Launch a local open-source LLM like Qwen or Gemma by dropping a pre-made YAML recipe into your models folder.
Switch between different local models by toggling recipes in lmswitch instead of writing new configs each time.
Run large, GPU-accelerated models through vLLM in Docker using ready-made settings for memory and batching.
Run smaller GGUF models locally with llama.cpp without needing any Docker setup.
| jvr0x/ai-models | 0verflowme/alarm-clock | 0xhassaan/nn-from-scratch | |
|---|---|---|---|
| Stars | 0 | — | 0 |
| Language | — | CSS | Python |
| Last pushed | — | 2022-10-03 | — |
| Maintenance | — | Dormant | — |
| Setup difficulty | hard | easy | moderate |
| Complexity | 3/5 | 2/5 | 4/5 |
| Audience | developer | vibe coder | developer |
Figures from each repo's GitHub metadata at analysis time.
Requires the separate lmswitch tool, a GPU for most recipes, and Docker for the vLLM-based configs.
This repository is a collection of ready made YAML configuration files, called recipes, for a companion tool called lmswitch, which is a local LLM launcher. Instead of writing your own configuration by hand to run a local AI model, you can pick one of these files, drop it into the ai-models folder, and lmswitch will detect it and be ready to serve that model right away. Each recipe targets one of two runtimes. The first is llama, which uses llama.cpp to run models saved in the GGUF file format and needs no Docker. The second is vllm, which uses the vLLM project to run models in formats like safetensors, FP8, or NVFP4, and does require Docker. The repository includes dozens of recipes covering many different open models and model families, along with quantized and distilled variants, some models tuned for coding, and a few that support both text and images. A recipe file is a simple YAML document with settings such as which runtime to use, the path to the model file, which network port to serve it on, how much context length to allow, a human readable display name, and how long to wait for the model to become ready. Depending on which runtime you choose, there are additional settings: llama recipes let you control GPU layer offloading, thread count, and batch sizes, while vllm recipes let you set GPU memory usage limits, choose a Docker image, and enable features like tool calling or reasoning support for models that support them. To use the recipes, you first clone the lmswitch tool itself, since this repository is meant to be used as a submodule inside it, install lmswitch, copy one of the provided YAML files, edit a few fields like the model path and port, and then launch lmswitch and select the model you want to turn on. The project also welcomes contributions of new recipes through a fork and pull request workflow, and asks contributors to note the model's source and quantization details. It is released under the MIT license.
A collection of ready-made YAML configuration recipes for lmswitch, a tool that launches and switches between local LLMs.
Use freely for any purpose, including commercial use, as long as you keep the copyright notice.
Setup difficulty is rated hard, with roughly 1h+ to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.