laya-typed-decisions-smoketest

Laya (Fine-Tuned on Typed-Decisions Benchmark)

This is a smoke test, not a full benchmark run. It was fine-tuned on 40 training examples (1 epoch) and evaluated on just 20 cases (100 decisions) from a single workflow category — Agent Trace Observability — not the official 400-case/2,000-decision test set. On that small slice, it achieves 0.460 Accuracy. For reference, the full published benchmark reports TypeSafe Jev 1.13.0 at 0.727 accuracy and a Teacher Self-Agreement ceiling of 0.735 — but those numbers aren't directly comparable to this run given the difference in scale. Treat these results as a pipeline sanity check, not a real evaluation of Laya's capabilities.

Head-to-Head Benchmark Results

Installation & Quickstart

pip install laya

import laya

# Load the fine-tuned model directly from Hugging Face
agent = laya.load("convaiinnovations/laya-typed-decisions")

# Evaluate any workflow state and typed questions in a single forward pass
result = agent.predict(state, questions)
print(result["answers"])

License

Apache 2.0. Developed by Convai Innovations.