Opening Remarks
summary
Quick overview of the program and concrete takeaways: a recent release Llama 3 405B is explained end to end with specifics on the training data, optimization tweaks, and how inference and serving scale to billions of requests; framework work focuses on PyTorch and partnering with Databricks to accelerate open source tooling; hardware sessions show concrete hardware co-design examples that reduced training time or cost and a panel closes with measurable trends and deployment lessons from large-scale ops.
AI-written summary · about
related talks
33mLive Q&A Session 2
01h 00mPANEL - Data, Code, and Life with LLMs.
01mWelcome to the PyTorch Ecosystem for LLM Fine-tuning Mini Summit
30mUnderstanding the end-to-end LLM training and inference pipeline
16mKeynote: An Open-Source Stack for AI Compute: Kubernetes + Ray + PyTorch + vLLM
36mPanel Discussion
19mServerless inferencing: an infrastructure point of view
33mKeynote: Fine-tuning our way towards openness in AI