ConceptW8A8 Quantization
W8A8 Quantization
A quantization scheme that compresses both model weights and activations to 8-bit integers. Compared with FP16 it substantially reduces memory footprint and improves NPU throughput, making it a compromise choice on edge chips that balances speed against runnability.
Timeline (last 90 days)
Oct 15
W8A8 quantization means both weights and activations are compressed to 8-bit integers; compared with FP16 it can reduce memory usage and improve NPU throughput, at the cost of possibly introducing slight accuracy loss.
Unverified50%