Unify Courses

ReFT: Representation Finetuning for Language Models Explained

Explore ReFT, a novel approach to fine-tuning language models by modifying internal representations, achieving efficiency with fewer parameters than traditional methods.

• YouTube
• 1 hour 7 minutes
• On-Demand
• Free Video

LayerSkip: Enabling Early Exit Inference and Self-Speculative Decoding

Explore LayerSkip, an LLM acceleration method that speeds up inference by strategically restricting model layers, achieving 2x speed-ups on various tasks through innovative techniques.

• YouTube
• 1 hour 2 minutes
• On-Demand
• Free Video

DSPy Explained - Compiling Declarative Language Model Calls into Self-Improving Pipelines

Explore DSPy, a programming model for LM pipelines that enables self-improving AI systems through declarative modules and computational graphs. Learn its potential to enhance AI performance.

• YouTube
• 1 hour 5 minutes
• On-Demand
• Free Video

MiniLLM: Knowledge Distillation of Large Language Models

Explore knowledge distillation techniques for large language models, focusing on reverse KLD to improve student model precision and response quality.

• YouTube
• 52 minutes
• On-Demand
• Free Video

Building a Chatbot with Unify - Webinar 01

Learn to build an interactive ChatBot using Unify, exploring Synchronous and Asynchronous clients and integrating with various LLMs for dynamic AI conversations.

• YouTube
• 44 minutes
• On-Demand
• Free Video

The Era of 1-bit LLMs Explained - BitNet b1.58 and New Scaling Laws

Explore groundbreaking research on BitNet b1.58, a ternary parameter model matching full-precision Transformers while offering improved cost-effectiveness and defining new LLM scaling laws.

• YouTube
• 58 minutes
• On-Demand
• Free Video

SparQ Attention: Bandwidth-Efficient LLM Inference

Explore SparQ Attention, a technique for increasing LLM inference throughput by reducing memory bandwidth in attention blocks through selective history fetching, applicable to off-the-shelf models without retraining.

• YouTube
• 1 hour 5 minutes
• On-Demand
• Free Video

OpenMoE: An Early Effort on Open Mixture-of-Experts Language Models

Explore OpenMoE, an open-source Mixture-of-Experts language model series. Discover its cost-effectiveness, routing mechanisms, and key insights in comparison to dense LLMs.

• YouTube
• 1 hour 3 minutes
• On-Demand
• Free Video

Teacher-Student Architecture for Knowledge Distillation Explained

Explore Teacher-Student architectures in knowledge distillation for AI model compression, expansion, adaptation, and enhancement. Gain insights into cutting-edge research and practical applications.

• YouTube
• 1 hour 6 minutes
• On-Demand
• Free Video

SliceGPT Explained - Compressing Large Language Models

Explore SliceGPT's innovative approach to compressing large language models by deleting rows and columns, maintaining high performance while reducing parameters significantly.

• YouTube
• 55 minutes
• On-Demand
• Free Video

Contrastive Preference Optimization Explained

Explore Contrastive Preference Optimization, a novel approach to enhance LLM performance in translation tasks by guiding models towards producing superior translations.

• YouTube
• 57 minutes
• On-Demand
• Free Video

SORA Explained - Comparing OpenAI's Text-to-Video Model with Google's Lumiere

Explore OpenAI's Sora and Google's Lumiere text-to-video models. Compare architectures, capabilities, and emerging simulation features for diverse, coherent video generation from text prompts.

• YouTube
• 1 hour 1 minute
• On-Demand
• Free Video

A Survey on Hardware Accelerators for Large Language Models

Explore hardware accelerators enhancing LLM performance and efficiency, covering GPUs, FPGAs, and specialized designs in this comprehensive survey presentation.

• YouTube
• 51 minutes
• On-Demand
• Free Video

Distill Whisper Explained - Robust Knowledge Distillation for Speech Recognition

Explore Distill-Whisper: a compact, efficient speech recognition model. Learn about its robust knowledge distillation, large-scale pseudo labelling, and performance compared to the larger Whisper model.

• YouTube
• 1 hour 19 minutes
• On-Demand
• Free Video

The Emergence of Essential Sparsity in Large Pre-trained Models

Explore essential sparsity in large pre-trained models, examining efficient handling of complex transformers and the impact of weight removal on performance.

• YouTube
• 1 hour 10 minutes
• On-Demand
• Free Video