Here, we are doomed again.

llm is subset of deep learning.

LLM refer to large , general-purpose language models that can be pre-trained and then fine-tuend for specific purposes.

LLM are trained for solving the common language problems like text classification, Question Answering, Document Summarization, Text Generation.

Then LLM can be fine tuned with small data.

Large : large training dataset ( petabytes ) and Large numver of parameters.

LLM use cases.

PaLM : pathways Language madel, 340 billion parameters.

new pathway architecture which can handle different tasks.

LLM developement Vs. Traditional Dev

1. No ML expertise needed
2. no traing examples
3. No need to train a model
4. Just think about the Prompt Desing

Traditional dev

ML expertise, training examples, model, compute time and hardware, minimizing the loss funciton.

Prompt Design Vs. Prompt Engineering

Prompt Desing is the process of creating a prompt that is tailored for the specific task that is asked to perform

Prompt Engineering : is the process of creating the prompt that is desinged to improve the performance.

Its like How to do specific thing and Improving the performance Or accuracy for the same.

3 Types of LLM Models

Generic Model : is something like a next word predictor. ( seach completion )

Instruction Tuned LM : Tuned for perfoming a task. ( generate a poem in style of x )

Dialog-Tuned LM : these are special Instruction Tuned LMs, to context is gathered.

Chain-of-thoughts .

MORE efficient Tunning Model

Parameter-efficient tuning method ( PETM ) : tuning the existing model on custom data without duplicating the model.