Model Serving Minefield

Judgement

What it is

List, course or other material

What it does

A community registry of bugs in LLM serving setups (chat templates, tool parsers, quantization kernels, eval harnesses) that produce wrong benchmark numbers, each listed by symptom with a check and a fix.

Adopt it when

You run or benchmark local LLMs with vLLM, llama.cpp or similar and a result looks off in a way you cannot explain.

Watch out

Entries carry different evidence levels, and only some are reproduced by the maintainers, so read each entry's status before trusting it.

Ready for

Usable

Built with

Python, Jinja and Shell

How established

New: started 3 months ago, latest release v0.2.2.

What the license allows

You can build it into a product you sell, as long as you keep its copyright notice.

Judged on 2026-11-06 from the project's README and GitHub data. The license and age lines are worked out directly.

Built from

Follow it down: Model Serving Minefield → CPython → GCC → LLVM project → runs on a microchip

How compilers were built, back to punched paper tape

Open-source alternatives to Model Serving Minefield

1. 1
   Ollama ollama

   Get up and running with Kimi, GLM, MiniMax, DeepSeek, gpt-oss, Qwen, Gemma and other models.

   Developer tool Production-ready You want to run local LLMs for development, privacy or offline use with minimal setup.

   ★ 186k MIT Go Active, pushed today
   
2. 2
   Llama.cpp ggml-org

   LLM inference in C/C++

   Building block Production-ready You need efficient local or edge LLM inference and want the engine many other tools are built on.

   ★ 133k MIT C++ Active, pushed today
   
3. 3
   Whisper openai

   Robust Speech Recognition via Large-Scale Weak Supervision

   Building block Production-ready You need accurate offline transcription and want the reference implementation of the models.

   ★ 112k MIT Python Active, pushed 2 months ago
4. 4
   vLLM vllm-project

   A high-throughput and memory-efficient inference and serving engine for LLMs

   Building block Production-ready You serve open-weight models to many users on your own GPUs and need throughput.

   ★ 96k Apache-2.0 Python Active, pushed today
   
5. 5
   AnythingLLM Mintplex-Labs

   Stop renting your intelligence. Own it with AnythingLLM. Everything you need for a powerful local-first agent experience

   Complete system Usable You want private document chat for a small team without building a RAG stack yourself.

   ★ 68k MIT JavaScript Active, pushed yesterday
   
6. 6
   whisper.cpp ggml-org

   Port of OpenAI's Whisper model in C/C++

   Building block Production-ready You need transcription embedded in a native, mobile or edge application without Python.

   ★ 55k MIT C++ Active, pushed 6 days ago
7. 7
   LocalAI mudler

   LocalAI is the open-source AI engine. Run any model - LLMs, vision, voice, image, video - on any hardware. No GPU required.

   Complete system Usable You want a drop-in local replacement for OpenAI's API covering more than just text.

   ★ 50k MIT Go Active, pushed today
8. 8
   faster-whisper SYSTRAN

   Faster Whisper transcription with CTranslate2

   Building block Production-ready You run Whisper transcription in Python at volume and want lower cost and latency without changing models.

   ★ 26k MIT Python Active, pushed 3 days ago

Commercial tools for running AI models locally

1. 1
   LM Studio Element Labs

   Bionic is LM Studio's agent for work and code. Create documents, slides, PDFs, and software with local or frontier open models.

   Complete system Usable You want to run open models privately on your own machine without command-line setup.

   Commercial