Schmidt: Decentralized training of 400B parameter AI models is solvable by 2025
Jeff Schmidt · The Quest for Community-Trained Open Source AI Models · Oct 1, 2024 · at 54:41
Jeff Schmidt (Jeffrey Quesnelle) is co-founder of Nous Research. He is addressing whether a decentralized community using DisTrO could train a model as large as Meta's Llama 3 405B.
0:00 / 0:41exact quote · 41.4s
▶ Watch the full episode on YouTube → 720p mp4 · rendered on demand · StarZero watermark
“I think it still is, it would still be, you know, like a next year sort of environment thing that we would have to do. There are some scaling problems, or not scaling problems, but technical things about how you shard the model, because at that point you get to this problem where you have to put the model on more than one GPU or more, and like when you get to sort of that threshold, now there is a communication requirement between those ones to even shard it. We'll have, we're looking at what, how the activations work, like how you could shard those pieces to make it work. So there's still technical work to be done to scale it up infinitely to those like that, those super large model sizes. But it's not intractable by anything. It's an engineer. It's truly just like an engineering question that we need to tackle.”
quote is from the automated transcript, cleaned for reading: filler sounds and stutters are removed, nothing is rephrased. names can be misheard (the analysis reads context, assessments check outside sources). how →
More from Jeff Schmidt
Schmidt: AI industry relies on outdated 1990s architecture assumptions ripe for disruption
“It turns out that most assumptions in the AI space right now are a product of that's just how things had been done when there was not nearly as much energy and attention to it. So someone made an assumption maybe in the early nineties that everyone just kind o…”
Schmidt: Relying solely on fine-tuning is an existential threat to open-source AI
“Because that for us, if we're just fine tuning models. Right. That's like an existential threat, right? That's like an actual existential threat because the closed providers will continue to get better and we would be like dead in the water in a lot of sense.”
Schmidt: Sharing only key signals in distributed training yields equivalent model learning
“We know that like what we, what needs to be communicated between these things, the two, the different nodes are just these few key pieces of information. And that is necessary. That is a necessary condition or rather a sufficient condition To get the equivalen…”
Schmidt: Decentralized training will force NVIDIA to redesign chips around VRAM ratios
“What might happen sooner would be a redesign of the types of chips that NVIDIA or someone would make. Okay, under this model, we can dedicate more VRAM versus, there's like this question of how much VRAM versus how much processing power is on a die, and that, …”
Schmidt: Compute chips in NVIDIA's RTX 4090 and H100 are almost identical
“I think people don't actually realize that like a forty-ninety and like an H 100 are in a lot of ways the same card. For the non-gamers in the room, explain the forty-ninety. The chip that's inside of them is almost identical. The chip, the actual compute chip…”
Schmidt: Consumer gaming GPUs will become the sweet spot for distributed training
“Because you're able to distribute it so wide, I think the gaming GPU angle is really going to be like the sweet spot. If, as long as there's continued to be sort of like higher end gaming GPUs, and those are on comparison with the high end training GPUs, even …”