Model Card for MediaTek Research Breeze-7B-32k-Base-v1_0

MediaTek Research Breeze-7B (hereinafter referred to as Breeze-7B) is a language model family that builds on top of Mistral-7B, specifically intended for Traditional Chinese use.

Breeze-7B-Base is the base model for the Breeze-7B series. It is suitable for use if you have substantial fine-tuning data to tune it for your specific use case.

Breeze-7B-Instruct derives from the base model Breeze-7B-Base, making the resulting model amenable to be used as-is for commonly seen tasks.

Breeze-7B-32k-Base is extended from the base model with more data, base change, and the disabling of the sliding window. Roughly speaking, that is equivalent to 44k Traditional Chinese characters.

Breeze-7B-32k-Instruct derives from the base model Breeze-7B-32k-Base, making the resulting model amenable to be used as-is for commonly seen tasks.

Practicality-wise:

• Breeze-7B-Base expands the original vocabulary with additional 30,000 Traditional Chinese tokens. With the expanded vocabulary, everything else being equal, Breeze-7B operates at twice the inference speed for Traditional Chinese to Mistral-7B and Llama 7B. [See Inference Performance.]
• Breeze-7B-Instruct can be used as is for common tasks such as Q&A, RAG, multi-round chat, and summarization.
• Breeze-7B-32k-Instruct can perform tasks at a document level (For Chinese, 20 ~ 40 pages).

A project by the members (in alphabetical order): Chan-Jan Hsu 許湛然, Feng-Ting Liao 廖峰挺, Po-Chun Hsu 許博竣, Yi-Chang Chen 陳宜昌, and the supervisor Da-Shan Shiu 許大山.

Features

• Breeze-7B-32k-Base-v1_0

  • Expanding the vocabulary dictionary size from 32k to 62k to better support Traditional Chinese
  • 32k-token context length

• Breeze-7B-32k-Instruct-v1_0

  • Expanding the vocabulary dictionary size from 32k to 62k to better support Traditional Chinese
  • 32k-token context length
  • Multi-turn dialogue (without special handling for harmfulness)

Model Details

• Breeze-7B-32k-Base-v1_0
  • Pretrained from: Breeze-7B-Base
  • Model type: Causal decoder-only transformer language model
  • Language: English and Traditional Chinese (zh-tw)
• Breeze-7B-32k-Instruct-v1_0
  • Finetuned from: Breeze-7B-32k-Base
  • Model type: Causal decoder-only transformer language model
  • Language: English and Traditional Chinese (zh-tw)

Long-context Performance

Needle-in-a-haystack Performance

We use the passkey retrieval task to test the model's ability to attend to different various depths in a given sequence. A key in placed within a long context distracting document for the model to retrieve. The key position is binned into 16 bins, and there are 20 testcases for each bin. Breeze-7B-32k-Base clears the tasks with 90+% accuracy, shown in the figure below.

Long-DRCD Performance

Short-Benchmark Performance

Use in Transformers

First, install direct dependencies:

pip install transformers torch accelerate

Flash-attention2 is strongly recommended for long context scenarios.

pip install packaging ninja
pip install flash-attn

Then load the model in transformers:

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model = AutoModelForCausalLM.from_pretrained(
    "MediaTek-Research/Breeze-7B-32k-Base-v1_0",
    device_map="auto",
    torch_dtype=torch.bfloat16,
    attn_implementation="flash_attention_2" # optional but highly recommended
)
from transformers import AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained("MediaTek-Research/Breeze-7B-32k-Base-v1_0")
tokenizer.tokenize("你好，我可以幫助您解決各種問題、提供資訊和協助您完成許多不同的任務。例如：回答技術問題、提供建議、翻譯文字、尋找資料或協助您安排行程等。請告訴我如何能幫助您。")
# Tokenized results
# ['▁', '你好', '，', '我', '可以', '幫助', '您', '解決', '各種', '問題', '、', '提供', '資訊', '和', '協助', '您', '完成', '許多', '不同', '的', '任務', '。', '例如', '：', '回答', '技術', '問題', '、', '提供', '建議', '、', '翻譯', '文字', '、', '尋找', '資料', '或', '協助', '您', '安排', '行程', '等', '。', '請', '告訴', '我', '如何', '能', '幫助', '您', '。']

Citation

@article{MediaTek-Research2024breeze7b,
      title={Breeze-7B Technical Report}, 
      author={Chan-Jan Hsu and Chang-Le Liu and Feng-Ting Liao and Po-Chun Hsu and Yi-Chang Chen and Da-Shan Shiu},
      year={2024},
      eprint={2403.02712},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}