Meta Llama 3 70B Instruct is an open-source language model by rajkosto. Features: 70b LLM, VRAM: 40.3GB, Context: 8K, License: other, Quantized, Instruction-Based, LLM Explorer Score: 0.12.
Meta Llama 3 70B Instruct Parameters and Internals
Model Type
text generation, code generation
Use Cases
Areas:
Commercial, Research
Applications:
Dialogue use cases, Natural language generation
Primary Use Cases:
Assistant-like chat, Adaptation for various tasks
Limitations:
Use in non-English languages bypassing the license, Violation of laws or use policies
Considerations:
Developers can fine-tune for other languages following the license and use policy.
Additional Notes
In developing these models, the goal was to optimize for helpfulness and safety. Emphasis on community contributions for improving the Llama technology.
Supported Languages
English (Primary)
Training Details
Data Sources:
Publicly available online data
Data Volume:
15 trillion tokens
Methodology:
Supervised fine-tuning and reinforcement learning with human feedback
Hardware Used:
Meta's Research SuperCluster, Third-party cloud compute
Model Architecture:
Auto-regressive language model with transformer architecture, tuned using SFT and RLHF.
Safety Evaluation
Methodologies:
Red-teaming, Adversarial evaluations
Findings:
Residual risks remain, mitigations implemented
Risk Categories:
Cybersecurity, Child Safety
Ethical Considerations:
Developed considering responsible AI practices, with guides and resources provided to the community.
Responsible Ai Considerations
Mitigation Strategies:
Llama Guard and Code Shield safeguards provided for safety mitigation.
Input Output
Input Format:
Text
Accepted Modalities:
Text
Output Format:
Text, Code
Release Notes
Version:
April 18, 2024
Notes:
Meta released Llama 3 models optimized for helpfulness and safety in two sizes, 8B and 70B parameters.