LARGE LANGUAGE MODELS - Prompt Engineering

LLM Special Tokens

The ** symbol is used in LLM prompts to indicate a special token that should be replaced with a specific value. These tokens are often used to control the behavior of LLMs and modify the generated text. They are a key component in building templates to shape LLM responses.

For example, the **temperature** parameter can be set using a token in the prompt. The temperature parameter controls the randomness of the generated text. A higher temperature value will result in more random and diverse outputs, while a lower temperature value will result in more predictable and conservative outputs.

Here is an example of a prompt that uses the temperature token:

Generate a paragraph about artificial intelligence with a temperature of **0.5**.

This prompt would generate a paragraph about artificial intelligence with a temperature of 0.5.

Special tokens can be used to template the output to replace sections where the special token appears. For example **adjective1** in a sentence would be replaced by an adjective keeping the rest of the sentence as entered.

Here are some examples of special tokens that can be used in prompts:

• **adjective1**, **adjective2**, **adjective3**: These tokens can be used to prompt for adjectives. For example, “Roses are **adjective1**” might produce “Roses are red” using the **adjective1** token to prompt for an adjective where the word “red” appears in the output.
• **noun1**, **noun2**, **noun3**: These tokens can be used to prompt for nouns. For example, “The **noun1** sat on the mat” might produce “The cat sat on the mat” using the noun1 token to prompt for a noun.
• **verb1**, **verb2**, **verb3**: These tokens can be used to prompt for verbs. For example, “She **verb1** to catch the bus” might produce “She ran to catch the bus” using the verb1 token to prompt for a verb.
• **plural_noun**: This token can be used to prompt for a plural noun. For example, “Roses are red, violets are blue, sugar is sweet, and so are **plural_noun**” might produce “Roses are red, violets are blue, sugar is sweet, and so are puppies” uses the plural_noun token to prompt for a plural noun.
• **number**: This token can be used to prompt for a number. For example, “I have been to **number** different countries” might produce “I have been to three different countries” uses the number token to prompt for a number.

These special tokens can be used in prompts to generate creative and engaging content using LLMs. By using tokens in prompts, prompt engineers can fine-tune LLMs to generate high-quality and relevant outputs.

Using Special Tokens – A more Advanced Example - MadLibs

Special tokens are used to create prompt patterns and templates. We shall explore using multiple special tokens in a single LLM prompt to create a prompt pattern and play the MadLibs game.

This is an example of a MadLib:

Roses are adjective1,
Violets are adjective2,
Sugar is adjective3,
And so are plural_noun.

In the game you would just fill in the blanks with your own words and enjoy the silly poem!

In a prompt we use template to direct the LLM, which ends up being almost exactly the same as the example above. For example:

“Using the following Madlib template create a funny poem: Roses are **adjective1**,  Violets are **adjective2**,  Sugar is **adjective3**,  And so are **plural_noun**. “

This prompt might produce:

“Roses are red,
Violets are blue,
Sugar is sweet,
And so are puppies.”

...Place Holders for Prompt Engineering....