This podcast episode highlights the pivotal advancements and challenges in AI infrastructure, showcasing the journey from early concepts to modern large language models like Llama 3, with an emphasis on open innovation, collaboration, and the evolving landscape of AI technology that seeks to democratize access and boost efficiency for developers and researchers alike.
AI Infra at Scale 2024: Opening Remarks and Overview
AI's Journey: From Turing to Llama 3.1
Scaling AI Models: Data, Compute, and Algorithms
The Future of AI: Scale, Openness, and the Quest for Generalized Intelligence
Llama 3: Open Intelligence for Everyone
Building the Llama System: Agentic Applications and Open Trust & Safety
Llama 3 Data Infrastructure: Scaling for Quality and Diversity
Llama 3 Data Lifecycle: Curation, Ablation, and Feedback Loops
Llama 3 Training Infrastructure: Scaling to 16,000 GPUs
Llama 3 Training: Scheduling, Parallelism, and Observability
The Future of AI Training: Heterogeneous Hardware, Asynchronous Training, and Beyond
Llama 3 Inference Infrastructure: Serving at Scale
Llama 3 Inference: Production Challenges and Optimizations
Llama 3 Inference: Scaling to a Global Audience
The Future of AI Infrastructure: Multimodality, Heterogeneity, and User Experience
Unlocking AI's Potential Through Open Innovation
Open Innovation in AI: Systems, Models, and Efficiency
The Cost of AI: Deploying Large Models Efficiently
PyTorch and Large Language Models: Enabling Open Innovation
PyTorch for Large-Scale Training: Distributed Algorithms and Techniques
PyTorch for Fine-Tuning Large Language Models: Memory Efficiency and TorchTune
Enabling On-Device Inference of LLMs: Quantization, Sparsity, and Torch.io
MTIA: Meta's In-House Training and Inference Accelerator
MTIA Silicon Architecture: Optimizations for Meta AI Models
MTIA Software Stack: PyTorch Integration, Triton Support, and Model Enablement
MTIA Results and Future Directions: Co-design and Performance Gains
AI Infra at Scale: Panel Discussion
Democratizing AI: The Challenges of Deployment and the Need for a Reset
The Future of AI: Production-Ready Solutions and the Importance of Openness
Open Innovation and the Future of AI: A Force Multiplier for Progress
Sign in to continue reading, translating and more.
Open full episode in Podwise 
