community pool / model answer

Laguna-S-2.1-NVFP4

chat · poolside/Laguna-S-2.1-NVFP4

101.4 tok/s · model median · 8 runs in the model record.

Reference rigQuantizationMedian resultBasisTTFT (ms)Peak VRAMMax contextEngines
GB10 Grace Blackwell 121GB4-bit20.6 tok/sreported426- GB128,000vllm
GB10 Grace Blackwell 128GB4-bit73.8 tok/smeasured328109 GB262,144vllm
RTX 5090 32GB ×44-bit157.1 tok/sreported-- GB262,144vllm
RTX PRO 6000 Blackwell 96GB4-bit149.6 tok/sreported57- GB2,048vllm

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.