📌 Strategy: Tries tokens per model first, then falls back to credits individually.
📌 Best for: Users who want to only use credits on models that need them.
How it Works #
- The system checks if tokens are available for each model individually.
- If a model has no tokens, it falls back to credits for that specific model only.
- If neither tokens nor credits are available for a model, the system skips it.
Example Scenario #
Model | Priority | Available Tokens | Credit Needed | User Credits |
---|---|---|---|---|
DeepSeek-7B | 1 | 100 | 20 | 50 |
LLaMA 3-8B | 2 | 0 | 15 | 10 |
Mistral-7B | 3 | 0 | 10 | 60 |
Final Selection: #
✅ DeepSeek-7B → Uses Tokens
❌ LLaMA 3-8B is skipped (not enough tokens or credits)
✅ Mistral-7B → Uses Credits
📌 Why? Unlike Sequential Model Fallback, this strategy does not force credit fallback if a model is unavailable in tokens.