4095c68822
- Use case-insensitive matching for model names and routing - Default max_tokens/num_predict to 8192 for all Ollama models to prevent truncation - Increase default context window and add more large-context model families - Ensure DeepSeek routing handles Ollama-hosted variants correctly