Here are the recommended settings for using the cosmosrp-3.5 model with ChubAI:
Base Setup (ChubAI)
- API: OpenAI / Reverse Proxy
- Base URL:
https://api.pawan.krd/cosmosrp-3.5/v1 - API Key: Your Pawan.Krd key
Model Parameters
- Model Name:
gpt-4(on ChubAI interface) - Temperature:
0.8(recommended for cosmosrp-3.5) - Top P:
1.0(default, keeps output diverse) - Top K:
40(if available—set to default if not shown) - Repetition Penalty:
1.15(slight penalty to avoid repetition, if available) - Frequency Penalty:
0.0 - Presence Penalty:
0.0 - Minimum Length:
0 - Max New Tokens:
2048-4096(recommended; keep under 4096 for stable output) - Context Size: Up to
65536tokens (64k); but ChubAI might use less—adjust as needed - Lorebook Scan Depth:
200-500tokens (default/depend on your lorebook size) - Lorebook Token Budget:
500-1000(adjust based on character and lore size)
Prompt/Instructions
- Pre History Instructions: (Leave empty or use a single space; do NOT use jailbreaks/custom pre-prompts. The model is unfiltered & works best with default settings.)
- Post History Instructions: (Leave empty or use a single space.)
- Impersonation Prompt: (Use your character’s default persona/roleplay introduction, or leave empty for standard behavior.)