Documentation
Documentation
Guides for installing, running, configuring and extending DynaLM. Pages on this site are being prepared; until then, each entry links to the current guide in the repository where one exists.
Documentation menu
Getting started
Running models
- Supported models Model families, file formats and quantization types, with known gaps. Being prepared
- Backends CPU and GPU backends, their status, and how one is selected. Being prepared
- Configuration config.yaml, environment variables and command-line precedence. Being prepared
- OpenAI-compatible API Endpoints, streaming, authentication and client setup. Being prepared
Execution
- CPU execution SIMD paths, threading, execution modes and AutoTune. Being prepared
- GPU execution Metal, Vulkan and CUDA: setup, model coverage and fallback. Being prepared
- Scheduling and batching How requests are admitted, batched and prefilled. Being prepared
- KV cache Paged allocation, prefix sharing and cache quantization. Being prepared