flashinfer.cake_kda_nondet_train_kcp.cake_kda_nondet_train_kcp_comm_bytes¶
- flashinfer.cake_kda_nondet_train_kcp.cake_kda_nondet_train_kcp_comm_bytes(ctx: Any, *, H: int = 64, C: int = 8192, save_transfer: bool | None = None, boundary_mode: str = 'saved', checkpoint_keep: str | None = None) dict¶
Bytes this rank sends and receives in one training step (forward and backward) for
Hheads andCconvolution channels: the state summaries, the boundary states and the halo rows of the context’s collectives, as a table per collective and per boundary (cake_kcp_plan.cake_kda_nondet_train_kcp_comm_bytes).checkpoint_keep: the traffic of a checkpointed step with a recompute handle of that tier (the tier fixessave_transfer;NoneisFalsewithout a tier; an explicit value that contradicts the tier raisesValueError).- Parameters:
ctx – This rank’s context from
cake_kda_nondet_train_kcp_build_context().H (int, default=64) – Number of KDA heads.
C (int) – Convolution channel count used to size halo rows.
save_transfer (bool, optional) – Whether to retain forward transfer buffers; default resolution is described above.
boundary_mode (str) – Boundary-state mode of the call,
"saved"by default.checkpoint_keep (str, optional) – Recompute-handle tier,
"states"or"transfer";Nonequeries the ordinary call.