Playing with Open Source LLMs
9 months ago
- #AI
- #OpenSource
- #Coding
- The author explores open-source LLMs for coding, preferring local tools over vendor-locked solutions.
- Evaluated models: deepseek-r1:8b (reasoning model with accuracy issues), mistral:7b (fast but inaccurate), qwen3:8b (balanced performance).
- Setup involves Ollama for local LLM API and aider for pair programming, guiding AI through coding tasks.
- Testing scenarios: refactoring (partial success), greenfield project (failed), troubleshooting (effective for error explanation).
- Qwen CLI attempted but failed due to context size limitations and file writing issues.