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 module
kernels
Registers core tensor graph ops (range, copy, reshape, and related utilities).
comptime values
logger
comptime logger = Logger(stdout, prefix=String(""), source_location=False)
Structs
-
CausalConv1DVarlenFwd: Varlen causal 1D convolution forward pass. -
CausalConv1DVarlenFwdRows:causal_conv1d_varlen_fwdin a mixed verify graph. -
Copy: Registers thecopygraph op with the graph compiler. -
EaglePrefillShiftTokens: Registers thespeculative.eagle_prefill_shift_tokensgraph op with the graph compiler. -
Fold: Registers thefoldgraph op with the graph compiler. -
GatedDeltaConv1dFwd: Gated DeltaNet causal conv1d forward pass (Pass 1 of two-pass prefill). -
GatedDeltaConv1dVerifyFwd:gated_delta_conv1d_fwdin a speculative verify graph. -
GatedDeltaRecurrenceFwd: Gated DeltaNet recurrence forward pass (Pass 2 of two-pass prefill). -
GatedDeltaRecurrenceVerifyRingFwd: Gated DeltaNet recurrence over a speculative verify window. -
GatedDeltaStateFold: Applies a speculative verify's accepted records to the state pool. -
GatedGroupRMSNorm: Fused silu-gate + group RMSNorm + weight-scale for the Mamba-2 mixer. -
InplaceMemcpy: Copies the contents ofsrcintodstin place. -
IRFFT: Registers theirfftgraph op with the graph compiler. -
LaunchHostFunc: Enqueues a pre-packed host callback on the device's default stream. -
Learnable2DInterpPosEmb: Registers thelearnable_2d_interp_pos_embgraph op with the graph compiler. -
Mamba2ReplayFold: Commits a speculative verify's accepted rows to every Mamba-2 layer. -
Mamba2SSDChunkScanVarlenFwdInplace: Varlen Mamba-2 SSD chunked scan with an in-place SSM state pool. -
Mamba2SSDChunkScanVarlenFwdInplaceRows:mamba2_ssd_chunk_scan_varlen_fwd_inplacein a mixed verify graph. -
MaxPool: Registers themo.max_poolgraph op with the graph compiler. -
MaxPoolCeilModeTrue: Registers themo.max_pool_ceil_mode_truegraph op with the graph compiler. -
NanCheckCountOp: Counts NaN/Inf values in a floating-point tensor. -
NanCheckRaiseOp: Raises an error if NaN or Inf counts are non-zero. -
NonMaximumSuppression: Registers themo.non_maximum_suppressiongraph op with the graph compiler. -
PackConvTransposeFilterShape: Registers thepack_conv_transpose_filter_shapegraph op with the graph compiler. -
RandomNormal: Registers themo.random.normalgraph op with the graph compiler. -
RandomUniform: Registers themo.random.uniformgraph op with the graph compiler. -
Range: Registers themo.rangegraph op with the graph compiler. -
RepeatInterleave: Registers therepeat_interleavegraph op with the graph compiler. -
ROIAlign: Registers themo.roi_aligngraph op with the graph compiler. -
Sleep: Registers theruntime.sleepgraph op with the graph compiler. -
SpatialMerge: Registers thevision.spatial_mergegraph op with the graph compiler. -
StateSpaceCopyRows: Copies selected state rows into a separately addressed scratch pool. -
Struct_accept_pair: Registers thesampler.accept_pairgraph op. -
Struct_argmax_with_bias: Registers thesampler.argmax_with_biasgraph op. -
Struct_chunked_qkv_rope_store_ragged_paged: Registers chunked QKV unpacking, partial NeoX RoPE, and paged stores. -
Struct_dspark_draft_sample: Registers thesampler.dspark_draft_samplegraph op. -
Struct_first_rejected: Registers thesampler.first_rejectedgraph op. -
Struct_fused_token_sampling: Registers thesampler.fused_token_samplinggraph op with the graph compiler. -
Struct_fused_token_sampling_with_dist: Registers thesampler.fused_token_sampling_with_distgraph op. -
Struct_gumbel_argmax_from_probs: Registers thesampler.gumbel_argmax_from_probsgraph op. -
Struct_gumbel_argmax_from_residual: Registers thesampler.gumbel_argmax_from_residualgraph op. -
Struct_hyper_connection_gates: Registers thehyper_connection.gatesgraph op with the graph compiler. -
Struct_ihc_collapse_rms_norm: Registers thehyper_connection.ihc.collapse_rms_normgraph op with the graph compiler. -
Struct_ihc_mix_gemv: Registers thehyper_connection.ihc.mix_gemvgraph op with the graph compiler. -
Struct_k_rope_v_store_ragged_paged: Registers head-major K rotation and paged K/V stores, with no query. -
Struct_keyed_uniform: Registers thesampler.keyed_uniformgraph op. -
Struct_mhc_post: Registers thehyper_connection.mhc.postgraph op with the graph compiler. -
Struct_mhc_pre: Registers thehyper_connection.mhc.pregraph op with the graph compiler. -
Struct_min_p_sampling: Registers themin_p_samplinggraph op with the graph compiler. -
Struct_moe_create_indices: Registers themoe.create.indicesgraph op with the graph compiler. -
Struct_moe_create_indices_packed: Registers themoe.create.indices.packedgraph op with the graph compiler. -
Struct_moe_create_indices_with_scales_offset: Registers themoe.create.indices.with.scales.offsetgraph op with the graph compiler. -
Struct_moe_create_indices_with_slots: Registers themoe.create.indices.with.slotsgraph op with the graph compiler. -
