Hello, and welcome to "Introduction to Large Language Models". My name is John Ewald and I am a Training Developer here at Google Cloud.

In this course, you learn to: Define Large Language Models (LLMs) Describe LLM Use Cases Explain Prompt Tuning Describe Google’s Gen AI Development tools

Large Language Models (or LLMs) are a subset of Deep Learning. To find out more about Deep Learning, see our Introduction to Generative AI course video. LLMs and Generative AI intersect and they are both a part of Deep Learning.

Another area of AI you may be hearing a lot about is generative AI. This is a type of artificial intelligence that can produce new content –including text images, audio, and synthetic data.

So, what are large language models?

Large language models refer to large, general-purpose language models that can be pre-trained and then fine-tuned for specific purposes. What do pre-trained and fine-tuned mean?

Imagine training a dog. Often you train your dog basic commands such as sit, come, down, and stay. These commands are normally sufficient for everyday life and help your dog become a good canine citizen.

However, if you need a special-service dog such as a police dog, a guide dog, or a hunting dog, you add special trainings.

This similar idea applies to large language models.

These models are trained for general purposes to solve common language problems such as text classification, question answering, document summarization, and text generation across industries.

The models can then be tailored to solve specific problems in different fields such as retail, finance, and entertainment, using a relatively small size of field datasets.

Let’s further break down the concept into three major features of large language models. Large indicates two meanings, First is the enormous size of the training dataset, sometimes at the petabyte scale. Second it refers to the parameter count (in ML, parameters are often called hyperparameters). Parameters are basically the memories and the knowledge that the machine learned from the model training. Parameters define the skill of a model in solving a problem, such as predicting text. General-purpose means that the models are sufficient to solve common problems. Two reasons lead to this idea: First is the commonality of a human language regardless of the specific tasks, and second is the resource restriction. Only certain organizations have the capability to train such large language models with huge datasets and a tremendous number of parameters. How about letting them create fundamental language models for others to use? This leads to the last point, pre-trained and fine-tuned, meaning to pre-train a large language model for a general purpose with a large dataset and then fine-tune it for specific aims with a much smaller dataset.

The benefits of using large language models are straightforward: First, a single model can be used for different tasks. This is a dream come true. These large language models that are trained with petabytes of data and generate billions of parameters are smart enough to solve different tasks including language translation, sentence completion, text classification, question answering, and more. Second, large language models require minimal field training data when you tailor them to solve your specific problem. Large language models obtain decent performance even with little domain training data. In other words, they can be used for few-shot or even zero-shot scenarios. In machine learning, "few-shot" refers to training a model with minimal data and "zero-shot" implies that a model can recognize things that have not explicitly been taught in the training before. Third, the performance of large language models is continuously growing when you add more data and parameters.

Let’s take PaLM as an example. In April 2022, Google released PaLM(short for Pathways Language Model), a 540 billion-parameter model that achieves a state-of-the-art performance across multiple language tasks.

PaLM is a dense decoder-only Transformer model. It has 540 billion parameters. It leverages the new Pathways system, which enabled Google to efficiently train a single model across multiple TPU v4 Pods. Pathway is a new AI architecture that will handle many tasks at once, learn new tasks quickly, and reflect a better understanding of the world. The system enables PaLM to orchestrate distributed computation for accelerators.

We previously mentioned that PaLM is a transformer model. A Transformer model consists of encoder and decoder. The encoder encodes the input sequence and passes it to the decoder, which learns how to decode the representations for a relevant task.

We’ve come a long way from traditional programming, to neural networks, to generative models! In traditional programming, we used to have to hard code the rules for distinguishing a cat - type: animal, legs: 4, ears: 2, fur: yes, likes: yarn, catnip.

In the wave of neural networks, we could give the network pictures of cats and dogs and ask - "Is this a cat" - and it would predict a cat.

In the generative wave, we - as users - can generate our own content - whether it be text, images, audio, video, etc. For example, models like PaLM (or Pathways Language Model or LaMDA (or Language Model for Dialogue Applications) ingest very, very large data from multiple sources across the Internet and build foundation language models we can use simply by asking a question - whether typing it into a prompt or verbally talking into the prompt. We - as users - can use these language models to generate text or answer questions or summarize data, among other things. So, when you ask it "what’s a cat", it can give you everything it has learned about a cat.

Let’s compare LLM Development using pre-trained models with traditional ML development. First, with LLM Development, you don’t need to be an expert. You don’t need training examples, and there is no need to train a model. All you need to do is think about prompt design, which is the process of creating a prompt that is clear, concise, and informative. It is an important part of natural language processing (NLP). In traditional machine learning, you need expertise, training examples, train a model and compute time and hardware.

Let’s take a look at an example of a Text Generation use case.

Question answering (QA) is a subfield of natural language processing that deals with the task of automatically answering questions posed in natural language. QA systems are typically trained on a large amount of text and code, and they are able to answer a wide range of questions, including factual, definitional, and opinion-based questions. The key here is that you needed domain knowledge to develop these Question Answering models.

For example, domain knowledge is required to develop a question answering model for customer IT Support, or Healthcare or Supply Chain.

Using Generative QA, the model generates free text directly based on the context. There is no need for domain knowledge.

