Skip to content

decision2:sol

Decision 2.0 · vLLM Semantic Router · 1.88B · ONNX Runtime · Apache-2.0

Score
50.0
95 % interval 47.8–52.2
Rank
31
of 66 scored models
Public benchmark
67.8
macro accuracy, %
typed-decisions
55.5
accuracy, %

By task

Chance-corrected accuracy on each of the 14 tasks: 0 is guessing, 100 is every answer right.

Workflows
typed-decisions
34.7
Classification
MASSIVE en
75.5
MASSIVE de
74.0
Civil Comments
60.0
Aegis 2
47.2
Understanding
MultiNLI
85.5
VitaminC
61.7
PAWS
27.2
Knowledge
BoolQ
54.7
SQuAD 2
37.1
PubMedQA
43.6
Judging
HelpSteer 2
24.2
SummEval rel.
7.3
SummEval cons.
67.0

Accuracy and calibration

BenchmarkAccuracyChoiceNoulScoreECEBrierRejected
Public benchmark67.8(66.3–69.2)75.972.646.30.0530.4120 / 3,880
typed-decisions55.5(53.0–57.7)49.567.351.00.0570.5540 / 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

Devicep50p90LoadVRAMRAM peakPrecision
RTX 5090 CUDA89.4 ms104 ms44.4 s10.8 GB11.4 GBF32
Threadripper 3960X CPU4.64 s4.74 s44.6 s–11.2 GBF32

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

DeviceSame decisionMax logit diff.ResultRecord
RTX 5090 CPU463/4638.4e-6passresults/runs/2026-10-25-parity-rtx5090-cpu-cob561.json
RTX 5090 CUDA463/4631.0e-5passresults/runs/2026-10-25-parity-rtx5090-cuda-cob561.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 decision2:sol

# 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 decision2:sol \
  --url http://127.0.0.1:11435 --data typed.jsonl --out typed/decision2-sol

# Bespoke Labs' public benchmark; build its 13 subsets first
NIMBLE_SRC=nimble python -m ollaya_convert.bench_public run --model decision2:sol \
  --url http://127.0.0.1:11435 --data public --out public/decision2-sol

# speed and memory
python -m ollaya_convert.bench_latency --models decision2:sol --memory \
  --machine my-machine --device cuda --out latency.json