Struct_moe_eplb_remap: Registers themoe.eplb.remapgraph op with the graph compiler. -
Struct_moe_exit_act_concat: Registers themoe.exit.act.concatgraph op with the graph compiler. -
Struct_moe_finalize: Registers themoe.finalizegraph op with the graph compiler. -
Struct_moe_finalize_bounded: Registers themoe.finalize.boundedgraph op with the graph compiler. -
Struct_moe_router_group_limited: Registers themoe.router.group.limitedgraph op with the graph compiler. -
Struct_moe_sigmoid_gemv_group_limited_router_packed: Registers themoe.sigmoid.gemv.group.limited.router.packedgraph op. -
Struct_moe_sigmoid_gemv_single_group_router: Registers themoe.sigmoid.gemv.single.group.routergraph op. -
Struct_moe_single_group_router: Registers themoe.single.group.routergraph op with the graph compiler. -
Struct_moe_single_group_router_eplb: Registers themoe.single.group.router.eplbgraph op with the graph compiler. -
Struct_moe_single_group_router_image_bias: Registers themoe.single.group.router.image.biasgraph op. -
Struct_moe_single_group_router_packed: Registers themoe.single.group.router.packedgraph op. -
Struct_moe_sink_gate_router: Registers themoe.sink.gate.routergraph op with the graph compiler. -
Struct_rope_qkv_store_ragged_paged: Registers head-major Q/K rotation and paged K/V stores. -
Struct_rope_ragged_paged: Registers themo.composite.rope.raggedgraph op with the graph compiler. -
Struct_rope_ragged_paged_with_position_id: Registers therope.ragged.with_position_idgraph op with the graph compiler. -
Struct_rope_split_store_ragged_paged: Registers thekv_cache.rope_split_store.ragged.pagedgraph op with the graph compiler. -
Struct_rope_split_store_ragged_paged_with_position_id: Registers thekv_cache.rope_split_store.ragged.paged.with_position_idgraph op with the graph compiler. -
Struct_rope_split_store_ragged_paged_with_qk_norm: Registers thekv_cache.rope_split_store.ragged.paged.with_qk_normgraph op with the graph compiler. -
Struct_sampler_apply_packed_bitmask_with_penalties: Registerssampler.apply_packed_bitmask_with_penalties: the packed grammar mask plus presence/frequency penalties in one output, so the penalties ride the mask's full pass over the logits instead of a copy. -
Struct_sampler_apply_penalties: Registers thesampler.apply_penaltiesgraph op with the graph compiler. -
Struct_sampler_update_frequency_data: Registers thesampler.update_frequency_datagraph op with the graph compiler. -
Struct_topk_topp_masked_probs: Registers thesampler.topk_topp_masked_probsgraph op. -
TPoolPatchMerger: Registers thetpool_patch_mergergraph op with the graph compiler. -
WaitHostValue: Stalls the stream until a host-visible flag reaches a given value. -
WaitHostValueWithDep: Variant ofruntime.wait_host_valuethat takes a fake mutable dependency operand.
Functions
-
causal_conv1d_varlen_fwd_shape: Computes the output shape for thecausal_conv1d_varlen_fwdgraph op. -
concat_from_list_shape_impl: Computes the concatenated output shape for a variadic list of input tensors. -
concat_shape_impl: Computes the concatenated output shape from the input tensor shapes. -
export: Keeps the managed-tensor-slice load/store entry points alive for export when this package is linked. -
fold_kernel_shape: Computes the output shape for thefoldgraph op. -
gated_delta_conv1d_fwd_shape: Computes the output shape for thegated_delta_conv1d_fwdgraph op. -
gated_delta_recurrence_fwd_shape: Computes the output shape for thegated_delta_recurrence_fwdgraph op. -
gated_delta_recurrence_verify_ring_fwd_shape: Computes the output shape forgated_delta_recurrence_verify_ring_fwd. -
gated_group_rmsnorm_shape: -
layout_transform_conv_filter_common: Packs a convolution filter into the layout expected by the conv kernels. -
layout_transform_conv_transpose_filter_common: Packs a transposed-convolution filter into the layout expected by the conv_transpose kernels. -
mamba2_ssd_chunk_scan_varlen_fwd_inplace_shape: Computes the output shape for themamba2_ssd_chunk_scan_varlen_fwd_inplacegraph op. -
max_pool_ceil_mode_true_shape: Computes the output shape for themo.max_pool_ceil_mode_truegraph op. -
max_pool_shape: Computes the output shape for themo.max_poolgraph op. -
non_maximum_suppression_shape: Computes the output shape for themo.non_maximum_suppressiongraph op. -
pack_conv_transpose_filter_shape_shape: Computes the output shape for thepack_conv_transpose_filter_shapegraph op. -
random_normal_shape: Computes the output shape for themo.random.normalgraph op. -
random_uniform_shape: Computes the output shape for themo.random.uniformgraph op. -
range_shape: Computes the output shape for themo.rangegraph op. -
reduce_shape: Compute the output shape of areduceoperation, and assert the inputs are compatible. -
repeat_interleave_kernel_shape: Computes the output shape for therepeat_interleavegraph op. -
roi_align_shape: Computes the output shape for themo.roi_aligngraph op. -
tpool_patch_merger_shape: Computes the output shape for thetpool_patch_mergergraph op.