Skip to content

v4.1.1: a change of sampling parameters keeps the seed

Choose a tag to compare

@lloyal-research lloyal-research released this 20 Oct 15:17
· 22 commits to main since this release

A change of sampling parameters keeps the seed

lloyal.node 4.1.1 carries liblloyal v3.1.1. A branch's sampler no longer reseeds itself when its parameters change without a seed: a seedless setSamplerParams with nothing else changed keeps the chain as it is, its penalty window and its position in the random sequence included, and a rebuild draws on the seed the chain already had rather than a fresh clock seed. A branch created without a seed is seeded once from the clock and keeps that seed for its life.

Packages

Thirteen platform packages, each built and tested on its own runner: macOS arm64 and x64; Linux arm64 and x64, with CUDA and Vulkan variants; Windows arm64 and x64, with CUDA and Vulkan variants. The Linux x64 CUDA flavor also ships as a signed download manifest for CUDA 12.9. Every package passed the L4 GPU gate before publishing.

The warm multi-turn recall case now runs on Qwen3.5-4B as well as SmolLM2, with thinking off and each answer bounded, so a hybrid model's warm continuation is covered on every platform.

Ships with lloyal-ai 1.20.0, alongside liblloyal v3.1.1 and hdk v6.1.1-rig.

What's Changed

  • test: the warm recall test runs with thinking off and bounded, on SmolLM2 and Qwen3.5-4B by @lloyal-research in #61

Full Changelog: v4.1.0...v4.1.1