Extensibility is the value gained when an LLM can connect to existing systems or be adapted beyond its core prompting capabilities. It matters when selecting a model for a specific workflow, not merely when comparing response quality. section 2.5.1
Extension options differ by platform and training approach:
- OpenAI provides paid API access and instruction-based output adjustment, but does not offer private fine-tuning and deployment of its GPT models. section 2.5.1
- Meta’s open-sourced Llama models can be downloaded and trained for particular requirements, while deployment infrastructure must be built separately. section 2.5.1
- Evaluate integration and extension features alongside response quality, training data, and parameter count. section 2.5.1
Two extension models
A team choosing between hosted GPT access and downloadable Llama models must weigh managed API integration against control over further training and responsibility for deployment infrastructure. section 2.5.1
