Skip to content

Base Model Notes

Default

  • Qwen2.5-7B-Instruct: best default for local technical reasoning and practical LoRA iteration on Apple Silicon.

Alternatives

  • Qwen2.5-14B-Instruct: better depth, slower locally, more suitable for remote adapter jobs.
  • Llama 3.1 8B Instruct: broader ecosystem compatibility, slightly weaker default fit for this repository.
  • Mistral 7B Instruct: lighter and fast, but not the primary recommendation for cross-domain professional synthesis.
  • Phi-3: useful for smaller-footprint experiments, not the default for broad enterprise knowledge.