decima:agent
Decima · A. M. Madani · 321M · ONNX Runtime · Apache-2.0
- Score
- 27.9
- 95 % interval 25.7–29.9
- Rank
- 25
- of 35 scored models
- Public benchmark
- 53.3
- macro accuracy, %
- typed-decisions
- 46.1
- accuracy, %
By task
Chance-corrected accuracy on each of the 14 tasks: 0 is guessing, 100 is every answer right.
Accuracy and calibration
| Benchmark | Accuracy | Choice | Noul | Score | ECE | Brier | Rejected |
|---|---|---|---|---|---|---|---|
| Public benchmark | 53.3(51.8–54.7) | 68.3 | 60.7 | 15.7 | 0.211 | 0.622 | 0 / 3,880 |
| typed-decisions | 46.1(43.5–48.5) | 45.0 | 53.3 | 41.4 | 0.193 | 0.689 | 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 | 27.1 ms | 27.6 ms | 5.93 s | 2.7 GB | 1.8 GB | F32 |
| Threadripper 3960X CPU | 1.23 s | 1.27 s | 4.16 s | – | 1.1 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
Its parity check is in the model’s family documentation in the repository (docs/families/); the compiled parity file does not list it yet.
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 decima:agent
# 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 decima:agent \
--url http://127.0.0.1:11435 --data typed.jsonl --out typed/decima-agent
# Bespoke Labs' public benchmark; build its 13 subsets first
NIMBLE_SRC=nimble python -m ollaya_convert.bench_public run --model decima:agent \
--url http://127.0.0.1:11435 --data public --out public/decima-agent
# speed and memory
python -m ollaya_convert.bench_latency --models decima:agent --memory \
--machine my-machine --device cuda --out latency.json