laya for Ollaya
Ollaya package of convaiinnovations/laya by Convai Innovations. Ollaya runs open decision models locally, the way Ollama runs LLMs: typed questions in, calibrated answers out, behind a TypeSafe-compatible API.
ollaya run laya
What is in this repository
This repository holds only the files Ollaya derives, with no weights. Each graph is an ONNX export of the
original model whose weights reference the authors' own weight files by byte offset,
so ollaya pull downloads the weights from the upstream repositories, unmodified and pinned to a
commit, and verifies their sha256.
| Tag | Upstream | Files |
|---|---|---|
laya:en |
convaiinnovations/laya@aa8c91c | en/model-fp32.onnx, en/model-fp16.onnx, en/decision.json, en/calibration.json |
laya:multilingual |
convaiinnovations/laya@aa8c91c | multilingual/model-fp32.onnx, multilingual/model-fp16.onnx, multilingual/decision.json, multilingual/calibration.json |
laya:typed-decisions |
convaiinnovations/laya@aa8c91c | typed-decisions/model-fp32.onnx, typed-decisions/model-fp16.onnx, typed-decisions/decision.json, typed-decisions/calibration.json |
Each tag has an fp32 graph (CPU) and an fp16 graph (GPU). Each tag also has decision.json (sequence layout, special tokens) and
calibration.json (temperatures).
Parity
The exports are checked against the PyTorch reference on 2,383 questions per checkpoint. The checks use typed-decisions plus multilingual and edge cases:
- fp32: the same decision on 100% of questions, with probabilities within 1.1e-4.
- fp16: the same decision on 99.1–99.6% of questions. Nearly all of the differences are near-ties between the top two options.
The fp32 graphs (CPU) are opset 23: attention runs as fused Attention nodes, which ONNX Runtime's
CPU provider runs faster than the decomposed ops, so they need ONNX Runtime 1.23 or newer. The fp16
graphs (GPU) are opset 20.
License
Same as the upstream model (Apache-2.0). Ollaya itself is Apache-2.0.