| Model Type | | bilingual, text generation, instruction fine-tuned, decoder, causal-lm |
|
| Use Cases |
| Areas: | |
| Applications: | | Arabic NLP, Chat applications, Sentiment analysis, Summarization |
|
| Primary Use Cases: | | Chat assistants, Arabic cultural phenomena, Quantitative studies, Mechanistic interpretability |
|
| Limitations: | | Malicious use, Sensitive Information, Generalization across all languages, High-stakes decisions |
|
| Considerations: | | Models should comply with applicable laws and not violate sensitive information handling. |
|
|
| Additional Notes | | Extensible approaches for adapting low and medium resource languages. |
|
| Supported Languages | | Arabic (MSA) (bilingual), English (bilingual) |
|
| Training Details |
| Data Sources: | | Web, Code, Books, Scientific, Synthetic |
|
| Data Volume: | |
| Methodology: | | Auto-regressive, Transformer-based, Decoder-only |
|
| Context Length: | |
| Hardware Used: | | Cerebras CS-2 Wafer-Scale Engines, Condor Galaxy supercomputer |
|
| Model Architecture: | | SwiGLU, ALiBi position encoding, RoPE, Grouped Query Attention |
|
|
| Input Output |
| Input Format: | |
| Accepted Modalities: | |
| Output Format: | |
| Performance Tips: | | Enable `trust_remote_code=True` when loading the model. |
|
|
| Release Notes |
| Version: | |
| Date: | |
| Notes: | | Initial release of Jais Family models ranging from 590M to 70B parameters. |
|
|
|
