AEON RYS 15/20 Full Release Record
The comprehensive non-finetuned model record: source choice, layer-window selection, quantization results, runtime decisions, rejected variants, and caveats.
Documentation
Release notes, runtime guidance, and evidence pages for the AEON RYS 15/20 GGUF line, the SignalLatch behavior fine-tune, and the PatchCode agentic-coder fine-tune.
AEON RYS 15/20 + SignalLatch Release Page
A comprehensive public webpage covering the model line, runtime profile, evidence, known tradeoffs, and FAQ.
RYS Layer Duplication Guide
The detailed build guide for the RYS checkpoint transformation, tensor mapping, layer metadata, and validation process.
SignalLatch Fine-Tune Record
The historical first-release fine-tune report: why ckpt386 was trained, how it was merged, what tests were run, and why s0.10 became the original SignalLatch baseline.
SignalLatch v1.1 Production Quantization Record
The production-selection record for the v1.1 Lite and Plus mixed-quant GGUFs: hard-suite runs, real-world ClawCode harnesses, baseline comparisons, and release decisions.
SignalLatch v1.1 All-Results Supplement
The heavy appendix: every summarized candidate row, large score graphs, real-world harness rows, and recipe annotations for tested combinations.
PatchCode (Agentic-Coder) Fine-Tune Record
The agentic-coder behaviour distil on SignalLatch: exact dataset pipeline (synthetic coding-agent generator + curated action-first slice), the lambda=0.5 merge, the 5-phase quant bake-off, and why plain IQ4_NL shipped over BF16 / Q8 / mixed recipes.