You can find Part 2 here.

If you are a developer looking to add language processing capabilities to your project, this guide is for you. We’ll give you a quick tour of what’s possible with language AI using Cohere’s Large Language Model (LLM) API. Our guide is essentially your “Hello, World!” of language AI, and since this is all about language processing, we’ll do this by exploring the phrase Hello, World! itself!

We’ll cover three groups of tasks that you will typically work on when dealing with language data, including:

Generating text
Classifying text
Analyzing text

These map nicely to three Cohere endpoints — Chat, Classify, and Embed — which we will cover in the examples. We’ll cover the first in Part 1 and the other two in Part 2.

*The three groups of language AI tasks we will cover*

This guide assumes little or no background in machine learning or NLP. The examples are shown using Python. We include only key snippets of the code used, but you can run the full version if you wish.

Setting up

The first step is to install the Cohere Python SDK.

$ pip install cohere

Next, sign up for a Cohere account and create an API key, which you can generate from one of the following interfaces:

Cohere dashboard
Cohere CLI tool

Once that is done, you can set up the Cohere client as follows.

import cohere
co = cohere.Client("COHERE_API_KEY")

1 - Generating text

There are two broad themes in language processing — language generation and language understanding. This part covers the former, which is the capability that you can access by calling Cohere’s Chat endpoint.

Cohere’s language AI endpoints produce different types of outputs, but there is a similar pattern in how you work with them. This can be broken into three steps:

Prepare input
Define model settings
Get output

Prepare input

The Chat endpoint generates text given an input, called a “prompt.”

The prompt provides a context for the text that we want the model to generate. To illustrate this, let’s start with a simple prompt as the input.

prompt = "What is a Hello World program."

Define model settings

Next, we’ll define some model settings. There are a number of settings that you can modify with the Chat endpoint, but we’ll start with a few:

model — we’ll use the default which is command. Visit the API Reference for other available model types.
prompt — the input string to be used.
temperature — Lower temperature will cause the model to output text that is more predictable, while higher temperature means that the output will be more creative. There is no one right setting as it depends on your task and it requires some experimentation.

Generate output

Finally, we’ll generate the output, giving us the generations response. Putting everything together, we get the code below.

prompt = "What is a Hello World program."

response = co.chat(
  message=prompt,
  model='command-r')

print(response.text)

Here’s a sample output returned:

A "Hello World" program is a traditional way for beginners to start learning a new programming language. The program simply displays the text "Hello World" on the screen, usually in a command-line interface. The simplicity of this task makes it an ideal starting point for newcomers, as it requires only a basic understanding of syntax and program structure.

The concept of a "Hello World" program ...

The output is not bad, but it could be better. If we were writing a blog post for example, the tone of the generated output wouldn’t fit very well. Also, there is no natural ending to the generated text, which looks like it could continue for quite some time.

We need to find a way to make the output tighter and closer to how we want it to be, which is where we leverage prompt engineering.

Create a better prompt

Prompt engineering is a fascinating topic. It is about figuring out the optimal way to prompt a model for a particular task, so we can shape the output to be how we want it to be. There are many creative prompts that have proven effective out there, and the whole topic is an active area of research. Having said that, the basic format that generally works well includes an instruction followed by a few examples of instruction and response.

Let’s say we are writing a blog about Hello, World! Specifically, we want to write an introductory paragraph about Learning to Code with Hello, World!

The prompt is as follows:

prompt = """
Write the first paragraph of a blog post given a blog title.
--
Blog Title: Best Activities in Toronto
First Paragraph: Looking for fun things to do in Toronto? When it comes to exploring Canada's
largest city, there's an ever-evolving set of activities to choose from. Whether you're looking to
visit a local museum or sample the city's varied cuisine, there is plenty to fill any itinerary. In
this blog post, I'll share some of my favorite recommendations
--
Blog Title: Mastering Dynamic Programming
First Paragraph: In this piece, we'll help you understand the fundamentals of dynamic programming,
and when to apply this optimization technique. We'll break down bottom-up and top-down approaches to
solve dynamic programming problems.
--
Blog Title: Learning to Code with Hello, World!
First Paragraph:"""

Using the new prompt, our generation code now looks like this:

response = co.chat(
  message=prompt,
  model='command-r')

print(response.text)

Here’s a sample output returned:

Starting to code can be daunting, but it's actually simpler than you think! The famous first program, "Hello, World!" is a rite of passage for all coders and a great starting point to begin your coding journey. This blog will take you through the process of writing your very first line of code, choosing the right programming language, and understanding the logic behind those seemingly complicated lines of code. It's a terrific gateway into the world of programming, and before you know it, you'll be on your way to becoming a programming pro!

