A large language model (LLM) is a generative neural language model trained on extensive text data to create or transform natural-language content. Its behavior combines training data, model architecture, scale, and training for a specific use case. GPT is the most successful architecture associated with LLMs, while fine-tuning adapts a model toward tasks such as chat, completion, or instruction following. section 2.0
Figure 2.1 The difference between generative and predictive models
Contrasts generative models, which create content from input, with predictive or classification models, which classify input. section 2.0
Use an LLM through several practical paths:
- Connect applications through the OpenAI API SDK, including access to models such as GPT-4. section 2.6
- Run open-source models locally with tools such as LM Studio, while managing hardware and configuration requirements. section 2.6
- Improve responses with prompt engineering and select models by performance, parameters, use case, and training input. section 2.6
- Configure an LLM connection in prompt flow by installing its packages, creating a secure connection, and testing a flow. section 9.3.1
