Run Prompt
Create and test prompts for optimizing LLM responses
1. Select Dataset
Click on the dataset name you want to use to create prompts. If no dataset is showing in the dashboard, ensure you have followed the steps required to Add Dataset on the Future AGI platform.
2. Access Run Prompt Interface
You can view your dataset in a spreadsheet-like interface. On the top right corner, select Run Prompt option to create a prompt.
3. Configure Your Prompt
Basic Configuration
- Enter a descriptive name for your prompt
- Select your desired model from the dropdown
API Key Setup
After selecting the model, enter your API key in the popup window to access the selected model. In this example, we’re using gpt-4o.
Output Configuration
Choose the output type for your prompt:
- string: For simple text responses (e.g., “correct”/“incorrect”)
- object: For JSON-structured outputs
Writing Your Prompt
Access dataset columns using double curly braces. A dropdown menu will appear showing available columns. Selected column names will be automatically enclosed in braces.
4. Model Parameters
Configure these parameters to optimize your model’s performance:
| Parameter | Description | Impact |
|---|---|---|
| Concurrency | Number of simultaneous prompt processes | Higher values increase speed but may hit API limits |
| Temperature | Controls response randomness | 0: Deterministic 1: More creative but potentially less accurate |
| Top P | Controls token selection diversity | Lower: More focused Higher: More varied responses |
| Max Tokens | Maximum response length | Higher values allow longer responses but increase API usage |
| Presence Penalty | Controls topic repetition | Higher: More diverse topics Lower: More focused on single topic |
| Frequency Penalty | Controls word/phrase repetition | Higher: Less repetition Lower: Allows repetition |
Response Format
- Choose between
textorJSONoutput format - Configure tool interaction:
required: Force tool usageauto: Let model decidenone: Disable tool interaction
5. Execute Prompt
Click Save and Run to execute your prompt configuration. Results will appear in a new column named after your prompt.
The generated responses will be visible in the newly created column:
Best Practices
• Start with lower concurrency to test API limits
• Use temperature 0.0-0.3 for factual tasks
• Use temperature 0.7-1.0 for creative tasks
• Set reasonable max token limits to control costs
• Test prompts on a small subset before full execution