william falcon

deep learning

machine learning

Deep dive into tools, tips and tricks to make AI models blazing fast, accurate and performant.

As the creator of PyTorch Lightning and AI researcher for the last 8 years, I've learned many tips and tricks to scale some of the world's largest models on thousands of GPUs. Here I deep dive into all the workflows, tools, tips and tricks to make AI models blazing fast, accurate and performant.

There is a lot of hype and misinformation in AI. These videos are unedited, factual, and show real-world AI development including all the dirty, insider tricks we use to make things work.

Let's train a PyTorch model on multiple B200 GPUs (multi-GPU training)

Let's train a PyTorch model on multiple B200 GPUs (multi-GPU training)
Let's deploy a custom AI model container as an autoscaling API on your private cloud in 10 minutes

Let's deploy a custom AI model container as an autoscaling API on your private cloud in 10 minutes
Let's finetune and deploy DeepSeek R1 (8B) for under $10

Let's finetune and deploy DeepSeek R1 (8B) for under $10
Let's code on cloud GPUs with VSCode and Jupyter notebooks

Let's code on cloud GPUs with VSCode and Jupyter notebooks
Let's pretrain a 3B LLM from scratch: on 16+ H100 GPUs, no detail skipped.

Let's pretrain a 3B LLM from scratch: on 16+ H100 GPUs, no detail skipped.
I explain Fully Sharded Data Parallel (FSDP) and pipeline parallelism in 3D with Vision Pro

I explain Fully Sharded Data Parallel (FSDP) and pipeline parallelism in 3D with Vision Pro
Round 2 - I use CodeLlama 70B vs Mixtral MoE to write code to finetune a model on 16 GPUs 🤯🤯

Round 2 - I use CodeLlama 70B vs Mixtral MoE to write code to finetune a model on 16 GPUs 🤯🤯
Round 1 - Codellama70B vs Mixtral MoE vs Mistral 7B for coding

Round 1 - Codellama70B vs Mixtral MoE vs Mistral 7B for coding