pplx-decider:flash
pplx-decider · Perplexity AI · 4B · ONNX Runtime · Apache-2.0
- Score
- 73.7
- 95 % interval 72.0–75.2
- Rank
- 1
- of 41 scored models
- Public benchmark
- 82.4
- macro accuracy, %
- typed-decisions
- 72.6
- 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 | 82.4(81.2–83.7) | 88.7 | 90.3 | 58.9 | 0.016 | 0.224 | 0 / 3,880 |
| typed-decisions | 72.6(70.5–74.7) | 69.8 | 82.5 | 67.3 | 0.031 | 0.371 | 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 | 127 ms | 131 ms | 22.0 s | 10.4 GB | 1.9 GB | F32 |
| Threadripper 3960X CPU | 5.53 s | 5.59 s | 7.04 s | – | 8.2 GB | F32 |
pplx-decider:flash-fp16 RTX 5090 | 89.9 ms | 95.0 ms | 7.39 s | 10.3 GB | 1.9 GB | F16 |
pplx-decider:flash-fp16 is the same model computing in fp16 on the GPU, not ranked separately. Its decisions agree with pplx-decider:flash's on at least 99.5 % of the parity questions; typed-decisions 72.6 against 72.6, measured with Ollaya 0.14.0.
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 CUDA | 696/696 | 8.2e-5 | pass | results/runs/2026-10-19-parity-rtx4090-cuda.json |
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 pplx-decider:flash
# 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 pplx-decider:flash \
--url http://127.0.0.1:11435 --data typed.jsonl --out typed/pplx-decider-flash
# Bespoke Labs' public benchmark; build its 13 subsets first
NIMBLE_SRC=nimble python -m ollaya_convert.bench_public run --model pplx-decider:flash \
--url http://127.0.0.1:11435 --data public --out public/pplx-decider-flash
# speed and memory
python -m ollaya_convert.bench_latency --models pplx-decider:flash --memory \
--machine my-machine --device cuda --out latency.json