Laya Task Tags v2
Area-conditioned tagging specialist fine-tuned from convaiinnovations/laya. After an upstream area classifier chooses home, work, or personal, this checkpoint evaluates the ten tags defined for that area.
This checkpoint was trained on the tags configuration of chand1012/laya-task-routing-synthetic-v2.
Evaluation
On the held-out synthetic v2 test split (182 records, 1,820 tag decisions):
| Metric | Result |
|---|---|
| Decision accuracy | 94.40% |
| Micro-F1 | 85.99% |
| Supported-tag macro-F1 | 85.69% |
| Exact tag-set accuracy | 58.24% |
See evaluation_report.json for per-area and per-tag results, calibration, and latency. These are synthetic-set results and should not be interpreted as production performance.
Usage
import laya
agent = laya.load("chand1012/laya-task-tags-v2")
result = agent.predict(state_with_area, area_specific_tag_questions)
The selected area must be supplied in the state, and the question set must contain that area's tags. The exact taxonomy is documented in the dataset repository.
Limitations
The model assumes the upstream area is correct, so routing errors can propagate. Training and evaluation data are synthetic and English-only, with ten fixed tags per area. Validate thresholds and labels on real application data before deployment.
License
Apache License 2.0. This is a derivative of the Apache-2.0-licensed Laya base model.