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

reductions

Registers reduction graph ops (sum, mean, argmax, and related) over tensor axes.

`comptime` values

`logger`

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

Logger for the reductions module.

Structs

ApplyQKRMSNorm: Fused per-element QK-RMSNorm apply for two operands Q and K.
ArgMax: Registers the mo.reduce.arg_max graph op with the graph compiler.
ArgMin: Registers the mo.reduce.arg_min graph op with the graph compiler.
ArgNonZero: Registers the mo.arg_nonzero graph op with the graph compiler.
ArgSort: Registers the mx.argsort graph op with the graph compiler.
BottomK: Registers the mo.bottom_k graph op with the graph compiler.
CumSum: Registers the mo.cumsum graph op with the graph compiler.
LayerNorm: Registers the mo.reduce.layer_norm graph op with the graph compiler.
LogSoftmax: Registers the mo.reduce.logsoftmax graph op with the graph compiler.
Mean: Registers the mo.reduce.mean graph op with the graph compiler.
ReduceAdd: Registers the mo.reduce.add graph op with the graph compiler.
ReduceGroupNorm: Registers the mo.reduce.group_norm graph op with the graph compiler.
ReduceMax: Registers the mo.reduce.max graph op with the graph compiler.
ReduceMin: Registers the mo.reduce.min graph op with the graph compiler.
ReduceMinAndMax: Registers the mo.reduce.reduce_min_and_max graph op with the graph compiler.
ReduceMul: Registers the mo.reduce.mul graph op with the graph compiler.
ReduceRMSNorm: Registers the mo.reduce.rms_norm graph op with the graph compiler.
ReduceRMSNormFusedResidualAdd: Registers the mo.composite.rms_norm_fused_residual_add graph op with the graph compiler.
ReduceRMSNormRoPE: Fuses RMS normalization and Rotary Position Embedding (RoPE) into one operation.
RMSNormResidualAdd: Fused single-norm residual-add + RMSNorm.
RowMeanOfSquares: Per-row mean of squares over the last axis, accumulated in float32.
RowMeanOfSquaresQK: Fused per-row mean of squares for two operands Q and K.
Softmax: Registers the mo.reduce.softmax graph op with the graph compiler.
TopK: Registers the mo.top_k graph op with the graph compiler.

Functions

arg_nonzero_shape: Computes the output shape for the mo.arg_nonzero graph op.
bottom_k_shape: Computes the output shape for the mo.bottom_k graph op.
composite_rms_norm_fused_residual_add_shape: Computes the output shape for the mo.composite.rms_norm_fused_residual_add graph op.
composite_rms_norm_residual_add_shape:
composite_rms_norm_rope_shape: Computes the output shape for the mo.composite.rms_norm_rope graph op.
reduce_add_shape: Computes the output shape for the mo.reduce.add graph op.
reduce_group_norm_shape: Computes the output shape for the mo.reduce.group_norm graph op.
reduce_layer_norm_shape: Computes the output shape for the mo.reduce.layer_norm graph op.
reduce_max_shape: Computes the output shape for the mo.reduce.max graph op.
reduce_mean_shape: Computes the output shape for the mo.reduce.mean graph op.
reduce_min_shape: Computes the output shape for the mo.reduce.min graph op.
reduce_mul_shape: Computes the output shape for the mo.reduce.mul graph op.
reduce_reduce_min_and_max_shape: Computes the output shape for the mo.reduce.reduce_min_and_max graph op.
reduce_rms_norm_shape: Computes the output shape for the mo.reduce.rms_norm graph op.
reduce_row_mean_of_squares_qk_shape: Computes the output shape for the mo.reduce.row_mean_of_squares_qk graph op.
reduce_row_mean_of_squares_shape: Computes the output shape for the mo.reduce.row_mean_of_squares graph op.
top_k_shape: Computes the output shape for the mo.top_k graph op.

comptime values​

logger​

Structs​

Functions​

`comptime` values

`logger`

Structs

Functions