yarqa-attn has moved to the kernel repository

This model repository holds no loadable build. The kernels library (0.16 and later) loads kernels only from the kernel repository:

https://huggingface.co/kernels/SZLHOLDINGS/YARQA-ATTN

# pip install "kernels>=0.17.2" torch
from kernels import get_kernel

kernel = get_kernel("SZLHOLDINGS/YARQA-ATTN", version=1, trust_remote_code=True)

The source of truth is szl-holdings/YARQA-ATTN. Its hf-kernel-publish workflow builds the kernel, verifies it through get_kernel, and then updates this pointer. Commits of this model repository are not kernel revisions.

What else is in this repository

• yarqa_attn.npz is a small SYNTHETIC attention-gate silhouette (route vs refuse), described by TRAINING_RECEIPT.json next to it. It is not loaded by the kernel, is not a trained production model, and carries no benchmark or capability claim. The kernel itself ships no weights.
• bom/model-bom.cdx.json, LICENSE and og-card.png are kept as uploaded.