jeeves:9b
Jeeves · PostHog · 9B · ONNX Runtime · Apache-2.0
- Score
- 61.1
- 95 % interval 59.1–63.0
- Rank
- 4
- of 35 scored models
- Public benchmark
- 74.4
- 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
53.2
Classification
MASSIVE en
85.8
MASSIVE de
84.3
Civil Comments
58.0
Aegis 2
54.4
Understanding
MultiNLI
88.0
VitaminC
44.9
PAWS
64.0
Knowledge
BoolQ
74.0
SQuAD 2
65.9
PubMedQA
57.4
Judging
HelpSteer 2
20.2
SummEval rel.
25.5
SummEval cons.
80.0
Accuracy and calibration
| Benchmark | Accuracy | Choice | Noul | Score | ECE | Brier | Rejected |
|---|---|---|---|---|---|---|---|
| Public benchmark | 74.4(73.2–75.8) | 79.7 | 81.6 | 53.5 | 0.049 | 0.360 | 0 / 3,880 |
| typed-decisions | 68.0(65.7–70.5) | 65.7 | 80.8 | 60.3 | 0.028 | 0.429 | 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 | 641 ms | 643 ms | 35.4 s | 18.1 GB | 2.7 GB | F32 |
| Threadripper 3960X CPU | 20.5 s | 20.6 s | 34.9 s | – | 16.5 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 | 100 % | 1.9e-4 | pass | families/jeeves.md:61 |
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 jeeves: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 jeeves:9b \
--url http://127.0.0.1:11435 --data typed.jsonl --out typed/jeeves-9b
# Bespoke Labs' public benchmark; build its 13 subsets first
NIMBLE_SRC=nimble python -m ollaya_convert.bench_public run --model jeeves:9b \
--url http://127.0.0.1:11435 --data public --out public/jeeves-9b
# speed and memory
python -m ollaya_convert.bench_latency --models jeeves:9b --memory \
--machine my-machine --device cuda --out latency.json