jevk5:4brejects 12 %
JevK5 · alibiserikbay · 4B · llama.cpp · Apache-2.0
- Score
- 48.6
- 95 % interval 46.6–50.7
- Rank
- 19
- of 35 scored models
- Public benchmark
- 62.1
- macro accuracy, %
- typed-decisions
- 62.9
- accuracy, %
By task
Chance-corrected accuracy on each of the 14 tasks: 0 is guessing, 100 is every answer right.
Workflows
typed-decisions
45.6
Classification
MASSIVE en
-5.9
MASSIVE de
-5.9
Civil Comments
66.7
Aegis 2
60.0
Understanding
MultiNLI
76.4
VitaminC
52.4
PAWS
61.6
Knowledge
BoolQ
69.3
SQuAD 2
59.9
PubMedQA
61.6
Judging
HelpSteer 2
27.2
SummEval rel.
29.7
SummEval cons.
81.8
Accuracy and calibration
| Benchmark | Accuracy | Choice | Noul | Score | ECE | Brier | Rejected |
|---|---|---|---|---|---|---|---|
| Public benchmark | 62.1(60.7–63.2) | 45.4 | 81.8 | 57.0 | 0.040 | 0.373 | 700 / 3,880 |
| typed-decisions | 62.9(60.4–65.1) | 57.2 | 74.7 | 58.4 | 0.089 | 0.490 | 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 | 95.3 ms | 97.5 ms | 10.4 s | 5.8 GB | 4.6 GB | Q8_0 |
| Threadripper 3960X CPU | 6.98 s | 7.27 s | 11.3 s | – | 4.9 GB | Q8_0 |
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 | 593/593 | 7.7e-6 | pass | families/jevk5.md:105 |
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 jevk5:4b
# 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 jevk5:4b \
--url http://127.0.0.1:11435 --data typed.jsonl --out typed/jevk5-4b
# Bespoke Labs' public benchmark; build its 13 subsets first
NIMBLE_SRC=nimble python -m ollaya_convert.bench_public run --model jevk5:4b \
--url http://127.0.0.1:11435 --data public --out public/jevk5-4b
# speed and memory
python -m ollaya_convert.bench_latency --models jevk5:4b --memory \
--machine my-machine --device cuda --out latency.json