🚀 LAION LeoLM: Linguistically Enhanced Open Language Model
Meet LeoLM, the first open and commercially available German Foundation Language Model built on Llama-2. It extends Llama-2's capabilities into German through continued pretraining on a large corpus of German-language and mostly locality specific text. With the support of a compute grant at HessianAI's new supercomputer 42, we're releasing two foundation models with an 8k context length: LeoLM/leo-hessianai-7b
and LeoLM/leo-hessianai-13b
under the Llama-2 community license (a 70b version is also coming soon! 😎). This release aims to bring new opportunities to German open-source and commercial LLM research and accelerate adoption. For more details, check out our blog post or our upcoming paper (preprint soon).
A project by Björn Plüster and Christoph Schuhmann in collaboration with LAION and HessianAI.
🚀 Quick Start
✨ Features
- Built on Llama-2, extending its capabilities to the German language.
- Two foundation models with 8k context length, and a 70b model on the way.
- A chat model
LeoLM/leo-hessianai-7b-chat
finetuned for German instruction tasks.
📦 Installation
First, install the direct dependencies:
pip install transformers torch sentencepiece
If you want faster inference using flash-attention2, install these additional dependencies:
pip install packaging ninja
pip install flash-attn==v2.1.1 --no-build-isolation
pip install git+https://github.com/HazyResearch/flash-attention.git@v2.1.1#subdirectory=csrc/rotary
💻 Usage Examples
Basic Usage
from transformers import pipeline
import torch
system_prompt = """<|im_start|>system
Dies ist eine Unterhaltung zwischen einem intelligenten, hilfsbereitem KI-Assistenten und einem Nutzer.
Der Assistent gibt ausführliche, hilfreiche und ehrliche Antworten.<|im_end|>
"""
prompt_format = "<|im_start|>user\n{prompt}<|im_end|>\n<|im_start|>assistant\n"
prompt = "Erkläre mir wie die Fahrradwegesituation in Hamburg ist."
generator = pipeline(model="LeoLM/leo-hessianai-7b-chat", device="cuda", torch_dtype=torch.float16, trust_remote_code=True)
print(generator(prompt_format.format(prompt=prompt), do_sample=True, top_p=0.95, max_length=8192))
The model's response:
"Als KI kann ich keine persönlichen Beobachtungen teilen, aber ich kann einige allgemeine Informationen zur Fahrradwegesituation in Hamburg liefern. Im Vergleich zu vielen anderen großen Städten hat Hamburg eine hohe Anzahl von Fahrradfahrern und nimmt seine Verantwortung für nachhaltige Verkehrsmittel sehr ernst. Es gibt viele Fahrradwege und separate Fahrspuren, die Radfahrern ermöglichen, zusammen mit dem Autoverkehr zu fahren. Diese Fahrradspuren sind oft mit Markierungen gekennzeichnet und durch physische Trennungen von anderen Fahrspuren abgegrenzt. Darüber hinaus gibt es viele Fahrradstraßen, auf denen Radfahrer Vorfahrt haben und Autos langsamer fahren müssen.
In einigen städtischen Gebieten können Fahrradwege jedoch eng oder überfüllt sein, besonders während der Stoßzeiten. Es gibt auch viele Kreuzungen, an denen Radfahrer anhalten und auf Grün warten müssen, ähnlich wie Autofahrer. Insgesamt ist die Fahrradinfrastruktur in Hamburg ziemlich gut, aber wie überall gibt es immer Raum für Verbesserungen."
