afrideva

/

Mixtral-GQA-400m-v2-GGUF

like 1

Original uploader didn't specify, tested with alpaca but think it is a base model and needs to be finetuned before it does well with any format

the original tokenizer_config.json suggests it doesn’t support system prompts, it wraps user messages in llama2/mistral instruction tokens ([INST], [/INST])
https://huggingface.co./BEE-spoke-data/Mixtral-GQA-400m-v2/blob/6f8c51d1bf60da6f8e64ba7fb75fb747d9b124cf/tokenizer_config.json#L32