strands-decider:e4b
Strands Decider · Strands Agents · 8.9B · ONNX Runtime · Apache-2.0
- Score
- 64.6
- 95 % interval 62.8–66.5
- Rank
- 3
- of 66 scored models
- Public benchmark
- 76.5
- macro accuracy, %
- typed-decisions
- 69.9
- accuracy, %
By task
Chance-corrected accuracy on each of the 14 tasks: 0 is guessing, 100 is every answer right.
Workflows
typed-decisions
55.9
Classification
MASSIVE en
85.8
MASSIVE de
83.1
Civil Comments
84.0
Aegis 2
50.4
Understanding
MultiNLI
71.9
VitaminC
64.9
PAWS
76.0
Knowledge
BoolQ
74.0
SQuAD 2
61.9
PubMedQA
61.0
Judging
HelpSteer 2
18.7
SummEval rel.
32.8
SummEval cons.
84.4
Accuracy and calibration
| Benchmark | Accuracy | Choice | Noul | Score | ECE | Brier | Rejected |
|---|---|---|---|---|---|---|---|
| Public benchmark | 76.5(75.2–77.8) | 80.5 | 84.6 | 56.2 | 0.039 | 0.325 | 0 / 3,880 |
| typed-decisions | 69.9(67.3–72.4) | 70.8 | 77.2 | 63.7 | 0.082 | 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 | 213 ms | 226 ms | 48.6 s | 15.4 GB | 6.2 GB | F32 |
| Threadripper 3960X CPU | 9.44 s | 9.49 s | 9.21 s | – | 9.7 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 CPU | 650/650 | 1.7e-5 | pass | results/runs/2026-10-24-parity-rtx4090-cpu.json |
| RTX 4090 CUDA | 650/650 | 1.7e-5 | pass | results/runs/2026-10-24-parity-rtx4090-cuda.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 strands-decider:e4b
# 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 strands-decider:e4b \
--url http://127.0.0.1:11435 --data typed.jsonl --out typed/strands-decider-e4b
# Bespoke Labs' public benchmark; build its 13 subsets first
NIMBLE_SRC=nimble python -m ollaya_convert.bench_public run --model strands-decider:e4b \
--url http://127.0.0.1:11435 --data public --out public/strands-decider-e4b
# speed and memory
python -m ollaya_convert.bench_latency --models strands-decider:e4b --memory \
--machine my-machine --device cuda --out latency.json