📚 Documentation
LeoLM Chat
LeoLM/leo-hessianai-7b-chat
is a German chat model based on the foundation model LeoLM/leo-hessianai-7b
, finetuned on selected German instruction datasets. It performs well on writing, explanation, and discussion tasks but has some challenges with math and advanced reasoning. See our MT-Bench-DE scores:
{
"first_turn": 5.75,
"second_turn": 4.45,
"categories": {
"writing": 5.875,
"roleplay": 6.3,
"reasoning": 3.5,
"math": 2.85,
"coding": 2.95,
"extraction": 4.3,
"stem": 7.4,
"humanities": 7.625
},
"average": 5.1
}
Model Details
Prompting / Prompt Template
Prompt dialogue template (ChatML format):
"""
<|im_start|>system
{system_message}<|im_end|>
<|im_start|>user
{prompt}<|im_end|>
<|im_start|>assistant
"""
The model input can contain multiple conversation turns between the user and the assistant, for example:
<|im_start|>user
{prompt 1}<|im_end|>
<|im_start|>assistant
{reply 1}<|im_end|>
<|im_start|>user
{prompt 2}<|im_end|>
<|im_start|>assistant
(...)
🔧 Technical Details
Finetuning Details
Hyperparameter |
Value |
Num epochs |
3 |
Examples per epoch |
131214 |
Global batch size |
256 |
Learning rate |
3e-5 |
Warmup steps |
100 |
LR scheduler |
Cosine |
Adam betas |
(0.9, 0.95) |
Dataset Details
## Stats for 'Subset of OpenAssistant/OASST-DE' (3534 samples (100.0%))
-----------------
Accepted: 3534/3534 (100.0%)
Accepted tokens: 2259302
Skipped: 0 (0.0%)
Min tokens per sample: 29
Max tokens per sample: 2484
Avg tokens per sample: 639.3044708545557
-----------------
## Stats for 'Subset of FreedomIntelligence/evol-instruct-deutsch' (57841 samples (100.0%))
-----------------
Accepted: 57841/57841 (100.0%)
Accepted tokens: 42958192
Skipped: 0 (0.0%)
Min tokens per sample: 33
Max tokens per sample: 5507
Avg tokens per sample: 742.6944900675991
-----------------
## Stats for 'Subset of FreedomIntelligence/alpaca-gpt4-deutsch' (48969 samples (100.0%))
-----------------
Accepted: 48969/48969 (100.0%)
Accepted tokens: 13372005
Skipped: 0 (0.0%)
Min tokens per sample: 19
Max tokens per sample: 1359
Avg tokens per sample: 273.07082031489307
-----------------
## Stats for 'Subset of LeoLM/OpenSchnabeltier' (21314 samples (100.0%))
-----------------
Accepted: 21314/21314 (100.0%)
Accepted tokens: 8134690
Skipped: 0 (0.0%)
Min tokens per sample: 25
Max tokens per sample: 1202
Avg tokens per sample: 381.65947264708643
-----------------
## Stats for 'Subset of LeoLM/German_Poems' (490 samples (100.0%))
-----------------
Accepted: 490/490 (100.0%)
Accepted tokens: 618642
Skipped: 0 (0.0%)
Min tokens per sample: 747
Max tokens per sample: 1678
Avg tokens per sample: 1262.534693877551
-----------------
## Stats for 'Subset of LeoLM/German_Songs' (392 samples (100.0%))
-----------------
Accepted: 392/392 (100.0%)
Accepted tokens: 187897
Skipped: 0 (0.0%)
Min tokens per sample: 231
Max tokens per sample: 826
Avg tokens per sample: 479.3290816326531
-----------------
## Stats for 'total' (132540 samples (100.0%))
-----------------
Accepted: 132540/132540 (100.0%)
Accepted tokens: 67530728
Skipped: 0 (0.0%)
Min tokens per sample: 19
Max tokens per sample: 5507
Avg tokens per sample: 509.51205673758864
-----------------
📄 License
This project is released under the LLAMA 2 COMMUNITY LICENSE AGREEMENT.
⚠️ Important Note
LeoLM has been tested in English and German, but it cannot cover all scenarios. As with all LLMs, the potential outputs of LeoLM/leo-hessianai-7b-chat
cannot be predicted in advance, and the model may produce inaccurate, biased, or other objectionable responses to user prompts. Therefore, developers should perform safety testing and tuning tailored to their specific applications of the model before deployment. Please refer to Meta's Responsible Use Guide.