bitllm / app.py
ar08's picture
Create app.py
861cb4a verified
raw
history blame
401 Bytes
import subprocess
import os
# Go one directory up
os.chdir("..")
# Prepare the command
command = [
"python",
"run_inference_server.py",
"-m", "./models/ggml-model-i2_s.gguf",
"--host", "0.0.0.0",
"--port", "7860"
]
# Run it as a subprocess
try:
subprocess.run(command, check=True)
except subprocess.CalledProcessError as e:
print(f"Error running inference server: {e}")