DEV Community

Cover image for Article 1 : Intro to Gen AI ,LLMS and LangChain Frameworks(Part B)
Raunak ALI
Raunak ALI

Posted on

Article 1 : Intro to Gen AI ,LLMS and LangChain Frameworks(Part B)

#genai #llm #langchain #python

Chapter B: Introduction to LLMs And Free LLM Resources

1. What Are LLMs (Large Language Models)?

Imagine a system that doesn’t just store information like a database, but can converse, summarize, translate, write code, and even reason through problems. That’s what an LLM (Large Language Model) does.

  • To a business owner: You can think of them as engines that can draft reports, analyze long documents, summarize meetings, or even generate marketing content at scale,cutting both cost and time.
  • To a student or fresher: It’s helpful to imagine them as a much smarter autocomplete. They’ve been trained on massive datasets, so they can predict the “next word” in a way that feels surprisingly natural, whether you’re writing code, a paragraph, or even a story.

2. How Do LLMs Work (Architecture Basics)

At the core of most modern LLMs is the Transformer architecture (Vaswani et al., 2017).
Unlike older models that processed text one word at a time, transformers look at whole sequences in parallel and figure out which words matter most to each other. Here are the essentials:

  • Embeddings – Words (or tokens) are turned into numerical vectors that capture meaning.
  • Positional Encoding – Adds information about word order (since transformers don’t read sequentially by default).
  • Self-Attention – Each word decides which other words in the sentence it should pay attention to.
  • Multi-Head Attention – Multiple attention mechanisms run in parallel, capturing different patterns (syntax, context, semantics).
  • Feed-Forward Layers + Residuals – Nonlinear layers stacked deep, with shortcut connections to keep training stable.
  • Output Layer – Predicts the most likely next token, repeating the process to generate full sentences.

That’s the backbone: a stack of transformer blocks working together, with more layers = more power.

Want to dig deeper? Microsoft has a great course on the same.

3. Types of LLMs

  • Decoder-only (GPT-style) → Text generation, chat, coding.
  • Encoder-only (BERT-style) → Text classification, embeddings, search.
  • Encoder-Decoder (T5/FLAN-style) → Translation, summarization, Q&A.
  • Instruction-tuned models → Optimized for the following natural language prompts (e.g., Mistral-Instruct, Falcon-Instruct,Gemini).

4. Accessing Open-Source LLMs on Hugging Face

Hugging Face hosts 100,000+ models. Some are fully open, others are gated.

  • To use gated models like Mistral or LLaMA:
  1. Visit the model’s page (e.g., Mistral-7B-Instruct)
  2. Click “Access repository” and accept the license.
  3. Generate a Read token here → HF Tokens
  4. Authenticate in notebook: 
   from huggingface_hub import login
   login("YOUR_HF_TOKEN")
Enter fullscreen mode Exit fullscreen mode

5. Running a Free LLM (Google AI Studio)

Instead of heavy Hugging Face models, you can start quickly with Google AI Studio → free API keys, fast responses.

👉 Try it here: Google AI Studio

Step 1: Get API Key

  1. Go to Google AI Studio Keys
  2. Generate a free API key.
  3. Copy it.

Step 2: Use in Notebook 

!pip install -q -U google-genai

from google import genai

# The client gets the API key from the environment variable `GEMINI_API_KEY`.
client = genai.Client(api_key="your_api_key")

response = client.models.generate_content(
    model="gemini-2.5-flash", contents="Explain All about LLMS"
)
print(response.text)
Enter fullscreen mode Exit fullscreen mode

👉 Example Notebooks:
1) Using Hugging Face free Models -> Colab Quickstart
1) Using Google AI Model -> Colab Quickstart

6. Free LLM Resources Table

  • Free & Fun LLM Access for Students
Platform Official Page Tutorial/Setup Guide Quick Notes
Hugging Face Hugging Face Models FreeCodeCamp: How To Start Use online demos, Spaces, no install needed. Colab works too!
ChatGPT (OpenAI, web) ChatGPT WhyTryAI Guide Just sign up and use it; no local resources required.
Google Gemini AI Studio Gemini Studio Gemini API Quickstart Run directly in browser or minimal code, free quota!
Meta AI (Llama 3, web demo) Meta.ai WhyTryAI Guide Llama 3 demo free in supported regions.
  • Free LLM Tools for Business Owners
Platform Official Page Setup & Docs Quick Notes
Google Gemini API Gemini API Main Gemini Quickstart
AI Studio Guide 		Generous free tier, ready for business use.
Vercel AI Gateway AI Gateway Getting Started
API Authentication 		One-stop API hub for many models.
Groq API Groq Console Groq Python SDK
Client Libraries 		Lightning-fast, monthly free tokens.
Hugging Face (commercial ok) Hugging Face Models FreeCodeCamp Setup
Commercial Model List 		Many models with permissive licenses.
  • Hands-On & Learning (For All)
Resource Main Page Description
Free LLM & Gen AI Courses Evidently AI LLM Courses Curated list for free learning.[13]

Just pick a platform, follow the quickstart, and you can chat or code with an LLM in minutes!

7. Limitations of Free LLMs

  • Rate limits → Free APIs (Google AI, Hugging Face) restrict daily usage.
  • Model size → Smaller free/open models may give weaker answers vs GPT-4/Gemini Pro.
  • Latency → Free cloud GPUs can be slow (Colab queues, Hugging Face load times).
  • Privacy → Using free APIs means your inputs may be logged. For sensitive use cases, local/offline models are safer.

Previous Chapter (Part A:What Is Generative AI?)

Now that you know what LLMs are, how they work, and how to get free access, the next step is learning how to talk to them effectively — that’s where Prompt Engineering comes in.

Next Chapter (Part C:Basics of Prompt Engineering)

Got questions or ideas?Drop a comment below — I’d love to hear your thoughts.

Let’s connect: 🔗 My LinkedIn

Top comments (0)