Meta had a conference call last Wednesday, and while there were no major announcements, Mark Zuckerberg had interesting things to say about the upcoming Llama 4 models:

We’re training the Llama 4 models on a cluster that is bigger than 100k H100s or bigger than anything that I’ve seen reported for what others are doing. I expect that the smaller Llama 4 models will be ready first, and they’ll be ready, we expect sometime early next year, and I think that they’re going to be a big deal on several fronts — new modalities, capabilities, stronger reasoning, and much faster. It seems pretty clear to me that open source will be the most cost-effective, customizable, trustworthy, performant, and easiest-to-use option that is available to developers, and I’m proud that Llama is leading the way on this.

I’m particularly interested in the “easiest-to-use” claim. I still haven’t successfully deployed a Llama Stack server, so I access Llama using HuggingFace’s Transformers library. If there was a way to perform inference using PyTorch, I’ll use that in a heartbeat.

It still hurts to see that Nvidia’s CUDA has become the only language for GPU computing. I spent years writing a book and a blog about OpenCL, and I gave a handful of presentations on the topic. But now OpenCL is as dead as ActionScript, and that’s why Nvidia is in the Dow Jones Industrial Average instead of Intel.