--- language: - de - bg - cs - da - el - en - es - et - fi - fr - ga - hr - hu - it - lt - lv - mt - nl - pl - pt - ro - sl - sv - sk metrics: - accuracy - bleu pipeline_tag: text-generation library_name: transformers base_model: - openGPT-X/Teuken-7B-base-v0.4 --- # Model Card for Teuken-7B-instruct-v0.4 [Teuken-7B-base-v0.4](https://huggingface.co/openGPT-X/Teuken-7B-base-v0.4) is a 7B parameter multilingual large language model (LLM) pre-trained with 4T tokens within the research project OpenGPT-X. Teuken-7B-instruct-v0.4 is an instruction-tuned version of [Teuken-7B-base-v0.4](https://huggingface.co/openGPT-X/Teuken-7B-base-v0.4). ### Model Description - **Developed by:** Fraunhofer, Forschungszentrum Jülich, TU Dresden, DFKI - **Funded by:** German Federal Ministry of Economics and Climate Protection (BMWK) in the context of the OpenGPT-X project - **Model type:** Transformer based decoder-only model - **Language(s) (NLP):** bg, cs, da, de, el, en, es, et, fi, fr, ga, hr, hu, it, lt, lv, mt, nl, pl, pt, ro, sk, sl, sv - **Shared by:** OpenGPT-X ## Uses Teuken-7B-instruct-v0.4 is intended for commercial and research use in all official 24 European languages. Since Teuken-7B-instruct-v0.4 focuses on covering all 24 EU languages, it renders more stable results across these languages and better reflects European values in its answers than English-centric models. It is therefore specialized for use in multilingual tasks. ## Disclaimer Toxic Content: This Language Model (LLM) may generate content that is inappropriate, offensive, or harmful. While the dataset has been heavily filtered to minimize such outputs, the model may still produce text that is biased or toxic due to the large scale and diverse nature of the data. ### Out-of-Scope Use The model is not intended for use in math and coding tasks. ## Bias, Risks, and Limitations Teuken-7B-instruct-v0.4 is an instruction-tuned version of [Teuken-7B-base-v0.4](https://huggingface.co/openGPT-X/Teuken-7B-base-v0.4) that is not completely free from biases and hallucinations. ## How to Get Started with the Model ## Usage The model requires transformers, sentencepiece, and the torch library. After installation, here's an example of how to use the model: The prompt template for the fine-tuned model is defined as follows: ```python user="Hi!" lang_code = "DE" system_messages={ "EN": "A chat between a human and an artificial intelligence assistant." " The assistant gives helpful and polite answers to the human's questions.", "DE": "Ein Gespräch zwischen einem Menschen und einem Assistenten mit künstlicher Intelligenz." " Der Assistent gibt hilfreiche und höfliche Antworten auf die Fragen des Menschen.", } prompt = f"System: {system_messages[lang_code]}\nUser: {user}\nAssistant:" ``` ```python import torch from transformers import AutoModelForCausalLM, AutoTokenizer device = torch.device("cuda" if torch.cuda.is_available() else "cpu") model_name = "openGPT-X/Teuken-7B-instruct-v0.4" model = AutoModelForCausalLM.from_pretrained( model_name, trust_remote_code=True, torch_dtype=torch.bfloat16, attn_implementation="flash_attention_2", ) model = model.to(device).eval() tokenizer = AutoTokenizer.from_pretrained( model_name, use_fast=False, trust_remote_code=True, ) messages = [{"role": "User", "content": "Wer bist du?"