Occiglot-7B-EU5 is a generative language model with 7B parameters supporting the top-5 EU languages (English, Spanish, French, German, and Italian) and trained by the Occiglot Research Collective.
It is based on Mistral-7B-v0.1 and trained on 293B tokens of additional multilingual and code data with a block size of 8,192 tokens per sample.
Note that the model is a general-purpose base model and was not instruction-fine-tuned nor optimized for chat or other applications. We make an instruction tuned variant available as occiglot-7b-eu5-instruct
This is the first release of an ongoing open research project for multilingual language models.
If you want to train a model for your own language or are working on evaluations, please contact us or join our Discord server. We are open for collaborations!
You can use this model directly with a pipeline for text generation. Since the generation relies on some randomness, we
set a seed for reproducibility:
python
1>>>from transformers import pipeline, set_seed
2>>> generator = pipeline('text-generation', model='occiglot/occiglot-7b-eu5')3>>> set_seed(42)4>>> generator("Hallo, Ich bin ein Sprachmodell,", max_length=40, num_return_sequences=1)5[{'generated_text':'Hallo, Ich bin ein Sprachmodell, das dir bei der Übersetzung von Texten zwischen Deutsch und Englisch helfen kann. Wenn du mir einen Text in Deutsch'}]
Dataset
The training data was split amongst the 4 target languages (de, es, fr, it) and the continuous training in English and code.
The data distribution by language (estimated) is as follows:
English: ~13%
Code: ~5%
German: ~20%
Spanish: ~20%
French: ~20%
Italian: ~20%
The training data was prepared using lm-datasets.
The exact data configuration is here.
Training settings
Continual pre-training on 128 x A100-80GB on HessianAI's 42.
Preliminary evaluation results can be found below.
Please note that the non-English results are based on partially machine-translated datasets and English prompts (Belebele and Okapi framework) and thus should be interpreted with caution, e.g., biased towards English model performance.
Currently, we are working on more suitable benchmarks for Spanish, French, German, and Italian.
