Llama-2-7b-chat-hf
| Property | Value |
|---|---|
| Parameter Count | 6.74B |
| Model Type | Chat-optimized Language Model |
| Training Data | 2.0T Tokens |
| License | Meta Custom Commercial License |
| Context Length | 4k tokens |
What is Llama-2-7b-chat-hf?
Llama-2-7b-chat-hf is Meta's chat-optimized version of their foundational language model, specifically designed for dialogue applications. This model represents the smallest variant in the Llama 2 family, optimized for production deployment while maintaining strong performance across various tasks.
Implementation Details
The model utilizes an optimized transformer architecture, trained on 2 trillion tokens of publicly available data. It implements supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF) to enhance its dialogue capabilities and safety features.
- FP16 tensor format for efficient deployment
- 4k token context window
- Trained with a learning rate of 3.0 x 10-4
- Requires specific formatting with INST and SYS tags for optimal performance
Core Capabilities
- Achieves 45.3% accuracy on MMLU benchmarks
- Demonstrates 33.29% truthfulness in TruthfulQA evaluations
- Shows strong performance in code generation with 16.8% pass@1 score
- Exhibits enhanced safety features with only 21.25% toxicity rate in ToxiGen
Frequently Asked Questions
Q: What makes this model unique?
This model stands out for its optimized balance between performance and resource requirements, making it ideal for production deployment. It incorporates specific safety features and dialogue optimization while maintaining commercial usability.
Q: What are the recommended use cases?
The model is specifically designed for English language assistant-like chat applications. It excels in dialogue scenarios and can be used for various natural language generation tasks, though it should be implemented with appropriate safety testing for specific applications.
