Large language models

ChatGPT Node

Use the ChatGPT node to add ChatGPT capabilities to your apps, products, or services. This node can follow instructions and provide detailed responses based on your input.

  • Model: Select from various ChatGPT models to suit your needs.

  • Messages:

    • This field is essential for the ChatGPT node. Give some context to guide the model’s answers to help get accurate and relevant responses.

    • Add at least one message. You can also include more messages or info from other nodes if needed.

When you add Messages there are Roles:

  • User: Use this when asking the model to perform a task.

  • System: Use this to give specific instructions that shape the model’s answers.

Advanced settings

  • Temperature (0.0 to 1.0): Control how random the responses are. Lower values make replies more focused; higher values make them more creative.

  • Top P (0.0 to 1.0): Set a probability limit for picking words. Lower values mean more predictable replies; higher values allow more varied words.

  • Stop: Set sequences that tell the model to stop generating further text. This helps manage the length or format of the output.

  • N (Number of completions): Ask for multiple responses to compare and pick the best one.

  • Max Tokens: Limit the number of tokens (words) in the reply to control its length. The use of tokens varies by language. In English, a token is roughly equivalent to a word, but in other languages, such as European ones, a token can be as small as a single syllable.

  • Presence Penalty (-2 to 2): Penalize tokens based on their presence so far, to avoid repetition.

  • Frequency Penalty (-2 to 2): Penalize tokens based on how often they appear, to reduce repeated words.

Play around with these settings to see what works best for you. Remember, these tweaks help you shape the model's replies to get the response you want.

The same concept can be applied to Martian LLM Router and other LLM nodes. Let’s have a look at the fields of these nodes.

Gemini node

Use the Gemini node to embed Google's multimodal AI models in different systems.

The syntax is a bit different from ChatGPT. Gemini doesn’t need you to use the messages interface.

  • Prompt: Provide a text description to guide generation.

  • History: Include history to offer context, using Messages.

  • Temperature (0.0 to 1.0): Controls how random the responses are. Lower values make replies more focused; higher values make them more creative.

  • Top P (0.0 to 1.0): Sets a probability limit for picking words. Lower values mean more predictable replies; higher values allow more varied words.

Claude node

If you’ve used ChatGPT or Google Gemini, you’ll find Claude familiar: a flexible chatbot that writes for you and answers your questions.

  • Model: Select from various Claude versions.

  • Prompt: Provide a text description to guide generation.

  • Messages: Add your text prompt in the Messages section to provide context for Claude’s response.

  • Max Tokens To Sample: Limit the number of tokens (words or its parts) Claude can use in its reply to control the length.

  • Temperature (0.0 to 1.0): Control how random the responses are. Lower values make replies more focused; higher values make them more creative.

  • Top P (0.0 to 1.0): Set a probability limit for picking words. Lower values mean more predictable replies; higher values allow more varied words.

  • Top K: Limit choices to the top K most likely words. Pick a number (e.g., 40, 50) to restrict Claude to the most probable word.

  • Metadata User ID: Enter your user ID to track your requests.

By following these steps, you can effectively interact with these AI models and tailor their responses to your needs.

Perplexity node

Perplexity gives you easy-to-understand and reliable answers.

  • Model: Select the Perplexity version you want to use.

  • Messages: Write your question or prompt in the message. This gives the model the context it needs to generate a response

  • Max Tokens: Limit the number of tokens (words or its parts) in the output to control its length.

  • Temperature (0.0 to 1.0): Control how random the responses are. Lower values make replies more focused and predictable; higher values make them more creative and varied.

  • Top P (0.0 to 1.0): Set a probability limit for sampling tokens. Lower values mean more predictable replies; higher values allow more varied words.

  • Top K: Limit choices to the top K most likely words. Pick a number (e.g., 40, 50) to restrict the model to the most probable tokens.

  • Return Citations: Choose if you want citations in the response to include references to sources.

  • Return Images: Decide whether to include relevant visuals in the response.

  • Presence Penalty (-2 to 2): Penalize tokens based on their presence so far to avoid repetition.

Frequency Penalty (-2 to 2): Penalize tokens based on their frequency in the text so far to reduce repeated words.

Last updated