Llama-2-13B-fp16
| Property | Value |
|---|---|
| Parameter Count | 13 Billion |
| Training Tokens | 2 Trillion |
| Context Length | 4k tokens |
| License | Meta Custom Commercial License |
What is Llama-2-13B-fp16?
Llama-2-13B-fp16 is Meta's advanced language model converted to fp16 precision format by TheBloke. This model represents a significant advancement in the field of large language models, trained on 2 trillion tokens of publicly available data. It's designed to maintain high performance while offering improved efficiency through fp16 precision.
Implementation Details
The model utilizes an optimized transformer architecture, converted from the original PTH files to Hugging Face format using Transformers 4.32.0.dev0. It features a 4k token context window and was trained with a learning rate of 3.0 x 10^-4 and a global batch size of 4M tokens.
- Full fp16 precision support for efficient inference
- Optimized transformer architecture
- Compatible with standard Hugging Face implementations
- 4k token context window for handling longer sequences
Core Capabilities
- Strong performance in commonsense reasoning (66.9% accuracy)
- Effective world knowledge applications (55.4% accuracy)
- Advanced reading comprehension capabilities (65.8% accuracy)
- Improved truthfulness metrics compared to previous versions
Frequently Asked Questions
Q: What makes this model unique?
This model stands out for its balance of size and performance, offering strong capabilities across various tasks while maintaining efficiency through fp16 precision. It shows significant improvements in truthfulness and toxicity metrics compared to its predecessors.
Q: What are the recommended use cases?
The model is best suited for commercial and research applications in English, including text generation, analysis, and general language understanding tasks. It's particularly effective for applications requiring balanced performance across reasoning, knowledge retrieval, and comprehension.
