This is a phi 4 (14B) model, fine tuned for more engaging conversation, to limit sycofancy in language models and encouraging the models to (gently) push back and call out bad ideas.
This llama model was trained 2x faster with
Unsloth and Huggingface's TRL library.
