Block a user
feat(neuron): Laya decision-model engine (#336)
feat(neuron): Laya decision-model engine (#336)
feat(laya): port the language/script detector (part of #339)
feat(laya): port the language/script detector (part of #339)
L4 — Laya performance on the 3060: batched forward, flash-attn (ampere), micro-batching — GATE for the residency swap
First GPU numbers (quadbrat 3060, sm_86, CUDA 13.0, release build of the #336 engine, no flash-attn)
Run inside a guarded window: both resident models idle before; 6,525 MB free and no new…
L3 — Laya engine: arch/laya in neuron + device-worker jobs (logit parity)
Pushed as feat/336-laya-engine (ec315a7f). CUDA type-check pending in branch CI.
Parity against the real checkpoints (f32, CPU, env-gated test over #334's reference.json):
feat(systemone): route /v1/systemone through cortex and helexa-router (part of #338)
feat(systemone): route /v1/systemone through cortex and helexa-router (part of #338)
L1 — Laya reference fixtures and parity harness
Landed via #343.
- 28 cases: 21 on the English checkpoint, 6 on multilingual (5 Devanagari, Khmer, CJK, Arabic and non-English Latin states, plus the German JSON state the router sent there),…
L1 — Laya reference fixtures and parity harness