Leo Mistral Hessianai 7b

🚀 LAION LeoLM: Linguistically Enhanced Open Language Model

Meet LeoLM-Mistral, the first open and commercially available German Foundation Language Model built on Mistral 7b. Our models extend Llama-2's capabilities into German through continued pretraining on a large corpus of German-language and mostly locality specific text. Thanks to a compute grant at HessianAI's new supercomputer 42, we release three foundation models trained with 8k context length.

LeoLM/leo-mistral-hessianai-7b under Apache 2.0 and LeoLM/leo-hessianai-7b and LeoLM/leo-hessianai-13b under the Llama-2 community license (70b also coming soon! 👀). With this release, we hope to bring a new wave of opportunities to German open-source and commercial LLM research and accelerate adoption. Read our blog post or our paper (preprint coming soon) for more details!

A project by Björn Plüster and Christoph Schuhmann in collaboration with LAION and HessianAI.

🚀 Quick Start

📦 Installation

First install direct dependencies:

pip install transformers torch accelerate

If you want faster inference using flash-attention2, you need to install these dependencies:

pip install packaging ninja
pip install flash-attn

💻 Usage Examples

Basic Usage

Then load the model in transformers:

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model = AutoModelForCausalLM.from_pretrained(
    model="LeoLM/leo-mistral-hessianai-7b",
    device_map="auto",
    torch_dtype=torch.bfloat16,
    use_flash_attn_2=True # optional
)

✨ Features

• Multilingual Support: Supports both English and German, extending Llama - 2's capabilities to the German language.
• Large Context Length: Trained with an 8k context length, enabling more comprehensive text processing.
• Multiple Licensing Options: Different models are available under Apache 2.0 and Llama - 2 community license, providing flexibility for various usage scenarios.

📚 Documentation

Model Details

Training parameters

Note that for Mistral training, we changed learning rate to 1e-5 going down to 1e-6. We also used Zero stage 3 and bfloat16 dtype.

Benchmarks

📄 License

The models are released under different licenses:

• LeoLM/leo-mistral-hessianai-7b is under Apache 2.0.
• LeoLM/leo-hessianai-7b and LeoLM/leo-hessianai-13b are under the Llama-2 community license.