metadata

language:
  - fr
  - it
  - de
  - es
  - en
license: apache-2.0
tags:
  - moe
inference: false

Model Card for Mixtral-Extraction-4x7B-Instruct-v0.1

This model is an experimental model created by merging mistralai/Mixtral-8x7B-Instruct-v0.1 experts.

How we extracted experts

Experts are selected and extracted.
This model specifies 4 experts.

How To Convert

use colab cpu-high-memory.
You can extract experts 1-7 by selecting experts as bit string.

experts_extract_bit = "11110000"

convert_mixtral_8x7b_to_4x7b_extract.ipynb

Usage

pip install git+https://github.com/huggingface/transformers --upgrade
pip install torch accelerate bitsandbytes flash_attn

from transformers import AutoTokenizer, AutoModelForCausalLM, MixtralForCausalLM
import torch

model_name_or_path = "mmnga/Mixtral-Extraction-4x7B-Instruct-v0.1"

tokenizer = AutoTokenizer.from_pretrained(model_name_or_path)
model = MixtralForCausalLM.from_pretrained(model_name_or_path, load_in_8bit=True)

text = "[INST] What was John Holt's vision on education? [/INST] "
inputs = tokenizer("<s> " + text, return_tensors="pt")

outputs = model.generate(**inputs, max_new_tokens=128)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))