specify torch/huggingface version.

app.py

@@ -1,9 +1,12 @@
 import os
 import io
 #DEBUG
-os.environ["TORCH_USE_CUDA_DSA"] = "1"
+os.environ["CUDA_LAUNCH_BLOCKING"] = "1"
 
 import torch
+import torchvision
+import transformers
+import logging
 import json
 import base64
 import gradio as gr
@@ -15,13 +18,17 @@ from plots import get_pre_define_colors
 from utils.load_model import load_xclip
 from utils.predict import xclip_pred
 
-
+logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
 #! Huggingface does not allow load model to main process, so we need to load the model when needed, it may not help in improve the speed of the app.
 DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
-print(f"Not at Huggingface demo, load model to main process.")
+logging.info(f"Using device: {DEVICE}")
+# get the torch, torchvision, and transformers version
+logging.info(f"torch version: {torch.__version__}")
+logging.info(f"torchvision version: {torchvision.__version__}")
+logging.info(f"transformers version: {transformers.__version__}")
+
 XCLIP, OWLVIT_PRECESSOR = load_xclip(DEVICE)
 
-print(f"Device: {DEVICE}")
 
 XCLIP_DESC_PATH = "data/jsons/bs_cub_desc.json"
 XCLIP_DESC = json.load(open(XCLIP_DESC_PATH, "r"))

requirements.txt

@@ -1,9 +1,9 @@
-# torch
-# torchvision
+torch>=2.6.0
+torchvision>=0.21.0
 # gradio
 numpy
 Pillow
-transformers
+transformers==4.47.1
 ftfy
 regex
 pandas