⚠️ Common Reasons for Errors with Claude on Janitor AI/silly
Hello, many users on Janitor AI (j.ai) experience errors when using Claude, either immediately or after a short time. There are two primary reasons for this:
1. Context Limit Misunderstanding (only j.ai)
The core issue is often a misunderstanding of how Janitor AI calculates the context limit.
- The context limit you configure (e.g., 32k) applies exclusively to the ongoing messages within your active chat session.
- This limit does not include your Persona, your Custom Prompt, or the Character's Definition ({{char}}).
- These additional elements are added on top of your chat history.
Example:
- Your Persona: 200 tokens
- Your Custom Prompt: 1,000 tokens
- Character Definition (persona, starting message, etc.): 2,000 tokens
- Your set context limit: 32,000 tokens
In this scenario, you are actually sending 35,200 tokens (32,000 + 200 + 1,000 + 2,000) to the model, which exceeds its 32k limit and causes an error.
2. Unsupported Generation Settings
A second common reason is using advanced generation settings that are not compatible with the Claude model.
- Temperature: Claude models generally do not support a temperature higher than 1. Some versions may even require the temperature to be set exactly to 1 (not 0.9, 0.8, or anything over 1).
- Other Settings: Claude does not support settings such as:
- Top P
- Top K
- Repetition Penalty
- Settings that work: Features like 'Prefill' and 'Banned Words' definitely still work.