https://locollm.org/docs/https://locollm.org/docs/adapter-guide/https://locollm.org/docs/adapter-primer/https://locollm.org/docs/adr/https://locollm.org/docs/adr/0001-base-model-qwen3-4b/https://locollm.org/docs/adr/0002-adapter-registry-design/https://locollm.org/docs/adr/0003-single-evolving-router/https://locollm.org/docs/adr/0004-retire-cerebro-adopt-b250-multi-gpu/https://locollm.org/docs/adr/0005-weiho-8gpu-replace-b250-colmena/https://locollm.org/docs/adr/0006-gguf-ollama-inference-standard/https://locollm.org/docs/architecture-vision/https://locollm.org/docs/architecture/https://locollm.org/docs/base-model-selection/https://locollm.org/docs/benchmarking-guide/https://locollm.org/docs/benchmarks/https://locollm.org/docs/evaluation-standards/https://locollm.org/docs/faq/https://locollm.org/docs/ideas/https://locollm.org/docs/known-challenges/https://locollm.org/docs/project-ideas/https://locollm.org/docs/research-roadmap/https://locollm.org/docs/small-model-strategies/https://locollm.org/docs/train-math-adapter/https://locollm.org/docs/training-new-adapters/https://locollm.org/docs/why-locollm/