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).
Python module
max.nn.state_space
Python wrappers for the state_space Mojo kernel package.
The module layout mirrors max/kernels/src/state_space/. All ops here are
registered as graph-compiler builtins, so no custom_extensions plumbing
is needed. The legacy Mamba-1 eager wrappers intentionally stay in the Mamba
architecture (their sole consumer) to keep max.nn from depending on
max.experimental.functional.
Submodules
max.nn.state_space.kimi_delta | Python wrapper for the KDA (Kimi Delta Attention) recurrence op. |
|---|
Functions
causal_conv1d_varlen_fwd | Applies a slot-indexed varlen causal depthwise conv1d for prefill and decode. |
|---|---|
gated_delta_conv1d_fwd | Applies the causal conv1d pass, mutating a slot-indexed conv-state pool in place. |
gated_delta_conv1d_verify_fwd | Runs gated_delta_conv1d_fwd() in a speculative verify graph. |
gated_delta_recurrence_fwd | Applies the gated delta recurrence pass, mutating a slot-indexed state pool in place. |
gated_delta_recurrence_verify_ring_fwd | Runs the recurrence over a speculative verify window into a ring. |
gated_delta_state_fold | Applies a verify's accepted ring records to the state pool. |
gated_group_rmsnorm | Fuses the gated group-RMSNorm (HF Zamba2RMSNormGated with norm_before_gate=False) into a single dispatch. |
kda_chunk | Runs the KDA chunk-parallel prefill recurrence, mutating state_pool. |
kda_chunk_supports_head_dims | Reports whether kda_chunk is compiled for this head-dim pair. |
kda_decode | Runs the KDA recurrence, mutating state_pool in place. |
mamba2_ssd_chunk_scan_varlen_fwd_inplace | Performs the Mamba-2 SSD chunked-scan forward, writing final states back into the SSM pool in place. |
short_conv_ring | Returns x + conv(x) over a ragged batch, optionally committing the chunk to ring. |
verify_width_operand | Returns the operand the speculative state ops read the verify width from. |