Llama 2 70b

Introducing the WEB Llama 2 Family of Models

Unlocking New Possibilities in Natural Language Processing

Overview

Google AI and Microsoft are proud to announce the release of the WEB Llama 2 family of pretrained and fine-tuned LLMs. These models offer a range of capabilities and performance levels to meet the diverse needs of researchers and developers.

The WEB Llama 2 family consists of three model sizes: 7B, 13B, and 70B. All models are trained on 2 trillion tokens, double the context length of previous models, allowing them to process and generate more complex and comprehensive text.

Model Specifications

The following table summarizes the key specifications of each model:

| Model Size | Parameters | Token Count | |---|---|---| | 7B | 7 billion | 2 trillion | | 13B | 13 billion | 2 trillion | | 70B | 70 billion | 2 trillion |

Applications

The WEB Llama 2 family of models can be applied to a wide range of natural language processing tasks, including:

* Text generation and translation * Question answering * Summarization * Dialogue systems * Machine reading comprehension

Availability

The WEB Llama 2 models are available for use through the ONNX Runtime. Developers can access the official ONNX Llama 2 repo here and the ONNX Runtime here. Please note that to use the ONNX Llama models, you must have a supported GPU.

Code Llama 70B

In addition to the WEB Llama 2 models, we are also releasing Code Llama 70B, the largest and best-performing model in the Code family. This model is specifically designed for code-related tasks, such as:

* Code generation * Code completion * Bug detection * Code documentation

Conclusion

The WEB Llama 2 family of models and Code Llama 70B represent significant advancements in natural language processing and code-related tasks. These models empower researchers and developers to unlock new possibilities and create innovative applications.