crit-hubcrit-hubrtx5090Qwen/Qwen3.8-27B
QUASAR-QAT/Qwen3.8-27B-QUASAR-NVFP4
| Engine | Contributor | Benchmark | Greedy decode | Sampled decode | Prefill | TTFT | Perplexity | KL divergence | Context |
|---|---|---|---|---|---|---|---|---|---|
dflash2-int8kv external | giveen | self-reported · LocalMaxxing | 510.7 tok/s | – | – | – | 4.28116 (absolute) | – | 196,032 tok |