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 module

linalg

Registers matmul-family graph ops (matmul, grouped, batched, and quantized variants) and dispatches them to the linalg kernels.

`comptime` values

`logger`

comptime logger = Logger(stdout, prefix=String(""), source_location=False)

Structs

BatchMatmul: Registers the mo.batch_matmul graph op with the graph compiler.
FusedMatmulAdd: Registers the mo.composite.matmul_add graph op with the graph compiler.
LayoutTransformMatmulKN2KNkni: Registers the layout_transform_KN_to_KNkni graph op with the graph compiler.
LayoutTransformMatmulNK2KNkni: Registers the layout_transform_NK_to_KNkni graph op with the graph compiler.
LinalgBandPart: Registers the mo.linalg.band_part graph op with the graph compiler.
Matmul: Registers the mo.matmul graph op with the graph compiler.
MatmulDynamicScaledFloat8: Registers the mo.matmul_dynamic_scaled_fp8 graph op with the graph compiler.
MatmulFusedPartialRMSNorm: Fuses GEMV (M=1 matmul) with partial RMS normalization.
MatmulStaticScaledFloat8: Registers the mo.matmul_static_scaled_float8 graph op with the graph compiler.
MatmulSwiGLU: Fused GEMM+SwiGLU on SM100 for BF16 inputs.
MatmulSwiGLUBias: Fused GEMM+SwiGLU+bias on SM100 for BF16 inputs.
MergeRaggedTensors: Registers the mo.merge_ragged_tensors graph op with the graph compiler.
PackMatmulBShapeFunc: Registers the pack_matmul_b_shape_func graph op with the graph compiler.
Struct_batched_matmul_dynamic_scaled_fp8: Registers the mo.batched.matmul.dynamic.scaled.fp8 graph op with the graph compiler.
Struct_grouped_matmul_block_scaled: MOGG wrapper for grouped block-scaled matrix multiplication.
Struct_grouped_matmul_block_scaled_mxfp4: MOGG wrapper for grouped block-scaled matrix multiplication.
Struct_grouped_matmul_dynamic_scaled_fp8: Registers the mo.grouped.matmul.dynamic.scaled.fp8 graph op with the graph compiler.
Struct_grouped_matmul_ragged: Registers the mo.grouped.matmul.ragged graph op with the graph compiler.
Struct_grouped_matmul_rowwise_dynamic_scaled_fp8: MOGG wrapper for grouped (ragged MoE) rowwise/per-token scaled FP8 matmul.
Struct_grouped_matmul_swiglu_nvfp4: MOGG wrapper for fused grouped NVFP4 matmul + SwiGLU + NVFP4 quant.
Struct_lora_sgmv_qkv_expand_ragged:
Struct_lora_sgmv_qkv_shrink_ragged: Registers the mo.lora_sgmv.qkv_shrink.ragged graph op with the graph compiler.
Struct_lora_sgmv_ragged: Registers the mo.lora_sgmv.ragged graph op with the graph compiler.
Struct_matmul_dynamic_block_scaled: Registers the mo.matmul.dynamic.block.scaled graph op with the graph compiler.
Struct_matmul_dynamic_block_scaled_mxfp4: Registers the mo.matmul.dynamic.block.scaled.mxfp4 graph op with the graph compiler.
Struct_matmul_int8_w8a8_apple: Apple M5 int8 W8A8 matmul (no bias): out = dequant(quant(a) @ b^T).
Struct_matmul_int8_w8a8_apple_bias: Apple M5 int8 W8A8 matmul WITH per-output-channel bias.
Struct_matmul_mxfp4_dequant_fp8: Registers the mo.matmul.mxfp4.dequant.fp8 graph op with the graph compiler.
Struct_matmul_weight_only_block_scaled_apple: Apple M5 weight-only NVFP4 (W4A16) matmul: out = a @ dequant(b)^T.
Struct_matmul_weight_only_scaled_float8_apple: Apple M5 weight-only FP8 (W8A16) matmul: out = a @ dequant(b)^T.
Struct_router_gate_mixed_gemv: MOGG wrapper for the mixed-input router-gate GEMV.

Functions

batch_matmul_shape: Computes the output shape for the mo.batch_matmul graph op.
composite_matmul_fused_partial_rms_norm_shape: Computes the output shape for the mo.composite.matmul_fused_partial_rms_norm graph op.
pack_matmul_b_shape_func_shape: Computes the output shape for the pack_matmul_b_shape_func graph op.

comptime values​

logger​

Structs​

Functions​

`comptime` values

`logger`

Structs

Functions