cygnet:12b
Cygnet · blockbrain-ai · 12B · llama.cpp · Apache-2.0
- Score
- 64.6
- 95 % interval 62.6–66.5
- Rank
- 1
- of 35 scored models
- Public benchmark
- 76.8
- macro accuracy, %
- typed-decisions
- 68.4
- accuracy, %
By task
Chance-corrected accuracy on each of the 14 tasks: 0 is guessing, 100 is every answer right.
Workflows
typed-decisions
53.7
Classification
MASSIVE en
84.9
MASSIVE de
86.4
Civil Comments
63.3
Aegis 2
63.2
Understanding
MultiNLI
74.9
VitaminC
62.2
PAWS
66.4
Knowledge
BoolQ
73.3
SQuAD 2
72.6
PubMedQA
61.0
Judging
HelpSteer 2
26.7
SummEval rel.
31.8
SummEval cons.
83.5
Accuracy and calibration
| Benchmark | Accuracy | Choice | Noul | Score | ECE | Brier | Rejected |
|---|---|---|---|---|---|---|---|
| Public benchmark | 76.8(75.5–78.0) | 81.0 | 83.9 | 57.9 | 0.103 | 0.352 | 0 / 3,880 |
| typed-decisions | 68.4(66.1–70.8) | 63.7 | 75.2 | 66.9 | 0.146 | 0.467 | 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 | 152 ms | 153 ms | 28.0 s | 18.1 GB | 12.2 GB | Q8_0 |
| Threadripper 3960X CPU | 20.7 s | 21.0 s | 41.6 s | – | 17.2 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 | 502/502 | 7.7e-6 | pass | families/cygnet.md:59 |
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 cygnet:12b
# 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 cygnet:12b \
--url http://127.0.0.1:11435 --data typed.jsonl --out typed/cygnet-12b
# Bespoke Labs' public benchmark; build its 13 subsets first
NIMBLE_SRC=nimble python -m ollaya_convert.bench_public run --model cygnet:12b \
--url http://127.0.0.1:11435 --data public --out public/cygnet-12b
# speed and memory
python -m ollaya_convert.bench_latency --models cygnet:12b --memory \
--machine my-machine --device cuda --out latency.json