49 lines
1.6 KiB
Python
49 lines
1.6 KiB
Python
|
# main.py
|
||
|
|
import os
|
||
|
|
import sys
|
||
|
|
|
||
|
|
# 1. Block hardware-level multi-device discovery loops
|
||
|
|
os.environ["CUDA_VISIBLE_DEVICES"] = ""
|
||
|
|
|
||
|
|
import torch
|
||
|
|
|
||
|
|
# 2. Hard monkeypatch PyTorch's MPS detection before ANY machine learning imports occur
|
||
|
|
# This stops underlying huggingface models from forcing device="mps" internally
|
||
|
|
torch.backends.mps.is_available = lambda: False
|
||
|
|
torch.backends.mps.is_built = lambda: False
|
||
|
|
|
||
|
|
# Force standard CPU allocation for safety
|
||
|
|
torch.set_default_device("cpu")
|
||
|
|
|
||
|
|
import time
|
||
|
|
from docling.document_converter import DocumentConverter
|
||
|
|
|
||
|
|
def main():
|
||
|
|
source = "https://arxiv.org/pdf/2408.09869"
|
||
|
|
|
||
|
|
print("🚀 Initializing Docling engine with hardware-isolated CPU routing...")
|
||
|
|
start_time = time.time()
|
||
|
|
|
||
|
|
# This will now natively spin up on standard CPU threads
|
||
|
|
converter = DocumentConverter()
|
||
|
|
|
||
|
|
print("⏳ Processing document locally (Parsing layout, OCR, and tables)...")
|
||
|
|
result = converter.convert(source)
|
||
|
|
|
||
|
|
print("✨ Conversion finished! Compiling Markdown structure...")
|
||
|
|
markdown_output = result.document.export_to_markdown()
|
||
|
|
|
||
|
|
# Save output cleanly to project directory
|
||
|
|
output_filename = "parsed_paper.md"
|
||
|
|
with open(output_filename, "w", encoding="utf-8") as f:
|
||
|
|
f.write(markdown_output)
|
||
|
|
|
||
|
|
elapsed_time = time.time() - start_time
|
||
|
|
print(f"✅ Success! Processing complete in {elapsed_time:.2f} seconds.")
|
||
|
|
print(f"📁 Structured output saved cleanly to: {os.path.abspath(output_filename)}")
|
||
|
|
|
||
|
|
print("\n--- Preview of First 300 Characters: ---")
|
||
|
|
print(markdown_output[:300])
|
||
|
|
|
||
|
|
if __name__ == "__main__":
|
||
|
|
main()
|