AstrAI

Files

T

ViperEkura 7540acb43e perf: dispatch linear gemv by decode batch size and unify extension style

- replace the per-shape auto tables in the linear backend with an M-banded rule (M in [2,4] on compute capability 8.0+) that measured at the HBM bandwidth floor across every family, and fold the capability check into the capable guard
- drop the unreachable swiglu auto shape-table machinery so both backends share one env-mode ladder via the new dispatch.env_mode helper
- add __all__ across extension modules, name the rotary registration records, and unify typing to the typing-module style
- rewrite test_linear_dispatch.py around behavioral routing assertions and document the M-banded policy in the developer docs
- Benchmark: L20 SM89, Python dispatch overhead 2.9us to 1.5us, auto now covers every projection shape at M in [2,4].

2026-09-03 07:23:13 +08:00

architecture.md

feat: version rollout weight updates

2026-09-02 19:01:41 +08:00

cuda_kernels.md

perf: dispatch linear gemv by decode batch size and unify extension style

2026-09-03 07:23:13 +08:00

dataflow.md

docs: sync all documentation with current codebase

2026-08-07 23:52:27 +08:00

decode_linear_benchmark.md

perf: dispatch linear gemv by decode batch size and unify extension style