Quantize both weights and activations to INT8 for faster inference. W8A8 delivers compute speedup through INT8 tensor cores, unlike W8A16 which only reduces memory 👣.
Perfect for single GPU deployments where memory is constrained and latency matters.
