LAION LeoLM 70b: Linguistically Enhanced Open Language Model

Meet LeoLM, the first open and commercially available German Foundation Language Model built on Llama-2. 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 a series foundation models trained with 8k context length under the Llama-2 community license. Now, we're finally releasing the much anticipated leo-hessianai-70b, the largest model of this series based on Llama-2-70b. 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.

Model Details

• Finetuned from: meta-llama/Llama-2-70b-hf
• Model type: Causal decoder-only transformer language model
• Language: English and German
• License: LLAMA 2 COMMUNITY LICENSE AGREEMENT
• Contact: LAION Discord or Björn Plüster

Use in 🤗Transformers

First install direct dependencies:

pip install transformers torch

Then load the model in transformers. Note that this requires lots of VRAM and most-likely multiple devices. Use load_in_8bit=True or load_in_4bit=True to save some memory by using a quantized version. For more quantized versions, check out our models at TheBloke's page: (coming soon!)

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model = AutoModelForCausalLM.from_pretrained(
 model="LeoLM/leo-hessianai-70b",
 device_map="auto",
 torch_dtype=torch.bfloat16,
 use_flash_attention_2=False # Set to true to use FA2. Requires `pip install flash-attn`
)

Training parameters

👁 Training Hyperparameters

Benchmarks

👁 Benchmark Scores
👁 Translation Benchmark Scores