RUNTIME tutorials v1.1 — Laya and Needle

Laya and Needle guides with beginner setup, troubleshooting, synthetic examples and recorded outputs.

This supersedes needle-video-v1 for new installations: the Laya dependency pins now use PyTorch 2.13.0 and Transformers 5.10.0 after an advisory review. Laya's CPU examples passed in a fresh Python 3.12.3 / Linux ARM64 environment. The optional LLM integration was not rerun with those packages. Needle's seven cases were rerun during the original export and its files are unchanged.

Download runtime-tutorials.zip, extract it, open README.md and choose one guide. Each guide uses a separate environment. Models and environments are not bundled. Existing caches/checkpoints were reused during validation; other platforms and cold downloads are not qualified.

Needle video companion + beginner tutorial library

Historical release. For new installations use tutorials-v1.1. This bundle contains older Laya dependency pins with published security advisories. Its original files are retained unchanged for reference.

Needle video companion and the RUNTIME tutorial library.

Includes Laya and Needle guides, beginner setup, glossary, troubleshooting, pinned dependencies, synthetic inputs and recorded results. The seven Needle cases were rerun in a fresh Python 3.12 environment on Linux ARM64; calls and validation warnings matched the recorded demonstration. Model/runtime download caches were reused. No device commands execute.

Needle video: https://youtu.be/FnKINBWJPTs
Laya video: https://youtu.be/J-Cn9UUJtdA

Download runtime-tutorials.zip, extract it, open README.md and choose one guide. Models and virtual environments are not bundled.

Laya video companion r4 — updated dependencies

The Laya video examples with updated dependency pins and clearer CPU installation.

Use this release for new installations. It supersedes laya-v19-r3, whose older PyTorch and Transformers pins have published security advisories. Example code, model revision and historical video outputs are unchanged.

Validated in a fresh Python 3.12.3 / Linux ARM64 environment using PyTorch 2.13.0+cpu, Transformers 5.10.0 and the existing pinned checkpoint: pip check, first example, triage, clarification and custom categories. Labels, rounded probabilities and gate decisions matched the recording. The optional local-LLM integration was not rerun; Windows/macOS and a cold checkpoint download were not tested.

Download laya-tutorial.zip, extract it, open README.md and follow the commands for your platform. Models and environments are not included.

Video: https://youtu.be/J-Cn9UUJtdA

Laya local setup: video guide and demos

Historical release. For new installations use laya-v19-r4. This bundle contains older Laya dependency pins with published security advisories. Its original files are retained unchanged for reference.

The setup guide and synthetic examples accompanying RUNTIME’s Laya tutorial.

Watch: https://youtu.be/J-Cn9UUJtdA

Tested with Python 3.12 on Linux ARM64 using CPU inference. Includes dependency pins, model download code, request sorting, an uncertainty gate, an optional local-model draft, and custom categories. See the README for expected results and tested limitations.

Original example code and documentation are MIT licensed; upstream models and dependencies retain their own licenses. No weights, environments, credentials or real customer data are included.

The eight tutorial files match the tested video companion. This release adds the explicit MIT license.