Using LRE to Deploy your Encrypted Model¶

This notebook describes deployment of a LEIP Optimize encrypted model. Encryption of the model has to be done in your host environment described in the LEIP Optimize encryption tutorial.

You should have produced a bundle which has a modelLibrary.so and a modelKey.bin. You will also have used a password during the encryption. You will need these 3 artifacts to decrypt your model.

First, we will locate these artifacts for deployment.

• path to modelLibrary.so is set to model_path
• path to modelKey.bin is set to key_path
• password string is set to password

In [6]:

from pylre import LatentRuntimeEngine as LRE
encrypted_output_dir = "encrypted_output"
model_path = f"{encrypted_output_dir}/modelLibrary.so"
key_path = f"{encrypted_output_dir}/modelKey.bin"
password = "test_password"

from pylre import LatentRuntimeEngine as LRE encrypted_output_dir = "encrypted_output" model_path = f"{encrypted_output_dir}/modelLibrary.so" key_path = f"{encrypted_output_dir}/modelKey.bin" password = "test_password"

We will first check if the model can be loaded, just like we did in the Deploy a Model tutorial. You will run into an error saying this is an encrypted model.

In [ ]:

try:
    lre_llvm_insecure = LRE(model_path)
except Exception as e:
    print(f"Error initializing LRE: {e}")

try: lre_llvm_insecure = LRE(model_path) except Exception as e: print(f"Error initializing LRE: {e}")

Now let's load the model with the proper credentials.

In [ ]:

lre_llvm_secure = LRE(model_path, password="test_password", key_path=key_path)

lre_llvm_secure = LRE(model_path, password="test_password", key_path=key_path)

Now you can use the lre_llvm_secure LRE object like we did in the Deploy a Model tutorial. However, please note that once decrypted, the lre_llvm_secure is no longer secure. Please ensure your deployment environment is sandboxed before decrypting to maintain your model security.