}] prompt_ids = tokenizer.apply_chat_template(messages, chat_template="DE", tokenize=True, add_generation_prompt=True, return_tensors="pt") prediction = model.generate( prompt_ids.to(model.device), max_length=512, do_sample=True, top_k=50, top_p=0.95, temperature=0.7, num_return_sequences=1, ) prediction_text = tokenizer.decode(prediction[0]) print(prediction_text) ``` This example demonstrates how to load the model and tokenizer, prepare input, generate text, and print the result. ## Training Details ### Pre-Training Data [Teuken-7B-base-v0.4](https://huggingface.co/openGPT-X/Teuken-7B-base-v0.4) was pre-trained on 4 trillion tokens of data from publicly available sources. The pretraining data has a cutoff of September 2023. More information are available in our [preprint](http://arxiv.org/abs/2410.08800). ### Instruction-Tuning Data ### English | Dataset file | Sample Count | | ----------------------------------------------------- | ------------ | | en/bactrianx_EN_fastchat.jsonl | 66985 | | en/code_alpaca_fastchat.jsonl | 19990 | | en/evol_instruct_143k_fastchat.jsonl | 142968 | | en/evol_instruct_70k_fastchat.jsonl | 69968 | | en/lmsys_chat_1m_high_quality_train_en_fastchat.jsonl | 18651 | | en/open_orca_fastchat_aa.jsonl | 599968 | | en/open_orca_fastchat_ab.jsonl | 599968 | | en/open_orca_fastchat_ac.jsonl | 599968 | | en/open_orca_fastchat_ad.jsonl | 599968 | | en/open_orca_fastchat_ag.jsonl | 599968 | | en/open_orca_fastchat_ah.jsonl | 33891 | | en/sharegpt_v3_unfiltered_fastchat.jsonl | 93880 | | en/ultrachat_200k_fastchat.jsonl | 11525 | | **total** | **3457698** | ### German | Dataset file | Sample Count | | ----------------------------------------------------------- | ------------ | | de/bactrianx_DE_fastchat.jsonl | 67017 | | de/freedomintelligence_alpaca_gpt4_deutsch_fastchat.jsonl | 49969 | | de/freedomintelligence_evol_instruct_deutsch_fastchat.jsonl | 59022 | | de/freedomintelligence_sharegpt_deutsch_fastchat.jsonl | 6101 | | de/german_poems_fastchat.jsonl | 400 | | de/german_songs_fastchat.jsonl | 1000 | | de/ultrachat_de_1k_fastchat.jsonl | 959 | | **total** | **184468** | ### Training Procedure Instruction fined tuned version of [Teuken-7B-base-v0.4](https://huggingface.co/openGPT-X/Teuken-7B-base-v0.4). #### Training Hyperparameters - **Training regime:** bf16 mixed precision ## Evaluation ### Testing Data, Factors & Metrics #### Testing Data The model was evaluated in 21 languages on ARC, GSM8K, HellaSwag, TruthfulQA, Translation and MMLU. Results can be seen in the [European LLM Leaderboard](https://huggingface.co/spaces/openGPT-X/european-llm-leaderboard). ## Technical Specifications ### Model Architecture and Objective | Hyper-Parameter | Value | |----------------------------|----------| | Training Objective | CLM | | Activation Function | SwiGLU | | Seq Length | 4096 | | Position Embeddings | Rotary | | Num Layers | 32 | | Hidden Size | 4096 | | FFN Hidden Size | 13440 | | Num Attention Heads | 32 | | Head Dim | 128 | | Group Query Attention | yes | | Num Query Groups | 2 | | Normalization | RMSNorm | | Learning rate | 3e-4 | | Min learning rate | 3e-5 | | Disable bias in linear | yes | | Hidden dropout | 0.0 | | Attention dropout | 0.0 | | Optimizer | AdamW | | Beta1 | 0.9 | | Beta2 | 0.95 | | Sequence-parallelism | Data-type | bf16 | | Recompute-activations | yes | | Distributed-optimizers | yes | | Model