INT8 – TildAlice

ONNX INT8 vs FP16: 3x Latency Drop on Jetson Orin Nano

2026년 04월 11일

YOLOv8n INT8 cut latency from 47ms to 15ms on Jetson Orin Nano — but small-object mAP dropped 5.7%. Real tradeoff numbers with power benchmarks.

2026년 04월 03일

INT8 quantization slashes AWS inference costs 54% vs FP16 for 7B LLMs. Real g5.xlarge benchmarks reveal the accuracy-speed-cost tradeoffs.

2026년 02월 19일

Compare QAT vs PTQ to find when that 3% accuracy gap destroys real-world performance—and which quantization method saves your model.

2026년 01월 15일

Quantizing Whisper for mobile: post-training quantization failed, QAT crashed, but ONNX Runtime reduced model size 75% without accuracy loss.