This is the model card for the first instruction tuned model of the EuroLLM series: EuroLLM-1.7B-Instruct. You can also check the pre-trained version: [EuroLLM-1.7B](https://huggingface.co/utter-project/EuroLLM-1.7B).
- **Developed by:** Unbabel, Instituto Superior Técnico, Instituto de Telecomunicações, University of Edinburgh, Aveni, University of Paris-Saclay, University of Amsterdam, Naver Labs, Sorbonne Université.
The EuroLLM project has the goal of creating a suite of LLMs capable of understanding and generating text in all European Union languages as well as some additional relevant languages.
EuroLLM-1.7B is a 1.7B parameter model trained on 4 trillion tokens divided across the considered languages and several data sources: Web data, parallel data (en-xx and xx-en), and high-quality datasets.
EuroLLM-1.7B-Instruct was further instruction tuned on EuroBlocks, an instruction tuning dataset with focus on general instruction-following and machine translation.
### Model Description
EuroLLM uses a standard, dense Transformer architecture:
- We use grouped query attention (GQA) with 8 key-value heads, since it has been shown to increase speed at inference time while maintaining downstream performance.
- We perform pre-layer normalization, since it improves the training stability, and use the RMSNorm, which is faster.
- We use the SwiGLU activation function, since it has been shown to lead to good results on downstream tasks.
- We use rotary positional embeddings (RoPE) in every layer, since these have been shown to lead to good performances while allowing the extension of the context length.
For pre-training, we use 256 Nvidia H100 GPUs of the Marenostrum 5 supercomputer, training the model with a constant batch size of 3,072 sequences, which corresponds to approximately 12 million tokens, using the Adam optimizer, and BF16 precision.
model = AutoModelForCausalLM.from_pretrained(model_id)
text = '<|im_start|>system\n<|im_end|>\n<|im_start|>user\nTranslate the following English source text to Portuguese:\nEnglish: I am a language model for european languages. \nPortuguese: <|im_end|>\n<|im_start|>assistant\n'
We evaluate EuroLLM-1.7B-Instruct on several machine translation benchmarks: FLORES-200, WMT-23, and WMT-24 comparing it with [Gemma-2B](https://huggingface.co/google/gemma-2b) and [Gemma-7B](https://huggingface.co/google/gemma-7b) (also instruction tuned on EuroBlocks).
The results show that EuroLLM-1.7B is substantially better than Gemma-2B in Machine Translation and competitive with Gemma-7B.
We also compare EuroLLM-1.7B with [TinyLlama-v1.1](https://huggingface.co/TinyLlama/TinyLlama_v1.1) and [Gemma-2B](https://huggingface.co/google/gemma-2b) on 3 general benchmarks: Arc Challenge and Hellaswag.
For the non-english languages we use the [Okapi](https://aclanthology.org/2023.emnlp-demo.28.pdf) datasets.
Results show that EuroLLM-1.7B is superior to TinyLlama-v1.1 and similar to Gemma-2B on Hellaswag but worse on Arc Challenge. This can be due to the lower number of parameters of EuroLLM-1.7B (1.133B non-embedding parameters against 1.981B).
#### Arc Challenge
| Model | Average | English | German | Spanish | French | Italian | Portuguese | Chinese | Russian | Dutch | Arabic | Swedish | Hindi | Hungarian | Romanian | Ukrainian | Danish | Catalan |
| Model | Average | English | German | Spanish | French | Italian | Portuguese | Russian | Dutch | Arabic | Swedish | Hindi | Hungarian | Romanian | Ukrainian | Danish | Catalan |
EuroLLM-1.7B-Instruct has not been aligned to human preferences, so the model may generate problematic outputs (e.g., hallucinations, harmful content, or false statements).
## Paper
Paper: [EuroLLM: Multilingual Language Models for Europe](https://huggingface.co/papers/2409.16235)
