colbert qps
0
20
40
60
80
throughput, qps
quantization config, colbert rescorer (hybrid prefetch, k=20, pf=100, concurrency=8)
30
no-quant
(f32)
39
no-quant
(f16)
20
no-quant
(turbo4)
23
scalar
int8
23
product
x16
50
binary
(f32)
39
binary
(f16)
25
turbo4
tq1