Name: Edge AI Lab
Author: MySimulator

Compression settings

Quantization: 32-bit float Pruning: 0% of weights removed Target device

Baseline is a 25MB fp32 model at 94.0% accuracy and 48ms latency on a phone NPU. Compression trades some accuracy for smaller size and faster inference — up to a point, then accuracy falls off a cliff.

Result

Model size—

Estimated latency—

Estimated accuracy—

Fits on device?—

Accuracy vs. latency trade-off (marker = current config)