nimble:9b
Nimble · Bespoke Labs · 9B · ONNX Runtime · Apache-2.0
- Score
- 61.9
- 95 % interval 59.9–63.8
- Rank
- 2
- of 35 scored models
- Public benchmark
- 74.8
- macro accuracy, %
- typed-decisions
- 66.5
- accuracy, %
By task
Chance-corrected accuracy on each of the 14 tasks: 0 is guessing, 100 is every answer right.
Workflows
typed-decisions
50.9
Classification
MASSIVE en
83.1
MASSIVE de
79.7
Civil Comments
68.7
Aegis 2
62.4
Understanding
MultiNLI
73.9
VitaminC
65.2
PAWS
60.0
Knowledge
BoolQ
70.7
SQuAD 2
64.5
PubMedQA
63.4
Judging
HelpSteer 2
17.7
SummEval rel.
25.0
SummEval cons.
80.9
Accuracy and calibration
| Benchmark | Accuracy | Choice | Noul | Score | ECE | Brier | Rejected |
|---|---|---|---|---|---|---|---|
| Public benchmark | 74.8(73.5–76.1) | 80.0 | 82.6 | 52.9 | 0.022 | 0.339 | 0 / 3,880 |
| typed-decisions | 66.5(64.3–68.7) | 57.2 | 81.0 | 62.6 | 0.083 | 0.457 | 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 | 1.69 s | 1.69 s | 39.7 s | 20.3 GB | 3.0 GB | F32 |
| Threadripper 3960X CPU | 62.0 s | 62.1 s | 41.1 s | – | 18.6 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 | 100 % | 1.1e-4 | pass | families/nimble.md:72 |
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 nimble:9b
# 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 nimble:9b \
--url http://127.0.0.1:11435 --data typed.jsonl --out typed/nimble-9b
# Bespoke Labs' public benchmark; build its 13 subsets first
NIMBLE_SRC=nimble python -m ollaya_convert.bench_public run --model nimble:9b \
--url http://127.0.0.1:11435 --data public --out public/nimble-9b
# speed and memory
python -m ollaya_convert.bench_latency --models nimble:9b --memory \
--machine my-machine --device cuda --out latency.json