Status: Internal evaluationData ownership: Institute-ownedLicence: CC BY-NC 4.0 (non-commercial)
Local Inference Task Suite (evaluation set)
A small internal set of representative tasks used to evaluate local vs hosted inference. Composed of synthetic and public-source prompts; contains no personal data.
Purpose
Internal, reproducible comparison of inference settings. Not a benchmark of record.
Source
Synthetic prompts and adapted public-source tasks.
Composition
A few dozen tasks across summarisation, extraction and classification. Small by design.
Known biases
- Task selection reflects our internal priorities and is not representative of all workloads.
- English-only; no multilingual coverage.
Limitations
- Too small to support strong generalisation claims.
- Results are hardware-specific.
Permitted uses
- Non-commercial research reproducibility.
Prohibited uses
- Commercial use.
- Use as an authoritative benchmark.
Access conditions
Available on request for research reproducibility. Not published for download in Phase 1.