MiniChat-3B-exl2

MiniChat-3B-EXL2

Original model: MiniChat-3B
Model creator: GeneZC

4bpw h6 (main)
4bpw h8
4.65bpw h8
5bpw h8
5.5bpw h8
6bpw h8
8bpw h8

I originally planned to make both h6 and h8 versions for each quant but there was consistent 30MB difference between h6 and h8.
So I don't see much of a reason to upload the rest of h6.

How to run

This quantization method uses GPU and requires Exllamav2 loader which can be found in following applications:

Text Generation Webui

KoboldAI

ExUI

Original model card:

MiniChat-3B

📑 arXiv | 👻 GitHub | 🤗 HuggingFace-MiniMA | 🤗 HuggingFace-MiniChat | 🤖 ModelScope-MiniMA | 🤖 ModelScope-MiniChat

❗ Must comply with LICENSE of LLaMA2 since it is derived from LLaMA2.

A language model distilled and finetuned from an adapted version of LLaMA2-7B following "Towards the Law of Capacity Gap in Distilling Language Models".

Outperforming a wide range of 3B competitors in GPT4 evaluation and even competing with several 7B chat models.

<img src="https://huggingface.co/GeneZC/MiniChat-3B/resolve/main/teaser_b.jpg" alt="teaser_b" width="687" />

The following is an example code snippet to use MiniChat-3B:

import torch

from transformers import AutoModelForCausalLM, AutoTokenizer

from conversation import get_default_conv_template

# MiniChat
tokenizer = AutoTokenizer.from_pretrained("GeneZC/MiniChat-3B", use_fast=False)
# GPU.
model = AutoModelForCausalLM.from_pretrained("GeneZC/MiniChat-3B", use_cache=True, device_map="auto", torch_dtype=torch.float16).eval()
# CPU.
# model = AutoModelForCausalLM.from_pretrained("GeneZC/MiniChat-3B", use_cache=True, device_map="cpu", torch_dtype=torch.float16).eval()

conv = get_default_conv_template("minichat")

question = "Implement a program to find the common elements in two arrays without using any extra data structures."
conv.append_message(conv.roles[0], question)
conv.append_message(conv.roles[1], None)
prompt = conv.get_prompt()
input_ids = tokenizer([prompt]).input_ids
output_ids = model.generate(
    torch.as_tensor(input_ids).cuda(),
    do_sample=True,
    temperature=0.7,
    max_new_tokens=1024,
)
output_ids = output_ids[0][len(input_ids[0]):]
output = tokenizer.decode(output_ids, skip_special_tokens=True).strip()
# output: "def common_elements(arr1, arr2):\n    if len(arr1) == 0:\n        return []\n    if len(arr2) == 0:\n        return arr1\n\n    common_elements = []\n    for element in arr1:\n        if element in arr2:\n            common_elements.append(element)\n\n    return common_elements"
# Multiturn conversation could be realized by continuously appending questions to `conv`.

Bibtex

@article{zhang2023law,
    title={Towards the Law of Capacity Gap in Distilling Language Models},
    author={Zhang, Chen and Song, Dawei and Ye, Zheyu and Gao, Yan},
    year={2023},
    url={https://arxiv.org/abs/2311.07052}
}

Open LLM Leaderboard Evaluation Results

Detailed results can be found here