sura:2b
sura · Ollaya · 2B · ONNX Runtime · Apache-2.0
- Score
- 40.9
- 95 % interval 38.5–43.0
- Rank
- 39
- of 66 scored models
- Public benchmark
- 61.0
- macro accuracy, %
- typed-decisions
- 57.0
- accuracy, %
By task
Chance-corrected accuracy on each of the 14 tasks: 0 is guessing, 100 is every answer right.
Accuracy and calibration
| Benchmark | Accuracy | Choice | Noul | Score | ECE | Brier | Rejected |
|---|---|---|---|---|---|---|---|
| Public benchmark | 61.0(59.4–62.6) | 68.8 | 68.9 | 34.9 | 0.060 | 0.481 | 0 / 3,880 |
| typed-decisions | 57.0(54.1–59.8) | 58.7 | 60.2 | 53.4 | 0.102 | 0.547 | 0 / 400 requests |
Accuracy in %, with its 95 % bootstrap interval. A rejected request (the model can’t take the question, or the input is too long for it) counts as wrong.
Speed and memory
| Device | p50 | p90 | Load | VRAM | RAM peak | Precision |
|---|---|---|---|---|---|---|
| RTX 5090 CUDA | 62.6 ms | 68.3 ms | 25.3 s | 7.9 GB | 11.4 GB | F32 |
| Threadripper 3960X CPU | 1.94 s | 2.23 s | 26.3 s | – | 10.5 GB | F32 |
Five-question requests over HTTP, one at a time, each a different short message. VRAM: the rise of the GPU’s used memory while the model is loaded. RAM peak: the runner process’s peak resident memory.
Parity with the authors’ code
Its parity check is in the model’s family documentation in the repository (docs/families/); the compiled parity file does not list it yet.
Reproduce
The tools are in convert/ollaya_convert and need only Python’s standard library. Run them against a local Ollaya 0.13.0 with the same model digest.
ollaya pull sura:2b
# typed-decisions; export the test split once with
# uv run python -m ollaya_convert.bench_typed export --out typed.jsonl
python -m ollaya_convert.bench_typed run --model sura:2b \
--url http://127.0.0.1:11435 --data typed.jsonl --out typed/sura-2b
# Bespoke Labs' public benchmark; build its 13 subsets first
NIMBLE_SRC=nimble python -m ollaya_convert.bench_public run --model sura:2b \
--url http://127.0.0.1:11435 --data public --out public/sura-2b
# speed and memory
python -m ollaya_convert.bench_latency --models sura:2b --memory \
--machine my-machine --device cuda --out latency.json