Model Type | pretrained, instruction tuned, generative text models, language model |
|
Use Cases |
Areas: | commercial use, research use |
|
Primary Use Cases: | |
Limitations: | Use that violates laws or the Acceptable Use Policy |
|
Considerations: | Fine-tuning is encouraged for non-English deployment; safety evaluations required for specific use cases |
|
|
Training Details |
Data Sources: | publicly available online data |
|
Data Volume: | |
Methodology: | supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF) |
|
Context Length: | |
Hardware Used: | Meta's Research SuperCluster, production clusters, third-party cloud compute |
|
Model Architecture: | auto-regressive language model with optimized transformer architecture |
|
|
Safety Evaluation |
Methodologies: | red teaming, adversarial evaluations |
|
Risk Categories: | CBRNE, cyber security, child safety |
|
Ethical Considerations: | Iterative testing for CBRNE threats; use of external expertise for safety evaluation |
|
|
Responsible Ai Considerations |
Fairness: | Considerations for fairness and safety through responsible AI practices |
|
Transparency: | Commitment to transparency in safety implementations |
|
Accountability: | Developers are encouraged to implement safety tools and responsible safety practices |
|
Mitigation Strategies: | Updated Responsible Use Guide, community feedback and engagement |
|
|
Input Output | |
Release Notes |
Date: | |
Notes: | Meta Llama 3 officially released. |
|
|
|