Model Type | |
Use Cases |
Areas: | |
Applications: | Assistant chat, Natural language generation, Synthetic data generation |
|
Primary Use Cases: | Multilingual dialogue, Tool-use integrations, Long context management |
|
Limitations: | Use beyond supported languages without fine-tuning is not recommended, Compliance with acceptable use policy required |
|
Considerations: | Developers should apply safety testing and tuning for their specific applications. |
|
|
Additional Notes | Openly releasing the model allows developers to fine-tune for languages and use-cases beyond those explicitly supported. |
|
Supported Languages | English (Advanced), German (Advanced), French (Advanced), Italian (Advanced), Portuguese (Advanced), Hindi (Advanced), Spanish (Advanced), Thai (Advanced) |
|
Training Details |
Data Sources: | Publicly available online data |
|
Data Volume: | |
Methodology: | Supervised fine-tuning and reinforcement learning with human feedback |
|
Context Length: | |
Hardware Used: | |
Model Architecture: | Optimized transformer architecture |
|
|
Safety Evaluation |
Methodologies: | Red-teaming, Adversarial prompting |
|
Findings: | Model may produce inaccurate, biased or objectionable responses |
|
Risk Categories: | CBRNE, Child Safety, Cyber attack enablement |
|
Ethical Considerations: | Responsible use guidelines should be followed; specific capabilities should be evaluated for safety. |
|
|
Responsible Ai Considerations |
Fairness: | Inclusion of multiple languages, consideration of cultural perspectives. |
|
Transparency: | Extensive documentation and licensing information provided. |
|
Accountability: | Developers are responsible for the safe deployment and compliance with local laws. |
|
Mitigation Strategies: | Safety guidelines and resources are available to developers. |
|
|
Input Output |
Input Format: | |
Accepted Modalities: | |
Output Format: | |
Performance Tips: | Consider tool-use templates and tokenization strategies for large inputs |
|
|
Release Notes |
Version: | |
Date: | |
Notes: | A new collection of generative models optimized for multilingual dialogue with improvements in inference scalability. |
|
|
|