Fix tokenizer: add <IMG_CONTEXT> special token and chat template
#7
by whats2000 - opened
This PR fixes the KeyError: <IMG_CONTEXT> issue when using vLLM by adding <IMG_CONTEXT> as an additional special token. It also adds a standard Llama-3-style chat template so framework like vLLM can properly parse conversations.
whats2000 changed pull request status to closed