community pool / model answer

Huihui-Qwen3.6-27B-abliterated-NVFP4-MTP

chat · sakamakismile/Huihui-Qwen3.6-27B-abliterated-NVFP4-MTP

97.5 tok/s · model median · 6 runs in the model record.

Reference rigQuantizationMedian resultBasisTTFT (ms)Peak VRAMMax contextEngines
GB10 Grace Blackwell 128GB4-bit24 tok/sreported348- GB262,144vllm
RTX 5090 32GB4-bit97.5 tok/smeasured11532 GB131,072vllm

Context tested

The table above is the tested context for each exact rig and quantization cell. Empty fields remain empty rather than being filled with an estimate.