LLM Parameters

Understanding LLM Parameters in GTWY

Fine-tune your AI with precision.

LLMs may feel magical, but the real power comes from controlling how they respond.
GTWY gives you full control over the core parameters that shape creativity, tone, structure, and logic.
Small tweaks → huge differences in output.

Here’s the simplest breakdown of every key parameter.

1. Creativity Level (Temperature)

This is the AI’s creativity dial.

• Low temperature (0.1–0.3):
  Logical, precise, predictable.
  Use it for factual answers, technical help, or support workflows.

• High temperature (0.7–1.0):
  Creative, expressive, imaginative.
  Use it for storytelling, brainstorming, branding, and content creation.

Lower = accuracy
Higher = creativity

2. Max Tokens Limit (Maximum Tokens)

Controls how long the AI is allowed to respond.

• Low limit: short, crisp answers

• High limit: detailed explanations

Use it to choose between summaries or deep dives.

3. Probability Cutoff (Top P / Nucleus Sampling)

Controls diversity in the AI’s response.

• Lower Top-P (0.4–0.6): safer, predictable

• Higher Top-P (0.8–1.0): more variety and interesting outputs

It defines how wide the model’s "idea search space" should be.

4. Log Probability

Returns the confidence score for every token generated.

Useful for developers, researchers, or debugging.
If you don’t need analytics, keep this off.

5. Repetition Penalty (Frequency Penalty)

Prevents the AI from repeating the same words or phrases.

Higher values = more variation in wording.
Perfect for long answers or creative writing.

Example:
Without this penalty, the model might repeat a sentence like:

“This is very important. This is very important.”
With a frequency penalty, it learns to rephrase or move forward instead.

6. Novelty Penalty (Presence Penalty)

Encourages the AI to introduce new ideas instead of reusing previous ones.

Use it when you want fresh, original responses instead of looping or repeating concepts.

7. Response Count (Number of Completions)

Sometimes you want multiple versions of a response — for comparison or variety.
This parameter tells the model how many different completions to generate for one prompt.

For example:

• 1 completion: Get a single response (default).

• 3 completions: Get three unique responses to choose from.

8. Tool Choice

Tells the AI which external tool to use — any tool connected to your GTWY agent.

Examples:

• Code execution

• Database lookup

• External API fetch

This ensures the model uses the right tool at the right time.

9. Response Type (Response Format)

This defines the format of the output.

• Text: Default mode — gives natural language answers.

• JSON: For structured responses — useful in apps, dashboards, or automations.

Choose based on how you plan to use the response — for display, logging, or API integration.

10. STOP (Stop Key)

A Stop Key tells the model when to stop generating text.

You can set a specific word or symbol (like "stop" or "###") — once the model encounters it, it will stop writing.

This helps control unwanted or overly long responses.

In Short

Final Thoughts

These parameters are your control panel for shaping GTWY’s AI.
Adjust them to tune precision, creativity, structure, and personality — whether you're writing blogs, answering customer tickets, or powering automation workflows.

Experiment with small tweaks.
You’ll immediately see how each setting transforms the voice, logic, and intelligence of your AI agent.