Models
Chat-style requests
Structure conversations with roles and messages.
Chat requests send an ordered list of messages. Each message has a role and content.
| Role | Purpose |
|---|---|
| system | Instructions that shape behavior |
| user | Input from the end user |
| assistant | Previous model replies, for context |
Common parameters
- temperature — higher values produce more varied output.
- max_tokens — caps the length of the response.
- stop — sequences that end generation early.
Support for individual parameters varies by model. Check the model page for details.