Goethe 0.1.0

What's Changed

• Add nvfp4-kernels: Qwen3.8-27B NVFP4 on one RTX 5090 by @peti12352 in #2
• Unify repository as four-entry model catalog by @peti12352 in #3
• nvfp4-kernels: faster Qwen3.8-27B decode (tile patch and draft shortlist patch) by @peti12352 in #4
• goethe command, phase 1: serve, list, doctor by @peti12352 in #5
• goethe: call the presets profiles, and stop on a busy port by @peti12352 in #6
• goethe serve: fit the context to the memory the engine has by @peti12352 in #7
• Point a coding agent at the running server by @peti12352 in #8
• Drop the unused GPU power limit by @peti12352 in #9
• Report whether the install still matches the pin by @peti12352 in #10
• Name missing engine libraries by @peti12352 in #11
• Keep the 5090 checklist out of the repo by @peti12352 in #12
• Replay the agent traces by @peti12352 in #14
• Flip the engine link back by @peti12352 in #13
• Weight bench decode by tokens by @peti12352 in #15
• Treat a CUDA out-of-memory boot as a failed fit by @peti12352 in #16
• Record the measured 5090 spikes by @peti12352 in #17
• Drop the machine name from the spike note by @peti12352 in #18
• Accept OpenAI tool names up to 128 characters by @peti12352 in #19
• Use an unpacked ninfer-serve when it is already on disk by @peti12352 in #20
• Serve any row in models.toml by @peti12352 in #21
• Add mesh inference engine (mesh, meshd, mesh-core) by @JanosMozer in #22
• Goethe by @JanosMozer in #23
• CI/CD: PR checks for main and signed releases from main by @JanosMozer in #24
• Fix stage6_integrity on CI runners by @JanosMozer in #25
• Fix flaky unit tests that failed CI on main by @JanosMozer in #27
• Release: host the package on GitHub Pages by @JanosMozer in #26

New Contributors

• @peti12352 made their first contribution in #2
• @JanosMozer made their first contribution in #22

Full Changelog: https://github.com/meshapplied/goethe/commits/v0.1.0