The runs are an A/B between MTPLX's own verify kernels, with and without the compiled verify graph, and a port of a third-party kernel used as the control. The record cell is MTPLX's own kernels plus the compiled graph, twin-confirmed. The plain-decode baseline is flat across all seven runs (30.23 to 30.43 tok/s), so the multiplier rests on a flat baseline.
| Log | Configuration | AR tok/s | D3 tok/s | Multiplier |
|---|---|---|---|---|
| oursW2-1.log | MTPLX kernels + compiled verify (record) | 30.37 | 81.74 | 2.69x |
| oursW2-2.log | MTPLX kernels + compiled verify (twin) | 30.23 | 81.73 | 2.70x |
| portW2-1.log | ported kernel + compiled verify | 30.33 | 81.68 | 2.69x |
| portW2-2.log | ported kernel + compiled verify (twin) | 30.33 | 81.64 | 2.69x |
| port1.log | ported kernel, eager | 30.38 | 75.46 | 2.48x |
| port2.log | ported kernel, eager (twin) | 30.43 | 75.48 | 2.48x |
| ours-plain.log | MTPLX kernels, eager | 30.25 | 74.77 | 2.47x |
Acceptance by depth on every run: 50/53, 47/53, 42/53 (94.3, 88.7, 79.2 percent). Sampler: temperature 0.6, top-p 0.95, top-k 20, thinking off. The 192-token generation is the burst lane, the shape of a short chat reply. The same configuration on an uncapped 11,390-token Flappy Bird generation with reasoning on ran 62.95 tok/s the same afternoon (75.5 tok/s over the first 128 tokens, 39.9 over the last 128); the compiled verify graph shipped as the turbo default in MTPLX 2.0.0 four days later.