Elon Musk: @O42nl @kirillgroshkov Tesla 进行 INT8 推理。比 FP16 更有效，但我们花了很多精力来克服量化误差。

原推：@O42nl @kirillgroshkov Tesla does INT8 inference. Way more efficient than FP16, but took us a lot of effort to overcome quantization errors.
https://twitter.com/elonmusk/status/1630657812530339855