Datasets:

LocalLLaMA
/

typed-decisions

Model: tasksource/tasksource-decider-nano (tag v5), a 149M cross-encoder (Apache-2.0) that scores every option of a decision and returns a distribution over them.
Training: a general decision model; this benchmark's train split is 0.5% of its training mix, hence this table. No extra fitting on train, nothing from test.
Scoring: full test split (2,000 decisions), one forward pass per request.

Accuracy	KL from gold	Brier	ECE	p50 latency
0.7515	0.1125	0.0621	0.125	37 ms (A30, shared)

By type: choice 0.712, noul 0.833, score 0.720.
By workflow: agent_trace_observability 0.692, customer_service 0.790, invoice_processing 0.802, security_incidents 0.722.

codelion

LocalLLaMA org 2 days ago

Thanks for the submission. tasksource-decider-nano v5 is now on the card in the fitted / fine-tuned table at #19: 0.7515 accuracy, KL 0.1125, Brier 0.0621, ECE 0.125, 37 ms on a shared A30. The Kind column notes that the train split is 0.5% of its training mix.

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment