Unleashing LLMs: Training, Fine-Tuning and Evaluating
10 August 2024 | 09:30AM - 05:30PM | RENAISSANCE :- Race Course Rd, Madhava Nagar Extension
About the workshop
This workshop is designed to provide a comprehensive overview of LLMs, from foundational concepts to advanced applications. Whether you're a beginner or have intermediate experience, you will gain valuable insights and hands-on experience with some of the most cutting-edge technologies in the field.
- Key Takeaways:
- Understand the fundamentals of Language Models and Transformer architectures.
- Gain hands-on experience with LLMs and related concepts such as PEFT, Prompt Engineering, RAGs, and more.
- Explore advanced topics such as Reinforcement Learning from Human Feedback (RLHF) and Retrieval-Augmented Generation (RAG).
Instructor
Modules
- Overview of Generative AI and the basics of language modeling.
- NLP Basics for Embedding and Attention: Fundamental concepts in NLP, focusing on embeddings and attention mechanisms.
- Language Modeling: Basics of language modeling and its importance.
- Hands-On: Implementing a simple language model using basic NLP techniques.
- Transformer Architectures: Detailed look into the Transformer architecture that powers modern LLMs.
- GPT Series of Models: Overview of the evolution of GPT models.
- Hands-On: Training a mini Transformer model and experimenting with GPT-2 for text generation.
- Training Process and Scaling Laws: Understand how LLMs are trained and the laws governing their scaling.
- PEFT: Learn Parameter-Efficient Fine-Tuning methods.
- LoRA: Introduction to Low-Rank Adaptation.
- QLoRA: Exploring Quantized Low-Rank Adaptation.
- Instruction Tuning: Techniques for fine-tuning models using instructions.
- RLHF: Reinforcement Learning from Human Feedback and its applications.
- Evaluation Metrics and Benchmarks: Methods to evaluate and benchmark LLM performance.
- Beyond Prompting: Understanding Frameworks such as DSPY
- Hands-On:
- Fine-tuning a pre-trained model using different methods and evaluating it with standard benchmarks.
- Hands-on with DSPY
- OpenSource vs Commercial LLMs: Comparison between open-source and commercial LLM solutions.
- Prompt Engineering: Crafting effective prompts to get desired outputs.
- RAGs: Techniques for retrieval-augmented generation.
- Vector Databases: Using vector databases for efficient data retrieval.
- Chunking and Ingesting Documents: Methods for processing and ingesting documents.
- Securing LLMs
- Prompt Hacking and Backdoors
- Defensive Measures
- Hands-On:
- Implementing basic prompt engineering techniques and
- Building a simple RAG system.
- Multimodal: Integration of different data modalities in LLMs.
- Mixture of Experts: Using a mixture of expert models for improved performance.
- SLM: Introduction to Small LMs.
- Ethics and Bias in LLMs: Understanding and mitigating biases in LLMs.
- Next Steps: Speculative topics on future advancements.
- GPT5?: What to expect from the next generation of GPT.
- Beyond: Future possibilities and directions for LLM research.
- Hands-On: (If time permits) Experimenting with multi-modal models and mixture of experts.
- Basic understanding of Python programming
- Familiarity with fundamental machine learning concepts
- Experience with common NLP tasks and techniques - e.g summarization, QA, classification
- Comfortable running Jupyter Notebooks using Anaconda/VS Code or Google Colab.
- We will provide ample GPU credits to ensure a seamless and productive workshop experience.
- (Optional) Basic knowledge of deep learning frameworks (e.g., PyTorch, TensorFlow)
Certificate of Participation
Receive a digital (blockchain-enabled) and physical certificate to showcase your accomplishment to the world
- Earn your certificate
- Share your achievement
