Model Type | |
Use Cases |
Areas: | |
Applications: | general purpose AI systems |
|
Primary Use Cases: | memory/compute constrained environments, latency bound scenarios, strong reasoning, accelerate research on language and multimodal models |
|
Limitations: | not designed for all downstream purposes |
|
Considerations: | developed with mapping, measuring, and mitigating risks in mind |
|
|
Supported Languages | Arabic (supported), Chinese (supported), Czech (supported), Danish (supported), Dutch (supported), English (supported), Finnish (supported), French (supported), German (supported), Hebrew (supported), Hungarian (supported), Italian (supported), Japanese (supported), Korean (supported), Norwegian (supported), Polish (supported), Portuguese (supported), Russian (supported), Spanish (supported), Swedish (supported), Thai (supported), Turkish (supported), Ukrainian (supported) |
|
Training Details |
Data Sources: | publicly available documents, synthetic data |
|
Data Volume: | |
Methodology: | supervised fine-tuning, proximal policy optimization, and direct preference optimization |
|
Context Length: | |
Training Time: | |
Hardware Used: | |
Model Architecture: | dense decoder-only Transformer |
|
|
Safety Evaluation |
Methodologies: | red-teaming, adversarial tests |
|
Findings: | models may refuse undesirable outputs in English even in other languages, susceptible to longer multi-turn jailbreak techniques |
|
Risk Categories: | |
Ethical Considerations: | invest in quality safety datasets across multiple languages |
|
|
Responsible Ai Considerations |
Fairness: | language models can be unfair or offensive |
|
Transparency: | developers should inform users they are interacting with an AI |
|
Accountability: | developers should fine-tune models for specific uses |
|
Mitigation Strategies: | apply responsible AI best practices |
|
|
Input Output |
Input Format: | |
Accepted Modalities: | |
Output Format: | |
Performance Tips: | testing in specific AI applications is encouraged |
|
|
Release Notes |
Version: | |
Date: | |
Notes: | update over June 2024 release, improved multilingual and reasoning capability |
|
|
|