Nous-Hermes-2-Mixtral-8x7B-DPO-HQQ

This model is part of a series of HQQ tests. I make no claims on the performance of this model and it very well may change or be deleted.

This is a very extreme example of quantization.

from hqq.engine.hf import HQQModelForCausalLM, AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained('macadeliccc/Nous-Hermes-2-Mixtral-8x7B-DPO-HQQ', trust_remote_code=True)
model = HQQModelForCausalLM.from_pretrained(
    "macadeliccc/Nous-Hermes-2-Mixtral-8x7B-DPO-HQQ",
    torch_dtype=torch.float16,
    device_map="auto"
)