Initialization | | ### Compute Infrastructure We trained our models on JUWELS Booster which consists of 936 compute nodes, each equipped with 4 NVIDIA A100 GPUs. The GPUs are hosted by AMD EPYC Rome CPUs. The compute nodes are connected with HDR-200 InfiniBand in a DragonFly+ topology. #### Hardware The configuration of JUWELS Booster compute nodes is the following: CPU: AMD EPYC 7402 processor; 2 sockets, 24 cores per socket, SMT-2 (total: 2×24×2 = 96 threads) in NPS-4 1 configuration Memory: 512 GB DDR4-3200 RAM (of which at least 20 GB is taken by the system software stack, including the file system); 256 GB per socket; 8 memory channels per socket (2 channels per NUMA domain) GPU: 4 × NVIDIA A100 Tensor Core GPU with 40 GB; connected via NVLink3 to each other Network: 4 × Mellanox HDR200 InfiniBand ConnectX 6 (200 Gbit/s each), HCA Periphery: CPU, GPU, and network adapter are connected via 2 PCIe Gen 4 switches with 16 PCIe lanes going to each device (CPU socket: 2×16 lanes). PCIe switches are configured in synthetic mode. #### Software [Megatron-LM](https://github.com/OpenGPTX/Megatron-LM) **BibTeX:** If you find our model useful in your research, please consider citing our [preprint](https://arxiv.org/abs/2410.03730): ``` @misc{ali2024teuken7bbaseteuken7binstructeuropean, title={Teuken-7B-Base & Teuken-7B-Instruct: Towards European LLMs}, author={Mehdi Ali and Michael Fromm and Klaudia Thellmann and Jan Ebert and Alexander Arno Weber and Richard Rutmann and Charvi Jain and Max Lübbering and Daniel Steinigen and Johannes Leveling and Katrin Klug and Jasper Schulze Buschhoff and Lena Jurkschat and Hammam Abdelwahab and Benny Jörg Stein and Karl-Heinz Sylla and Pavel Denisov and Nicolo' Brandizzi and Qasid Saleem and Anirban Bhowmick and Lennard Helmer and Chelsea John and Pedro Ortiz Suarez and Malte Ostendorff and Alex Jude and Lalith Manjunath and Samuel Weinbach and Carolin Penke and Oleg Filatov and Shima Asaadi and Fabio Barth and Rafet Sifa and Fabian Küch and Andreas Herten and René Jäkel and Georg Rehm and Stefan Kesselheim and Joachim Köhler and Nicolas Flores-Herr}, year={2024}, eprint={2410.03730}, archivePrefix={arXiv}, primaryClass={cs.CL}, url={https://arxiv.org/abs/2410.03730}, } ``` # Team ## Data Team Anirban Bhowmick (IAIS), Nicolo Brandizzi (IAIS), Lennard Helmer (IAIS), Benny Jörg Stein (IAIS), Karl-Heinz Sylla (IAIS), Pavel Denisov (IAIS), Qasid Saleem (IAIS), Johannes Leveling (IAIS), Hammam Abdelwahab (IAIS), Luzian Hahn (IIS), Farzad Naderi (IIS), Md Saiful Islam (IIS), Alexander Schwirjow (IIS), Pedro Ortiz Suarez (ex. DFKI), Malte Ostendorff (ex. DFKI) ## Model-Training Team ### Core contributors Mehdi Ali (IAIS), Michael Fromm (IAIS), Jan Ebert (FZJ), Chelsea John (FZJ), Lena Jurkschat (TUD), Alexander Weber (IAIS) ### Contributors: Richard Rutmann (IAIS), Daniel Steinigen (IAIS), Lalith Manjunath (TUD), Carolin Penke (FZJ) ## Evaluation Team ### Core contributors Klaudia Thellmann (TUD), Alex Jude (IAIS), Jasper Buschhoff (IAIS) ### Contributors: Shima Assadi (IIS), Fabio Barth (DFKI) ## Management Joachim Köhler (IAIS), Nicolas Flores-Herr (IAIS), Stefan Kesselheim (FZJ), Andreas Herten (FZJ), Georg Rehm (DFKI), René Jäkel (TUD), Fabian Küch (IIS), Nicole Hildebrandt (IAIS), Ines Wendler (IAIS)

Contact Information

You can reach out to the following model card contact: