Structured data exercise

Controlling LLM Output with Stop Sequences and Message Prefilling

• Purpose of Control Mechanisms: Message prefilling and stop sequences are used to precisely guide the LLM's response, ensuring the output meets specific structural requirements (e. g. , no commentary, single block).
• Using Stop Sequences:
• A stop sequence (e. g. , three backticks: `) tells the LLM when to stop generating text.
• This is effective for preventing the LLM from adding unnecessary headers or conversational commentary.
• Markdown Code Blocks and Language Identifiers:
• When using markdown code blocks (e. g. , ), the LLM may optionally include a language identifier (e. g. , bash`) for syntax highlighting.
• If the LLM adds an unwanted identifier, the prefill message can be adjusted to explicitly include the desired identifier (e. g. , bash `) to guide the LLM.
• Advanced Message Prefilling (Structural Guidance):
• Message prefilling is not limited to character sequences (like backticks).
• It can be used to provide complex, structural instructions to the LLM.
• Goal: To ensure the LLM provides all required items in a single, clean block without internal explanatory comments.
• Example: Prefilling with a directive like "Here are all three commands in a single block without any comments. " leads to more reliable output.

Takeaways

• Stop sequences and message prefilling are used to precisely guide the LLM's response, ensuring the output meets specific structural requirements.
• Stop sequences (e. g. , `) instruct the LLM when to stop generating text, preventing unnecessary headers or conversational commentary.
• Message prefilling can be used for structural guidance to ensure the LLM provides all required items in a single, clean block.
• When using markdown code blocks, prefilling can be adjusted to explicitly include desired language identifiers (e. g. , bash `).