Expand description
2026-09-28: Plan to program. layout states the storage the emitters’ kernels need (the
residual stream and the logits bound to the model’s buffers, in-place outputs, packed
outputs); the caller places the rest in one workspace allocated at boot; compile asks
each group’s emitter for its launches.
Owner: model-layers circuit executor. Invariants:
- Every stream edge (a block’s
stream_inorstream_out) is the model’shiddenbuffer, updated in place, as the legacy layers update it; every declared output is the model buffer its producer writes (logits,tokens). - The program has exactly
plan.launches()launches; any other count is a compile error. - A group’s weights are checked against the format its nodes were resolved to.
Re-exports§
pub use super::fixed::DraftFixed;pub use super::fixed::Fixed;