About the FPGA Performance Metrics

[SinnLiu]

Hi, I have a question about FPGA Performance Metrics. In the "The single core implementation" part of section 5.2, which show above

Table 3. Given the 370M parameter model where L = 24, d = 512, the total projected runtime is 16.08ms, and a throughput of approximately 62 tokens per second. The 1.3B parameter model, where L = 24 and d = 2048, has a projected runtime of 42ms, and a throughput of 23.8 tokens per second.

But as shown in Table 3, the total projected runtime is 43ms where L = 24, d = 512, and the same as the case of L = 24 and d = 2048. Is there something wrong with my understanding？It would be my honor if you could reply.

[SinnLiu]

So, why won't anyone answer my question？

[ridgerchu]

Hello,
I believe our FPGA team will try to figure that out, but it seems they haven't replied yet... @sifferman could you please help solve this issue if possible?