Tokenizer issue
#2
by
apoorvumang
- opened
There is something wrong with the tokenizer, it keeps generating multiple chat turns until max_tokens is reached.
I'm currently bypassing it by using tokenizer from llama-3-8b
tokenizer = load_tokenizer(get_model_path("mlx-community/Meta-Llama-3-8B-Instruct-4bit"))
if someone can fix tokenizer here pls help, model is quite good (uncensored)