dbw6/Mixtral-8x7B-AQLM-2Bit-2x8-hf

This repository contains a Hugging Face export of Mixtral-8x7B-v0.1 quantized with AQLM using the 2-bit 2x8 scheme.

Base model

• mistralai/Mixtral-8x7B-v0.1

Quantization

• Method: AQLM
• Scheme: 2x8
• Effective label: 2-bit
• Source checkpoint: /work/bduan1/quantized_models/Mixtral-8x7B-AQLM-2bit-2x8

Conversion

This repo was produced with convert_to_hf.py from the AQLM project, then exported with --save_safetensors and --save_tokenizer.

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "dbw6/Mixtral-8x7B-AQLM-2Bit-2x8-hf"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype="auto",
    device_map="auto",
    trust_remote_code=True,
)