Let’s look at three questions given to Bard, a large language model chatbot developed by Google AI.

Question 1 - This year’s sales are 100,000 dollars. Expenses are 60,000 dollars. How much is net profit? Bard first shares how net profit is calculated then performs the calculation. Then, Bard provides the definition of net profit.

Here is another question: Inventory on hand is 6,000 units. New order requires 8,000 units. How many units do I need to fill to complete the order? Again, Bard answers the question by performing the calculation.

And our last example, We have 1,000 sensors in ten geographic regions. How many sensors do we have on average in each region? Bard answers the question with an example on how to solve the problem and some additional context.

In each of my questions, a desired response was obtained. This is due to Prompt Design.

Prompt design and prompt engineering are two closely related concepts in natural language processing. Both involve the process of creating a prompt that is clear, concise, and informative. However, there are some key differences between the two. Prompt design is the process of creating a prompt that is tailored to the specific task that the system is being asked to perform. For example, if the system is being asked to translate a text from English to French, the prompt should be written in English and should specify that the translation should be in French. Prompt engineering is the process of creating a prompt that is designed to improve performance. This may involve using domain-specific knowledge, providing examples of the desired output, or using keywords that are known to be effective for the specific system. In general, prompt design is a more general concept, while prompt engineering is a more specialized concept. Prompt design is essential for, while prompt engineering is only necessary for systems that require a high degree of accuracy or performance.

There are three kinds of Large Language Models - Generic Language Models, Instruction Tuned, and Dialog Tuned. Each needs prompting in a different way. Generic Language Models predict the next word (technically token)based on the language in the training data.

This is an example of a generic language model - The next word is a token based on the language in the training data. In this example, the cat sat on —- the next word should be "the" and you can see that "the" is the most likely next word. Think of this type as an "auto-complete" in search.

In Instruction tuned, the model is trained to predict a response to the instructions given in the input.

For example, summarized a text of "x", generate a poem in the style of ‘x", give me a list of keywords based on semantic similarity for "x".

And in this example, classify the text into neutral, negative or positive.

In Dialog tuned, the model is trained to have a dialog by the next response.

Dialog-tuned models are a special case of instruction tuned where requests are typically framed as questions to a chat bot. Dialog tuning is expected to be in the context of a longer back and forth conversation, and typically works better with natural question-like phrasings.

Chain of thought reasoning is the observation that models are better at getting the right answer when they first output text that explains the reason for the answer. Let’s look at the question: Roger has 5 tennis balls. He buys 2 more cans of tennis balls. Each can has 3 tennis balls. How many tennis balls does he have now? This question is posed initially with no response. The model is less likely to get the correct answer directly. However, by the time the second question is asked, the output is more likely to end with the correct answer.

A model that can do everything has practical limitations. Task-specific tuning can make LLMs more reliable.

Vertex AI provides task specific foundation models. Let’s say you have a use case where you need to gather sentiments (or how your customers are feeling about your product or service), you can use the classification task sentiment analysis task model. Same for vision tasks - if you need to perform occupancy analytics, there is a task specific model for your use case.

Tuning a model enables you to customize the model response based on examples of the task that you want the model to perform. It is essentially the process of adapting a model to a new domain or set of custom use cases by training the model on new data. For example, we may collect training data and "tune" the model specifically for the legal or medical domain.

You can also further tuned the model by "fine-tuning", where you bring your own dataset and retrain the model by tuning every weight in the LLM. This requires a big training job and hosting your own fine-tuned model.

Here is an example of a medical foundation model trained on Healthcare data. The tasks include question answering, image analysis, finding similar patients, etc.

Fine-tuning is expensive and not realistic in many cases. So, are there more efficient methods of tuning?

Yes. Parameter-Efficient Tuning Methods are methods for tuning a large language model on your own custom data without duplicating the model. The base model itself is not altered. Instead, a small number of add-on layers are tuned, which can be swapped in and out at inference time.

Generative AI Studio lets you quickly explore and customize generative AI models that you can leverage in your applications on Google Cloud. Generative AI Studio helps developers create and deploy generative AI models by providing a variety of tools and resources that make it easy to get started. For example, there is a: Library of pre-trained models Tool for fine-tuning models Tool for deploying models to production Community forum for developers to share ideas and collaborate Generative AI App Builder lets you create Gen AI apps without having to write any code. Gen AI App Builder has A: Drag-and-drop interface that makes it easy to design and build apps. Visual editor that makes it easy to create and edit app content. Built-in search engine that allows users to search for information within the app. Conversational AI engine that allows users to interact with the app using natural language. You can create your own: Chatbots Digital assistants Custom search engines Knowledge bases Training applications And more PaLM API let’s you test and experiment with Google’s Large Language Models and Gen AI tools. To make prototyping quick and more accessible, developers can integrate PaLM API with MakerSuite and use it to access the API using a graphical user interface. The suite includes a number of different tools, such as a model training tool, a model deployment tool, and a model monitoring tool. The model training tool helps developers train ML models on their data using different algorithms. The model deployment tool helps developers deploy ML models to production with a number of different deployment options. The model monitoring tool helps developers monitor the performance of their ML models in production using a dashboard and a number of different metrics.