Refactor app.py to remove Python and Torch version logging; install flash-attn package at runtime. Update requirements.txt to remove flash-attn wheel link.

app.py

@@ -3,13 +3,11 @@ from PIL import Image
 from transformers import AutoTokenizer, AutoProcessor, AutoModelForImageTextToText
 import torch
 import spaces
-import sys
 
-model_path = "nanonets/Nanonets-OCR-s"
+import subprocess
+subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, shell=True)
 
-# Print python version, torch and cuda version
-print(f"Torch version: {torch.__version__}") 
-print(f"Python version: {sys.version}")
+model_path = "nanonets/Nanonets-OCR-s"
 
 # Load model once at startup
 print("Loading Nanonets OCR model...")

requirements.txt

@@ -1,5 +1,4 @@
 transformers
 torch
 accelerate
-https://github.com/mjun0812/flash-attention-prebuild-wheels/releases/download/v0.0.8/flash_attn-2.7.4.post1+cu126torch2.7-cp310-cp310-linux_x86_64.whl
 spaces