flashinfer.quantization.nvfp4_e4m3_max

flashinfer.quantization.nvfp4_e4m3_max(nvfp4_4over6_config: NVFP44Over6Config | None = None) → float

E4M3 block-scale clamp implied by a resolved recipe.

None (the default) is standard NVFP4 (448). Unlike the quantizers, omitting the recipe here does not read the environment: pass the recipe resolved for the matching quantize call.