This looks much better! Because we have created a couple of examples with the tone of a blog post, the model captured that context and was then able to generate something with a similar tone.

Automate the process

In actual applications, you will likely need to produce these text generations on an ongoing basis, given different inputs. Let’s simulate that with our example.

*Automating generations from multiple prompts*

First, we create a list of new topics, so we can iterate on them and get the paragraphs generated.

topics = ["How to Grow in Your Career",
          "The Habits of Great Software Developers",
          "Ideas for a Relaxing Weekend"]

Then we set up the function to generate text.

def generate_text(topic):
  prompt = f"""
Write the first paragraph of a blog post given a blog title.
--
Blog Title: Best Activities in Toronto
First Paragraph: Looking for fun things to do in Toronto? When it comes to exploring Canada's
largest city, there's an ever-evolving set of activities to choose from. Whether you're looking to
visit a local museum or sample the city's varied cuisine, there is plenty to fill any itinerary. In
this blog post, I'll share some of my favorite recommendations
--
Blog Title: Mastering Dynamic Programming
First Paragraph: In this piece, we'll help you understand the fundamentals of dynamic programming,
and when to apply this optimization technique. We'll break down bottom-up and top-down approaches to
solve dynamic programming problems.
--
Blog Title: {topic}
First Paragraph:"""
  # Generate text by calling the Chat endpoint
  response = co.chat(
    message=prompt,
    model='command-r')

  return response.text

And finally, we generate the paragraphs.

paragraphs = []

for topic in topics:
  paragraphs.append(generate_text(topic))
  
for topic,para in zip(topics,paragraphs):
  print(f"Topic: {topic}")
  print(f"First Paragraph: {para}")
  print("-"*10)

And here’s a sample output returned:

Topic: How to Grow in Your Career
First Paragraph: Advancing in your career can seem like a daunting task, especially if you're unsure of the path ahead. In this ever-changing professional landscape, there are numerous factors to consider. This blog aims to shed light on the strategies and skills that can help you navigate the complexities of career progression and unlock your full potential. Whether you're looking to secure a promotion or explore new opportunities, these insights will help you chart a course for your future. Let's embark on this journey of self-improvement and professional growth, equipping you with the tools to succeed in your career aspirations.
----------
Topic: The Habits of Great Software Developers
First Paragraph: Great software developers are renowned for their ability to write robust code and create innovative applications, but what sets them apart from their peers? In this blog, we'll delve into the daily habits that contribute to their success. From their approach to coding challenges to the ways they stay organized, we'll explore the routines and practices that help them excel in the fast-paced world of software development. Understanding these habits can help you elevate your own skills and join the ranks of these industry leaders.
----------
Topic: Ideas for a Relaxing Weekend
First Paragraph: Life can be stressful, and sometimes we just need a relaxing weekend to unwind and recharge. In this fast-paced world, taking some time to slow down and rejuvenate is essential. This blog post is here to help you plan the perfect low-key weekend with some easy and accessible ideas. From cozy indoor activities to peaceful outdoor adventures, I'll share some ideas to help you renew your mind, body, and spirit. Whether you're a homebody or an adventure seeker, there's something special for everyone. So, grab a cup of tea, sit back, and get ready to dive into a calming weekend of self-care and relaxation!
----------

We have covered one example, but the ways to construct a prompt are only limited by your creativity. It is also highly dependent on the task at hand. Prompt engineering can be used not only in text completion, but also to do any form of text generation, such as summarization, rewriting, question answering, extraction, and many more.

In Part 2, we’ll complete our tour by looking at how we can use Language AI to classify and analyze text. In the meantime, to try out the Chat endpoint for yourself, sign up for a Cohere account now.