encode_pytorch_tensor

Engine.encode_pytorch_tensor(tensor, level)

Encodes a PyTorch tensor into a plaintext format for encryption. The tensor's device can be either CPU or CUDA and must match that of the engine. The message is encoded in segments of length Engine.slot_count. If the message length is not divisible by the slot count, the remaining slots are padded with zeros during encoding.

• Input:
  • tensor: torch.Tensor
  • level: optional
    • Specifies the maximum multiplication level for encoding. Lower levels reduce space usage but also reduce the maximum multiplication depth.
• Output:
  • Plaintext

Examples

CPU

import torch

from desilofhe import Engine

engine = Engine()

tensor = torch.tensor([1, 2, 3, 4])
encoded = engine.encode_pytorch_tensor(tensor)
encoded_level_2 = engine.encode_pytorch_tensor(tensor, level=2)

GPU

import torch

from desilofhe import Engine

engine = Engine(mode="gpu")

tensor = torch.tensor([1, 2, 3, 4], device="cuda")
encoded = engine.encode_pytorch_tensor(tensor)

Typically, the encrypt function handles encoding automatically. Pre-encoding a message can improve efficiency when the same message is reused across multiple operations, as it avoids redundant encoding.