HandsOnLLM/Hands-On-Large-Language-Models
Official code repo for the O'Reilly Book - "Hands-On Large Language Models"
Language:Jupyter Notebook
Total stars: 194
Stars trend:
16 Sep 2024
5pm ▊ +6
6pm ▊ +6
7pm ▉ +7
8pm ▎ +2
9pm ▍ +3
10pm ▌ +4
11pm ▍ +3
17 Sep 2024
12am ▏ +1
1am ▍ +3
2am ▋ +5
3am ██▎ +18
4am ██▏ +17
#jupyternotebook
#artificialintelligence, #book, #largelanguagemodels, #llm, #llms, #oreilly, #oreillybooks

Top 7 Open-Source LLMs in 2025

1️⃣ DeepSeek R1
An open-source reasoning model excelling in logic, math, and decision-making.

2️⃣ Qwen2.5-72B
Alibaba’s 72B LLM optimized for coding, multilingual tasks, and structured data.

3️⃣ Llama 3.3
Meta’s multilingual LLM with strong dialogue, reasoning, and 128K context.

4️⃣ Mistral-Large 
A 123B model excelling in reasoning, coding, and high factual accuracy.

5️⃣ Llama-3.1-70B
A robust instruction-tuned model for research, reasoning, and enterprise use.

6️⃣ Phi-4
Microsoft’s efficient small-scale model for programming and logical reasoning.

7️⃣ Gemma-2-9b-it
Google’s lightweight LLM for reasoning, summarization, and Q&A.


#llm

𝐇𝐨𝐰 𝐃𝐨 𝐋𝐚𝐫𝐠𝐞 𝐋𝐚𝐧𝐠𝐮𝐚𝐠𝐞 𝐌𝐨𝐝𝐞𝐥𝐬 (𝐋𝐋𝐌𝐬) 𝐖𝐨𝐫𝐤?

When I first worked with LLMs, they felt like magic. But once I learned how they really process language, it all started to make sense. Here’s how it works -

1. Tokenization
- Why it matters: Before the model understands language, it needs to slice it into chunks—words, subwords, even characters.
• Use case: In a chatbot for a retail client, tokenization helped capture slang and misspellings from user queries—so “gr8 deals” didn’t get lost in translation.

2. Embedding
- Why it's key: Those tokens turn into vectors—numbers that carry meaning and context.
• Use case: While building a resume parser, embeddings helped the model understand “developer” and “programmer” as similar—even though the words were different.

3. Attention (Self-Attention)
- Why this stands out: This is where the model learns what to pay attention to. It looks across the entire sentence to make sense of context.
• Use case: In a legal document assistant, attention mechanisms helped the model figure out that “he” referred to “the client” several sentences back.

4. Feed-Forward Layers
- Why it's helpful: It adds depth. These layers refine meaning and relationships even more.
• Use case: While generating product descriptions, this helped the model balance between specs and tone—so it sounded natural, not robotic.

5. Normalization + Dropout
- Why it's needed: Keeps learning stable and prevents the model from overfitting to noise.
• Use case: During fine-tuning for customer service tone, this made sure the model didn’t memorize one style too closely—and stayed flexible.

6. Prediction (Next-Token Generation)
- Why it's powerful: Based on what it saw so far, the model predicts the next word.
• Use case: In an AI assistant for internal reports, prediction steps helped craft bullet points from long texts, cutting writing time by 70%.

. .

But what’s the most sensitive step?
- Attention. If it focuses wrong, hallucinations happen—confusing facts or inventing things.

My learning?
- You don’t need to master it all at once. Stay curious. Build, break, repeat.

#llm