Kapil

Data Scientist/MLOPS Engineer at @Verisk || NIT Raipur

[Elon Musk @elonmusk]

We should send rockets not at each other, but rather to the stars

[Kapil @kappu2512]

#Finetuning helps #LLMs to overcome their limitations such as hallucination, missing context, etc.

Here is a example code on how to fine-tune a large language model like Llama2 on custom datasets using the QLoRA technique.

https://t.co/luYeO6bU81

[Kapil @kappu2512]

How to Reduce the Hallucinations from Large Language Models?

Hallucinations can be a serious problem for LLMs because they can lead to the spread of misinformation, expose confidential information, and create unrealistic expectations about what LLMs can do.

[Kapil @kappu2512]

For Example: User ask question it will first create embedding and extract most closest context from database and pass it into prompt and then LLM generates the output.

[Kapil @kappu2512]

Retrieval Augmented Generation aka RAG

RAG is a powerful tool that combines the benefits of retrieval-based models and generative models.

It involves retrieving relevant information from a large database and using that information to generate a response or solution.

[Kapil @kappu2512]

While both RAG and Prompt Engineering contribute to problem-solving. RAG focuses on retrieving relevant information from a database, while Prompt Engineering focuses on guiding the model's generation process through well-crafted prompts.

[Kapil @kappu2512]

Prompt

Positive This is awesome!
This is bad! Negative
Wow that movie was rad!
Positive
What a horrible show! --

[Kapil @kappu2512]

Prompt

A prompt is simply the input text given to an LLM, and is meant to guide the model to generate the desired output.

A prompt can consist of multiple components:

1. Instructions or Description
2. External information or context
3. User input or query
4. Output indicator

[Kapil @kappu2512]

Example:

"Classify the text into neutral, negative or positive.
Text: I think the vacation is okay.
Sentiment:"

2. Few Shot prompt: The model is given small number of examples (few shots) related to task before attempting it.

[Kapil @kappu2512]

For example the public IP address of the new instance is written to a local file ip_address.txt.

3. Remote-exec Provisioner: The remote-exec Provisioner runs commands on the remote machine. It connects via SSH or WinRM.

[Kapil @kappu2512]

In Terraform, "provisioners" are used as a last resort to execute scrips or action on a local or remote machine to prepare servers or other infrastructure objects.

[Kapil @kappu2512]

This provisioners is used to copy files from your local machine to the remote machine.

2. Local-exec Provisioners: This provisioner is used to execute a command on the machine where Terraform is being run. This could be to Runa cleanup or an initialisation script.

[Kapil @kappu2512]

8. Terraform state:

Performs operations on the Terraform state, such as listing resources or moving resources to a different state file.

terraform state list

[Kapil @kappu2512]

Below are the some useful commands in terraforms.

1. Terraform init: Initializes a Terraform working directory by downloading the necessary provider plugins and setting up the backend.

terraform init

[Kapil @kappu2512]

terraform validate

6. terraform fmt:

Rewrites Terraform configuration files to a canonical format with consistent indentation and formatting.

terraform fmt

7. Terraform show:

Outputs the current state or a saved plan in a human-readable format.

terraform show