search

bytedance / MRECG

Apache License 2.0
35 stars 4 forks source link

How to measure improviment of Quantization? #9

Open padeirocarlos opened 6 months ago

padeirocarlos commented 6 months ago

Hi, in this paper you reference that improvement of Quantization. How do you measure this improvement?

I mean, how do you measure the improvement of Memory, Inference time and MAC of Quantized model? Is it possible to calculate this improvement? Could you share a reference if it is possible?!

BobxmuMa commented 6 months ago

This improvement usually relies on the actual deployment of quantization models. the MQBench repository bridges the gap between algorithm and model deployment. It integrates different quantization algorithms as well as deployment modules.MRECG is also compatible with the MQBench codebase.

As for the improvements in the paper, the community usually consider the efficiency improvements to be consistent across algorithms for the same quantization configuration.