For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /max/get-started.md).

Mojo function

compute_scales_fp8_kernel

def compute_scales_fp8_kernel[out_type: DType, scales_type: DType, in_type: DType, input_fn: def[width: Int, alignment: Int](row: Int, col: Int) capturing -> SIMD[in_type, width], num_threads: Int, group_size: Int, simd_width: Int, scales_layout: TensorLayout](scales: TileTensor[scales_type, scales_layout, MutAnyOrigin], scale_ub: Scalar[scales_type])

Compute per-group FP8 scale factors without quantizing.

Each block scans its (row, group) tile via input_fn, computes the scale factor, and writes it to scales[group_idx, row]. This is the first half of quantize_fp8_kernel — used by the per-tensor path so the second kernel can find the tensor-wide max scale.

Was this page helpful?

Thank you! We'll create more content like this.

Thank you for helping us improve!

😔 What went wrong?

Some code doesn’t work

It includes inaccurate information

It's missing information I need

It was difficult to understand

Other

Submit