trust_remote_code=True,
+    torch_dtype=torch.float32  # Use float32 for CPU compatibility
 )
 tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
 
 load_time = time.time() - start_time
 print(f"Model loaded in {load_time:.2f} seconds")
 
+
 # --- Tools for GAIA Agent ---
 def web_search(query: str) -> str:
     """Search the web using DuckDuckGo or Serper API"""

requirements.txt

@@ -17,7 +17,7 @@ regex>=2023.10.3
 numexpr
 torch
 pdfminer.six
-transformers==4.39.3
+transformers>=4.0.0
 duckduckgo-search>=0.8
 beautifulsoup4>=4.12.0
 html2text>=2020.1.16