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

Mojo function

mxfp4_matmul_sm90

def mxfp4_matmul_sm90(c: TileTensor[Storage=c.Storage, address_space=c.address_space, linear_idx_type=c.linear_idx_type], a: TileTensor[Storage=a.Storage, address_space=a.address_space, linear_idx_type=a.linear_idx_type], b_packed: TileTensor[Storage=b_packed.Storage, address_space=b_packed.address_space, linear_idx_type=b_packed.linear_idx_type], b_scales: TileTensor[Storage=b_scales.Storage, address_space=b_scales.address_space, linear_idx_type=b_scales.linear_idx_type], ctx: DeviceContext)

MXFP4 matmul: dequant B weights to FP8, cast A to FP8, SM90 FP8 GEMM.

Args:

• ​c (TileTensor[Storage=c.Storage, address_space=c.address_space, linear_idx_type=c.linear_idx_type]): Output [M, N] in bfloat16.
• ​a (TileTensor[Storage=a.Storage, address_space=a.address_space, linear_idx_type=a.linear_idx_type]): Activations [M, K] in bfloat16.
• ​b_packed (TileTensor[Storage=b_packed.Storage, address_space=b_packed.address_space, linear_idx_type=b_packed.linear_idx_type]): Weights [N, K//2] in uint8 (packed MXFP4).
• ​b_scales (TileTensor[Storage=b_scales.Storage, address_space=b_scales.address_space, linear_idx_type=b_scales.linear_idx_type]): Weight scales [N, K//32] in float8_e8m0fnu.
• ​ctx (DeviceContext): Device context.