peonist-ai/crucible is a collection of tools that prunes the experts of MoE models (mixture of experts, a structure that computes only some specialized experts depending on the input) and merges them into a model for local inference. GitHub commit 1aa0d14 (2026-10-28 17:08:52 UTC) is dependency update PR #19 raised by dependabot, which raised transformers from 5.18.0 to 5.19.0 and ruff (a Python linter) from 0.16.10 to 0.17.0.
The changed files are three: packages/crucible/pyproject.toml, packages/crucible-bench/pyproject.toml, and uv.lock, and the change amount is 28 lines added and 28 lines deleted. The main package's minimum transformers version rose from 5.17.0 to 5.19.0, and crucible-bench's minimum ruff version for development rose from 0.16.8 to 0.17.0. The same pyproject files still carry a comment stating that transformers 5.5 or higher is required and that 4.x has no MoE modeling code for Gemma 4 and Qwen 3.5/3.6. In uv.lock, four additional dependency groups — ifeval, bfcl, local, dev — are registered.
What this change actually touches is the minimum required version of the runtime environment. A fresh install requires transformers 5.19.0 or higher, and since this repository already requires 5.x, environments using 4.x have to be migrated along with it. That said, this commit contains no figures on the quality or speed of the compressed model, and no license, star count, or release version of crucible itself is given. How much transformers 5.19.0 changed the MoE modeling code can only be known by checking the release notes, so for now it is more reasonable to view this as a change that raised the dependency floor rather than one that added functionality.