Llama 2 Chat Template

Replicate

Prompt Template Best Practices for Llama 2 Chat Models

Introduction

The Llama2 chat models, developed by Meta AI, follow a specific template for prompt engineering in a chat-style interface. This template involves using tags (e.g., INST) and adhering to a particular structure. By adhering to these best practices, users can optimize their prompts and effectively interact with the Llama2 models.

Template Components

INST Tag

The INST tag marks the beginning of a new instruction or input to the model. It is followed by a colon and a space, and then the instruction itself. For example:

``` INST: Translate the following sentence into Spanish: ```

Textual Input

The textual input follows the INST tag and provides the content or instructions for the model to process. For example:

``` The quick brown fox jumps over the lazy dog. ```

Optional Tags

In addition to the INST tag, users may also use optional tags to provide additional information or control the model's behavior. These tags include:

* `[USER]`: Indicates that the text following the tag is from the user. * `[MODEL]`: Indicates that the text following the tag is a response generated by the model. * `[SKIP]`: Skips the current instruction and moves on to the next one.

Example Prompt

Here is an example of a multi-turn Llama-2-chat prompt using the template:

``` INST: Greet the user with a friendly message: Hi there! Welcome to the Llama2 chat. How can I help you today? [USER]: I'm interested in learning about the prompt template for Llama2 models. INST: Give the user a brief overview of the prompt template: The prompt template for Llama2 models consists of three components: INST, text, and optional tags. You can use this template to instruct the model and provide additional information. [USER]: Can you provide an example of a prompt using the template? INST: Demonstrate a simple prompt using the template: INST: Translate the following sentence into French: The cat sat on the mat. [USER]: What is the purpose of the INST tag? INST: Explain the purpose and structure of the INST tag: The INST tag marks the beginning of a new instruction or input to the model. It consists of the INST keyword, a colon, and a space, followed by the instruction itself. ```

Conclusion

By adhering to the prompt template best practices for Llama2 chat models, users can optimize their prompts for effective interaction. The use of tags, such as INST, and a consistent structure ensures that the models receive clear instructions and can deliver accurate and relevant responses.

Medium