Tag: #INT8
Kubernetes, DevOps and MLOps, written from things we ran and measured ourselves · 1 posts
Reducing Numbers: FP16, BF16, INT8 Quantization, Post-Training Quantization, Quantization-Aware Training, and Calibration Data
Training is done in 32-bit floating-point (FP32), but inference can be done in 16-bit or 8-bit integers. Reducing bits decreases memory movement and enables the use of dedicated arithmetic units, making things faster. We
2026-10-15 · 18 min read #비전 모델#실시간#FP16#BF16#INT8