备份 - decalyu/HPC-AI-Resources GitHub Wiki

🚀 ✨ PC: Introduction to Large Language Models (LLMs) ✨ 🚀

🎯 Goal

🤖 Understand what a Large Language Model (LLM) is and how it can be used for text generation, question answering, and more, using Python. No prior experience needed—just bring your curiosity! 🚀

📌 What You Will Learn 🧠 💡

✅ What is a Large Language Model (LLM)?
✅ How do LLMs work?
✅ How to use pre-trained LLMs for text generation
✅ How to use LLMs for answering questions
✅ Hands-on coding with Google Colab
✅ Basics of Hugging Face library for LLMs

🤖 1. What is a Large Language Model (LLM)?

🧠 Understanding LLMs in Simple Terms

A Large Language Model (LLM) is a type of AI model that can understand and generate human-like text. It learns by analyzing huge amounts of text data, recognizing patterns, and predicting words based on context.

📌 Real-World Examples:

  • ✅ Chatbots like Siri, Google Assistant, ChatGPT 🗣

  • ✅ AI-powered writing assistants (Grammarly, Jasper AI) ✍

  • ✅ Search engines predicting your queries 🔍

  • ✅ AI-generated stories and essays 📖

🔥 2. How do LLMs work?

🛠 How Do LLMs Generate Text?

  • 📚 Training on Large Datasets

    • LLMs are trained on billions of words from books, articles, and websites.

  • 🔍 Predicting the Next Word

    • When given a sentence, the model predicts the most likely next word.

  • ✍ Generating Coherent Responses

    • By repeating this process, it forms complete sentences that make sense.

📌 Example: How AI predicts text

  • Input: "The sky is"

  • Predicted Output: "blue because of the way light scatters."

💡 *Quick Thought: Where else do you see AI predicting text? Jot down your ideas! 📝 *

🔧 3. How to use pre-trained LLMs for text generation

💻 Step 1: Open Google Colab

1️⃣ Open Google Colab
2️⃣ Click + New Notebook

📚 Step 2: Import Required Libraries

➕ 🐍 Add a New Code Cell

1️⃣ Click + Code in the top left to add a new code cell.
2️⃣ Copy and paste the following code into the new code cell.

🔗 ChatGPT prompt for generating the code


# Import the pipeline function from the transformers library
from transformers import pipeline

# Load a pre-trained text generation pipeline
# The 'text-generation' pipeline uses a model like GPT-2 by default
text_generator = pipeline("text-generation")

# Define a prompt for text generation
prompt = "Once upon a time, in a distant kingdom,"

# Generate text based on the prompt
# The max_length parameter controls the output length
output = text_generator(prompt, max_length=50)

# Print the generated text
print(output)

🔗 ChatGPT explanation for the code

3️⃣ Click Run (▶) and check the output!

✅ Libraries imported successfully! You’re now ready to use the Hugging Face pipeline. 🚀 🎉

🧠 Step 3: Load a Pre-Trained LLM

💡 We will use GPT-2, a popular model that generates human-like text.

➕ 🐍 Add a New Code Cell

1️⃣ Click + Code in the top left to add a new code cell.
2️⃣ Copy and paste the following code into the new code cell.

🔗 ChatGPT prompt for generating this code


# First, we need to install the transformers and torch libraries if they are not already installed.
# Uncomment and run the following line if you haven't installed them:
# !pip install transformers torch

# Import necessary libraries
from transformers import GPT2LMHeadModel, GPT2Tokenizer

# Step 1: Load the pre-trained GPT-2 model and tokenizer
# GPT2Tokenizer is used for encoding input text and GPT2LMHeadModel is used for text generation.

# Load pre-trained GPT-2 tokenizer
tokenizer = GPT2Tokenizer.from_pretrained("gpt2")

# Load pre-trained GPT-2 model
model = GPT2LMHeadModel.from_pretrained("gpt2")

# Step 2: Encode the input text
# Tokenize the input text using the tokenizer. 
# The model expects input in tokenized form, so the text is converted into token IDs.
input_text = "Once upon a time, there was a brave knight"
input_ids = tokenizer.encode(input_text, return_tensors='pt')

# Step 3: Generate text using the model
# The model generates a continuation of the input text based on the provided input.
# Here, 'max_length' specifies the maximum length of the generated text.
output = model.generate(input_ids, max_length=100, num_return_sequences=1, no_repeat_ngram_size=2, top_p=0.95, temperature=0.7)

# Step 4: Decode the generated output
# After the model generates the token IDs, we decode them back to text.
generated_text = tokenizer.decode(output[0], skip_special_tokens=True)

# Step 5: Display the generated text
print("Generated text:\n", generated_text)

🔗 ChatGPT explanation for the code

3️⃣ Click Run (▶) and check the output!

