coba:1brejects 8 %
coba · Ollaya · 1.1B · ONNX Runtime · Apache-2.0
- Score
- 36.3
- 95 % interval 34.3–38.6
- Rank
- 41
- of 66 scored models
- Public benchmark
- 56.1
- macro accuracy, %
- typed-decisions
- 60.8
- accuracy, %
By task
Chance-corrected accuracy on each of the 14 tasks: 0 is guessing, 100 is every answer right.
Workflows
typed-decisions
42.6
Classification
MASSIVE en
76.1
MASSIVE de
48.3
Civil Comments
76.7
Aegis 2
19.2
Understanding
MultiNLI
61.4
VitaminC
62.7
PAWS
20.0
Knowledge
BoolQ
51.3
SQuAD 2
33.8
PubMedQA
43.0
Judging
HelpSteer 2
-6.9
SummEval rel.
-16.1
SummEval cons.
-3.3
Accuracy and calibration
| Benchmark | Accuracy | Choice | Noul | Score | ECE | Brier | Rejected |
|---|---|---|---|---|---|---|---|
| Public benchmark | 56.1(54.8–57.4) | 68.0 | 70.1 | 13.0 | 0.090 | 0.449 | 453 / 3,880 |
| typed-decisions | 60.8(58.1–63.5) | 62.2 | 63.7 | 57.6 | 0.072 | 0.519 | 7 / 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 | 20.1 ms | 24.5 ms | 13.9 s | 3.6 GB | 4.7 GB | F16 |
| Threadripper 3960X CPU | 1.48 s | 1.53 s | 11.8 s | – | 4.3 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
| Device | Same decision | Max logit diff. | Result | Record |
|---|---|---|---|---|
| M4 Pro METAL | 2383/2383 | – | pass | results/runs/2026-10-23-parity-macm4pro-metal-coba.json |
| RTX 5090 CPU | 2383/2383 | – | pass | results/runs/2026-10-23-parity-rtx5090-cpu-coba.json |
| RTX 5090 CUDA | 2383/2383 | – | pass | results/runs/2026-10-23-parity-rtx5090-cuda-coba.json |
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 coba:1b
# 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 coba:1b \
--url http://127.0.0.1:11435 --data typed.jsonl --out typed/coba-1b
# Bespoke Labs' public benchmark; build its 13 subsets first
NIMBLE_SRC=nimble python -m ollaya_convert.bench_public run --model coba:1b \
--url http://127.0.0.1:11435 --data public --out public/coba-1b
# speed and memory
python -m ollaya_convert.bench_latency --models coba:1b --memory \
--machine my-machine --device cuda --out latency.json