2 ms·

Would be good to see some non-batch inference benchmarks for this. I couldn't find any numbers online.