File size: 7,848 Bytes
5479f93 e2ef0ce 5479f93 e2ef0ce |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 |
---
license: apache-2.0
language:
- en
- es
- de
- fr
- it
pipeline_tag: text-generation
---
# Occiglot-7B-EU5
> A [polyglot](https://en.wikipedia.org/wiki/Multilingualism#In_individuals) language model for the [Occident](https://en.wikipedia.org/wiki/Occident).
>
**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 [German Research Center for Artificial Intelligence (DFKI)](https://www.dfki.de/en/web).
It is based on [Mistral-7B-v0.1](https://huggingface.co/mistralai/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.
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. **We are open for collaborations!**
### Model details
- **Continued-pretraining from:** [Mistral-7B-v0.1](https://huggingface.co/mistralai/Mistral-7B-v0.1)
- **Model type:** Causal decoder-only transformer language model
- **Languages:** English, Spanish, French, German, Italian, and code.
- **License:** [Apache 2.0](https://www.apache.org/licenses/LICENSE-2.0.html)
- **Compute resources:** [HessianAI's 42](https://hessian.ai/)
- **Contributors:** Manuel Brack, Patrick Schramowski, Pedro Ortiz, Malte Ostendorff, Fabio Barth, Georg Rehm, Kristian Kersting
- **Research labs:** [SAINT](https://www.dfki.de/en/web/research/research-departments/foundations-of-systems-ai) and [SLT](https://www.dfki.de/en/web/research/research-departments/speech-and-language-technology)
### How to use
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
>>> from transformers import pipeline, set_seed
>>> generator = pipeline('text-generation', model='occiglot/occiglot-7b-eu5')
>>> set_seed(42)
>>> generator("Hallo, Ich bin ein Sprachmodell,", max_length=40, num_return_sequences=1)
[{'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](https://github.com/malteos/lm-datasets).
The exact data config will be released soon.
## Training settings
- Continual pre-training on 128 x A100-80GB on [HessianAI's 42](https://hessian.ai/).
- Framework: [Determined](https://www.determined.ai/)
- Precision: bf16
- Optimizer: AdamW (lr: 0.00001, warmup_steps: 420)
- Global batch size: 512 (with 8192 blocksize) split over 128 GPUs
- Cosine Annealing with Warmup
## Tokenizer
Tokenizer is unchanged from [Mistral-7B-v0.1](https://huggingface.co/mistralai/Mistral-7B-v0.1).
## Evaluation
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](https://huggingface.co/datasets/facebook/belebele) and [Okapi framework](https://github.com/nlp-uoregon/Okapi)) 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.
### All languages
| **model_name** | **arc_challenge** | **hellaswag** | **belebele** | **mmlu** | **avg** |
|--------------------------|-------------------|---------------|--------------|----------|---------|
| Mistral-7B-v0.1 | 0.5277 | 0.6825 | 0.7687 | 0.6287 | 0.6519 |
| leo-mistral-hessianai-7b | 0.4614 | 0.6423 | 0.6524 | 0.5440 | 0.5750 |
| Occiglot-7B-EU5 | 0.5083 | 0.7191 | 0.6758 | 0.5432 | 0.6116 |
### English
| **model_name** | **arc_challenge** | **hellaswag** | **belebele** | **mmlu** | **avg** |
|--------------------------|-------------------|---------------|--------------|----------|---------|
| Mistral-7B-v0.1 | 0.6143 | 0.8344 | 0.8444 | 0.6351 | 0.7321 |
| leo-mistral-hessianai-7b | 0.5213 | 0.7779 | 0.7356 | 0.5508 | 0.6464 |
| Occiglot-7B-EU5 | 0.5307 | 0.7900 | 0.7267 | 0.5467 | 0.6485 |
### German
| **model_name** | **arc_challenge** | **hellaswag** | **belebele** | **mmlu** | **avg** |
|--------------------------|-------------------|---------------|--------------|----------|---------|
| Mistral-7B-v0.1 | 0.4765 | 0.6101 | 0.7411 | 0.5274 | 0.5888 |
| leo-mistral-hessianai-7b | 0.4739 | 0.6818 | 0.6900 | 0.4887 | 0.5836 |
| Occiglot-7B-EU5 | 0.4944 | 0.6667 | 0.6467 | 0.4833 | 0.5728 |
### Spanish
| **model_name** | **arc_challenge** | **hellaswag** | **belebele** | **mmlu** | **avg** |
|--------------------------|-------------------|---------------|--------------|----------|---------|
| Mistral-7B-v0.1 | 0.5256 | 0.6728 | 0.7478 | 0.5432 | 0.6224 |
| leo-mistral-hessianai-7b | 0.4436 | 0.5970 | 0.6178 | 0.4359 | 0.5236 |
| Occiglot-7B-EU5 | 0.5085 | 0.7255 | 0.6778 | 0.4997 | 0.6029 |
### French
| **model_name** | **arc_challenge** | **hellaswag** | **belebele** | **mmlu** | **avg** |
|--------------------------|-------------------|---------------|--------------|----------|---------|
| Mistral-7B-v0.1 | 0.5244 | 0.6651 | 0.7744 | 0.5413 | 0.6263 |
| leo-mistral-hessianai-7b | 0.4354 | 0.5967 | 0.6222 | 0.4326 | 0.5217 |
| Occiglot-7B-EU5 | 0.5064 | 0.7125 | 0.6756 | 0.4959 | 0.5976 |
### Italian
| **model_name** | **arc_challenge** | **hellaswag** | **belebele** | **mmlu** | **avg** |
|--------------------------|-------------------|---------------|--------------|----------|---------|
| Mistral-7B-v0.1 | 0.4979 | 0.6303 | 0.7356 | 0.5372 | 0.6002 |
| leo-mistral-hessianai-7b | 0.4328 | 0.5580 | 0.5967 | 0.4311 | 0.5047 |
| Occiglot-7B-EU5 | 0.5013 | 0.7008 | 0.6522 | 0.4949 | 0.5873 |
## Acknowledgements
The model training was supported by a compute grant at the [42 supercomputer](https://hessian.ai/) which is a central component in the development of [hessian AI](https://hessian.ai/), the [AI Innovation Lab](https://hessian.ai/infrastructure/ai-innovationlab/) ([HMWK](https://wissenschaft.hessen.de) & [HMinD](https://innen.hessen.de)) and the [AI Service Centers](https://hessian.ai/infrastructure/ai-service-centre/) ([BMBF](https://www.bmbf.de/bmbf/en/home/home_node.html)).
The curation of the training data is partially funded by the [German Federal Ministry for Economic Affairs and Climate Action (BMWK)](https://www.bmwk.de/Navigation/EN/Home/home.html)
through the project [OpenGPT-X](https://opengpt-x.de/en/) (project no. 68GX21007D).
## License
[Apache 2.0](https://www.apache.org/licenses/LICENSE-2.0.html)
|