✅ Pre-trained LLM loaded successfully! You're now ready to generate text using GPT-2. 🧠 🚀 🎉

📝 Step 4: Generate Text Using the LLM

➕ 🐍 Add a New Code Cell

1️⃣ Click + Code in the top left to add a new code cell.
2️⃣ Copy and paste the following code into the new code cell.

🔗 ChatGPT prompt for generating this code


# Install the necessary libraries (uncomment if not installed)
# !pip install transformers torch

# Import required libraries from Hugging Face's transformers
from transformers import GPT2LMHeadModel, GPT2Tokenizer

# Define the input text prompt
input_prompt = "Once upon a time in a land far away"

# Set the maximum length for the generated text
max_length = 100  # You can adjust this to control the length of the output text

# Load the pre-trained GPT-2 model and tokenizer
# GPT-2 is a popular language model trained by OpenAI
tokenizer = GPT2Tokenizer.from_pretrained("gpt2")
model = GPT2LMHeadModel.from_pretrained("gpt2")

# Encode the input text prompt into tokens that the model understands
# This converts the input string into a format that can be fed into the model
input_ids = tokenizer.encode(input_prompt, return_tensors='pt')

# Generate text using the model based on the input prompt
# 'max_length' specifies the maximum length of the output sequence (including the input prompt)
generated_output = model.generate(input_ids, max_length=max_length, num_return_sequences=1, no_repeat_ngram_size=2, temperature=0.7)

# Decode the generated tokens back into human-readable text
# This step converts the token IDs back into a string
generated_text = tokenizer.decode(generated_output[0], skip_special_tokens=True)

# Print the generated text
print("Generated Text:\n", generated_text)

🔗 ChatGPT explanation for the code

3️⃣ Click Run (▶) and check the output!

✅ Text generated successfully! You should now see a creative continuation of your prompt. 📝 ✨ 🎉

🎯 Challenge: Try changing the prompt to something funny or mysterious!🔥

💡 Extra Tip: Edit the prompt_text = "..." line and try your own creative ideas. For example, start with “In a haunted bakery...” or “The cat who ruled the internet...”

🤖 4. How to use LLMs for answering questions

💡 LLMs can also answer questions based on a given context!

➕ 🐍 Add a New Code Cell

1️⃣ Click + Code in the top left to add a new code cell.
2️⃣ Copy and paste the following code into the new code cell.

🔗 ChatGPT prompt for generating this code


# Step 1: Install necessary libraries (if not installed yet)
# You can run these commands in your terminal or Jupyter notebook cell
# !pip install transformers torch

# Step 2: Import necessary libraries
from transformers import pipeline

# Step 3: Load the question-answering pipeline using the 'distilbert-base-uncased-distilled-squad' model
qa_pipeline = pipeline("question-answering", model="distilbert-base-uncased-distilled-squad")

# Step 4: Define the context (the passage that contains information to answer the question)
context = """
Hugging Face is a company that focuses on Natural Language Processing (NLP). They have developed a popular library called transformers, 
which provides pre-trained models for various NLP tasks, including text generation, sentiment analysis, and question answering. 
Hugging Face aims to democratize AI by making these models accessible to the wider community.
"""

# Step 5: Define the question you want to ask
question = "What does Hugging Face do?"

# Step 6: Use the QA pipeline to answer the question based on the context
answer = qa_pipeline(question=question, context=context)

# Step 7: Print the answer
print(f"Question: {question}")
print(f"Answer: {answer['answer']}")

🔗 ChatGPT explanation for the code

3️⃣ Click Run (▶) and check the output!

✅ Question answered successfully! You should now see the model's response based on the given context. 🤖 🎉

🎯 5. Wrap-Up & Next Steps

🎉 Congratulations! Today you learned how to:

  • ✅ Use LLMs for text generation ✍
  • ✅ Use LLMs for question answering 💬
  • ✅ Work with Hugging Face models 🤖

🚀 Next Workshop: 📚 LLM + RAG (AI-Powered Search)

🔗 Additional AI Resources 📚

🎉 Keep learning AI, and see you at the next workshop! 🚀

📝 Workshop Feedback Survey

Thanks for completing this workshop!🎆

We'd love to hear what you think so we can make future workshops even better. 💡

📌 Survey link