Why You Should Stop Using Ollama

Ollama gained popularity by wrapping llama.cpp for easy LLM usage but has since obscured its reliance on llama.cpp, misled users, and drifted from its local-first mission. Here’s why you should stop using it:

• Reliance on llama.cpp: Ollama’s entire inference capability comes from llama.cpp, yet it failed to credit it properly for over a year, omitting it from READMEs and not including the required MIT license notice.
• Misleading Model Naming: Ollama labeled distilled versions of models as full versions, misleading users about what they were running.
• Inferior Custom Backend: After distancing from llama.cpp, Ollama built a custom backend on ggml, reintroducing bugs and performing worse than llama.cpp, with benchmarks showing it’s significantly slower.
• Community Criticism: The local LLM community has long criticized Ollama for downplaying its reliance on llama.cpp and producing an inferior product after trying to go independent.