decider:4b
decider · Mapika · 4.2B · ONNX Runtime · Apache-2.0
- Score
- 62.9
- 95 % interval 61.0–64.9
- Rank
- 2
- of 35 scored models
- Public benchmark
- 75.6
- macro accuracy, %
- typed-decisions
- 68.0
- accuracy, %
By task
Chance-corrected accuracy on each of the 14 tasks: 0 is guessing, 100 is every answer right.
Workflows
typed-decisions
53.1
Classification
MASSIVE en
85.5
MASSIVE de
83.4
Civil Comments
74.7
Aegis 2
60.8
Understanding
MultiNLI
89.5
VitaminC
61.2
PAWS
53.6
Knowledge
BoolQ
73.3
SQuAD 2
57.9
PubMedQA
56.2
Judging
HelpSteer 2
31.2
SummEval rel.
22.4
SummEval cons.
77.4
Accuracy and calibration
| Benchmark | Accuracy | Choice | Noul | Score | ECE | Brier | Rejected |
|---|---|---|---|---|---|---|---|
| Public benchmark | 75.6(74.4–77.0) | 81.7 | 82.0 | 54.9 | 0.043 | 0.328 | 0 / 3,880 |
| typed-decisions | 68.0(65.5–70.5) | 68.2 | 74.2 | 63.2 | 0.020 | 0.434 | 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 | 393 ms | 394 ms | 22.1 s | 11.0 GB | 2.0 GB | F32 |
| Threadripper 3960X CPU | 11.9 s | 11.9 s | 20.2 s | – | 9.3 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 |
|---|---|---|---|---|
| i9-13900K; ONNX Runtime 1.28; weights_in_memory bf16 | 100 % | 3.0e-5 | pass | families/decider.md:298-303 |
| RTX 4090, CUDA 13; ONNX Runtime 1.28 | 100 % | 2.7e-5 | pass | families/decider.md:298-303 |
| RTX 4090 with a CUDA 13 driver, CUDA 12 pack | 100% | – | pass | distribution.md:126-130 |
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:4b
# 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:4b \
--url http://127.0.0.1:11435 --data typed.jsonl --out typed/decider-4b
# Bespoke Labs' public benchmark; build its 13 subsets first
NIMBLE_SRC=nimble python -m ollaya_convert.bench_public run --model decider:4b \
--url http://127.0.0.1:11435 --data public --out public/decider-4b
# speed and memory
python -m ollaya_convert.bench_latency --models decider:4b --memory \
--machine my-machine --device cuda --out latency.json