torchft + TorchTitan: 1200+ failures, no checkpoints, model convergence.
A Llama 3 model was trained across 300 L40S GPUs with synthetic failures every 15s. No restarts. No rollbacks. Just asynchronous recovery and continued progress.
📘 hubs.la/Q03t1Z0b0#PyTorch
Here's my attempt at training a 3D dynamic object detection model from scratch. I'm pretty happy with the result though it's far from perfect fn.lc/post/3d-detr/
This is what I've been up to for the last couple of months. Happy to answer any questions folks have 🙂
I'll be giving a talk about the new torch::deploy beta at #PyTorchConference on Dec. 2nd!
If anyone is around, I'd love to meet up and chat about PyTorch, ML and self-driving
Registration and live stream is at pytorchconference22.splashthat.com
Hope this makes it easier to run other conferencing apps -- I got Discord to work with the car cameras a year ago but haven't used it much since it's a pain to setup
NEWS: Zoom officially announces its video conferencing app is coming soon to Tesla vehicles.
The Zoom app will use the vehicle's cabin camera located above the rear-view mirror. driveteslacanada.ca/news/zoom-offi…