Model Type | text generation, instruction tuned |
|
Use Cases |
Areas: | |
Applications: | assistant-like chat (instruction tuned), natural language generation tasks (pre-trained) |
|
Primary Use Cases: | dialogue, text and code generation |
|
Limitations: | Use violating laws or regulations not allowed, English-only pretraining testing |
|
Considerations: | Users may fine-tune for other languages under the license. |
|
|
Additional Notes | Implemented safety and misuse mitigations; allows model fine-tuning for uses beyond initial capabilities under license conditions. |
|
Supported Languages | |
Training Details |
Data Sources: | publicly available online data |
|
Data Volume: | |
Methodology: | auto-regressive with optimized transformer architecture, RLHF |
|
Context Length: | |
Hardware Used: | Meta's Research SuperCluster, H100-80GB GPUs |
|
Model Architecture: | |
|
Safety Evaluation |
Methodologies: | red teaming, adversarial evaluations |
|
Findings: | Model was extensively tested to lower residual risks |
|
Risk Categories: | CBRNE, cybersecurity, child safety |
|
Ethical Considerations: | misuse mitigation measures in place. |
|
|
Responsible Ai Considerations |
Mitigation Strategies: | Implement safeguards like Llama Guard and Code Shield. |
|
|
Input Output |
Input Format: | |
Accepted Modalities: | |
Output Format: | |
|