Felladrin

/

Llama-68M-Chat-v1

like 11

Hi, @Clausss !

Yes, for models where config.json has "model_type": "llama", max_position_embeddings indicates the maximum context size, meaning that it was trained with texts up to 2048 tokens long. If the generation surpasses this limit, the model will probably start outputting nonsense sentences.

Thanks you for fast response and very well explain 🤗