strands-decider:e2b
Strands Decider · Strands Agents · 4.6B · ONNX Runtime · Apache-2.0
- Score
- 58.0
- 95 % interval 56.0–60.0
- Rank
- 18
- of 66 scored models
- Public benchmark
- 71.6
- macro accuracy, %
- typed-decisions
- 64.6
- accuracy, %
By task
Chance-corrected accuracy on each of the 14 tasks: 0 is guessing, 100 is every answer right.
Workflows
typed-decisions
48.1
Classification
MASSIVE en
82.2
MASSIVE de
68.8
Civil Comments
80.7
Aegis 2
57.6
Understanding
MultiNLI
54.9
VitaminC
58.4
PAWS
86.4
Knowledge
BoolQ
60.7
SQuAD 2
51.8
PubMedQA
49.0
Judging
HelpSteer 2
14.2
SummEval rel.
27.6
SummEval cons.
71.3
Accuracy and calibration
| Benchmark | Accuracy | Choice | Noul | Score | ECE | Brier | Rejected |
|---|---|---|---|---|---|---|---|
| Public benchmark | 71.6(70.2–72.8) | 72.4 | 83.7 | 50.2 | 0.065 | 0.394 | 0 / 3,880 |
| typed-decisions | 64.6(62.3–67.0) | 64.8 | 74.0 | 57.4 | 0.086 | 0.494 | 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 | 116 ms | 131 ms | 30.5 s | 10.1 GB | 5.2 GB | F32 |
| Threadripper 3960X CPU | 5.54 s | 5.57 s | 14.3 s | – | 5.5 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 | 2.1e-5 | pass | results/runs/2026-10-24-parity-rtx4090-cpu.json |
| RTX 4090 CUDA | 650/650 | 1.8e-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:e2b
# 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:e2b \
--url http://127.0.0.1:11435 --data typed.jsonl --out typed/strands-decider-e2b
# Bespoke Labs' public benchmark; build its 13 subsets first
NIMBLE_SRC=nimble python -m ollaya_convert.bench_public run --model strands-decider:e2b \
--url http://127.0.0.1:11435 --data public --out public/strands-decider-e2b
# speed and memory
python -m ollaya_convert.bench_latency --models strands-decider:e2b --memory \
--machine my-machine --device cuda --out latency.json