1

AI Free Basic Course | Lecture 33 | Introduction to LlamaIndex |

Live Session

Finally, we are going to delve into the topic of LlamaIndex, a

subject that students have shown great interest in. Following
our agenda, we will provide an introduction to LlamaIndex.
However, it's crucial to note that Llama, LlamaIndex, and Llama-
2 are distinct terms. When these terms are used interchangeably,
it can lead to confusion regarding which specific Llama is under
discussion. Therefore, we will clarify the distinctions of these
terms in such a way that whenever they come up, we can mentally
distinguish which technology is being referred to.

Subsequently, we will conduct a comparative analysis between

Langchain and LlamaIndex, exploring both their similarities and
differences. Finally, we will conclude by elucidating these
concepts through coding demonstrations.

Prepared by: Tahir Mahmood 0321-5111997, What up 0312-8536439

 2

Now, let's dive into the topic of LlamaIndex. Just as we are

aware that Langchain acts as the intermediary between data and
Large Language Models, LlamaIndex falls within the same
category. This means that, similar to Langchain, we can develop
chatbots using LlamaIndex. It's worth noting that LlamaIndex was
previously known as GPTIndex, so don't be puzzled if you come
across the term GPTIndex on certain websites; it's still in use
in some places.

LlamaIndex essentially performs the same role as Langchain, but

it adopts a different approach. LlamaIndex is a Python-based
platform that links your data with the Large Language Model,
making it akin to the younger sibling of Langchain. The key
distinction between them lies in the ease of implementation. We
will explore why LlamaIndex is easier to implement compared to
Langchain in more detail later. For now, it's important to grasp
that LlamaIndex is a technology utilized for chatbot development
and generating responses to our queries by transmitting data to
a large language model.

Our next agenda item pertains to the distinction between

LlamaIndex and Llama-2. Unlike LlamaIndex, which serves as a
tool for chatbot development and data interaction with Large
Language Models, Llama-2, much like GPT3.5, GPT4.0, Colab, or
Bard, is essentially a variant of a Large Language Model. What
sets Llama-2 apart is its development by Facebook and its
availability at no cost. While it can serve as an alternative to
GPT3.5 or GPT4.0, it's essential to be cautious of the
substantial computational resources it demands for smooth
operation

Prepared by: Tahir Mahmood 0321-5111997, What up 0312-8536439

 3

To elaborate further, Llama-2 offers a range of configurations,

starting from Llama-7B, which is the more straightforward
option, to Llama-70B, which is significantly more resource-
intensive. Llama-7B is equipped with 7 billion neurons, while
Llama-70B boasts an impressive 70 billion neurons. This
discrepancy in neuron count directly influences their ability to
perform inferences and generate text. Consequently, to fully
leverage the capabilities of Llama-70B with its extensive neuron
count, a substantial amount of memory is essential. Despite its
cost-free availability, its demanding resource requirements can
pose challenges to effective utilization.

To run Llama-2 smoothly, a considerable amount of RAM is

necessary. For instance, if you're using the 70-B version of
Llama-2, you would require 50 GB of Video RAM, while the 7B
version of Llama-2 demands 13 GB of video RAM. Therefore, if you
have a GPU with 12 GB to 14 GB of video RAM, you can comfortably
utilize the 7-B version of Llama-2, eliminating the need for GPT
licenses and bypassing other formalities typically associated
with GPT usage.
Now that we have a clear understanding of the distinctions
between LlamaIndex and Llama-2, the question arises: What is
Llama all about? Llama was also a Language Model that made its
debut prior to Llama-2, much in the same way that ChatGPT and
ChatGPT3.5 preceded ChatGPT4.0. In a parallel fashion, Facebook
initially introduced Llama as an open-source project. While it
showed promise, it didn't quite have the impressive "wow
factor."

Consequently, Llama-2, a large language model, was developed as

an evolution of the original Llama, following an iterative
process to enhance its capabilities and performance.

As mentioned earlier, Llama-2 is indeed freely available, but it

does not come with an API provided by Facebook/META. Instead,
Facebook/META offers the entire Llama-2 model for your use. This
means that you are responsible for deploying it on a cloud
platform or obtaining it from sources that have already deployed
it in the cloud.

It's worth noting that there are some instances of Llama-2 being
available for free on Hugging Face, but you would need to search
for these specific models on the Hugging Face platform to access
them.

Prepared by: Tahir Mahmood 0321-5111997, What up 0312-8536439

 4

Moving on to our main topic for today, which is LlamaIndex,

let's explore what we can create with the assistance of
LlamaIndex. In essence, LlamaIndex offers the capability to
develop a wide range of applications with significantly less
effort compared to what would be required using Langchain. Many
pre-implemented features and functionalities are readily
available within LlamaIndex, making it a convenient choice for
various projects.

However, there is a trade-off to this convenience. Once these

features have been implemented in LlamaIndex, they cannot be
easily customized or amended. You are bound to use them as they
have been originally implemented. This convenience, while
advantageous, is also a double-edged sword because if you
require modifications, you may need to consider changing the
entire technology stack to accommodate those changes. This is a
crucial aspect to keep in mind when working with LlamaIndex.
Many clients often express interest in developing websites using
WordPress, but I tend not to recommend it for a specific reason.
WordPress heavily relies on plugins for extended functionality,
and this can limit my capabilities as a developer. When a plugin
is installed, it brings along thousands of lines of code that
are not under my direct control. This can potentially lead to
compatibility issues, security vulnerabilities, and performance
concerns.

To educate the client about this aspect of WordPress, it's

essential to provide them with a gentle hint rather than
overwhelming them with every potential problem that may arise.
Instead of detailing every possible issue, you can subtly convey

Prepared by: Tahir Mahmood 0321-5111997, What up 0312-8536439

 5

the idea that using too many plugins can sometimes lead to
unforeseen challenges and potential business disruptions. This
approach allows the client to consider the trade-offs while
making an informed decision.

We cand build variety of applications using LlamaIndex as:

1. Question Answering
2. Summarization of a document
3. Retrieval
4. Search
5. Conversation
In order to implement all these there is need of loading,
splitting, vector store.

As previously discussed, LlamaIndex is primarily tailored for

the development of search and retrieval applications, and its
main focus is efficiency. Therefore, if you prioritize
efficiency over customization, LlamaIndex is the way to go.
However, it's important to note that a drawback of LlamaIndex is
its limited support for customization. For instance, if you wish
to modify the embedding model, this can only be achieved using
Langchain. With LlamaIndex, you're essentially locked into using
the pre-defined OpenAI embedding, as it's hardcoded, and
customization options are restricted.

On the other hand, Langchain can be likened to a Swiss army

knife, encompassing a wide array of essential tools. This is why
Langchain is often referred to as a general-purpose tool. It
offers a versatile range of capabilities, from data loading to
Prepared by: Tahir Mahmood 0321-5111997, What up 0312-8536439
 6

splitting, with numerous options. For example, it provides

various loaders tailored to different data types, such as PDFs,
text, CSVs, Excel files, and even GitHub repositories. Langchain
offers a multitude of possibilities that, unfortunately, are not
available within LlamaIndex.
Following table will explain the difference between LlamaIndex
and Langchain.

Prepared by: Tahir Mahmood 0321-5111997, What up 0312-8536439