decider:2b
decider · Mapika · 1.9B · ONNX Runtime · Apache-2.0
- Score
- 54.3
- 95 % interval 52.1–56.2
- Rank
- 16
- of 35 scored models
- Public benchmark
- 70.3
- macro accuracy, %
- typed-decisions
- 59.1
- accuracy, %
By task
Chance-corrected accuracy on each of the 14 tasks: 0 is guessing, 100 is every answer right.
Workflows
typed-decisions
40.1
Classification
MASSIVE en
80.9
MASSIVE de
78.2
Civil Comments
68.0
Aegis 2
44.0
Understanding
MultiNLI
78.4
VitaminC
45.4
PAWS
44.0
Knowledge
BoolQ
60.7
SQuAD 2
56.5
PubMedQA
59.2
Judging
HelpSteer 2
28.2
SummEval rel.
18.8
SummEval cons.
58.3
Accuracy and calibration
| Benchmark | Accuracy | Choice | Noul | Score | ECE | Brier | Rejected |
|---|---|---|---|---|---|---|---|
| Public benchmark | 70.3(69.0–71.7) | 76.7 | 77.3 | 48.1 | 0.087 | 0.402 | 0 / 3,880 |
| typed-decisions | 59.1(56.6–61.5) | 55.7 | 65.8 | 56.6 | 0.088 | 0.554 | 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 | 193 ms | 194 ms | 32.2 s | 11.9 GB | 11.4 GB | F32 |
| Threadripper 3960X CPU | 4.54 s | 4.56 s | 36.6 s | – | 14.6 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 |
|---|---|---|---|---|
| RTX 4090, CUDA 13; ONNX Runtime 1.28 | 100 % | 5.2e-5 | pass | families/decider.md:298-303 |
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 decider: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 decider:2b \
--url http://127.0.0.1:11435 --data typed.jsonl --out typed/decider-2b
# Bespoke Labs' public benchmark; build its 13 subsets first
NIMBLE_SRC=nimble python -m ollaya_convert.bench_public run --model decider:2b \
--url http://127.0.0.1:11435 --data public --out public/decider-2b
# speed and memory
python -m ollaya_convert.bench_latency --models decider:2b --memory \
--machine my-machine --device cuda --out latency.json