the (near) future is not 500B param models.
that is just what bloated companies with no imagination can and will continue to do.

another reason i am motivated to build a decentralized HF clone
twitter.com/ClementDelangu…

i dont believe a word coming out of him lol

whether we burn $50B a year, $500B or $5T a year, I genuinely don't care. we're building AGI, it's totally worth it

wake up new neural network just dropped (holy shit)

OOM

where's ilya, where's thebloke

Anyone knows what happened to WizardLM-2 models? Did Microsoft pull the plug because they uploaded a gpt4-level (allegedly) open weights model?

I managed to grab the 8x22B before it got pulled. Really weird. huggingface.co/alpindale/Wiza…

Mixtral 8x7b going 378 tokens per second on CPU.

comment with your ideas of what you want to see from a real Open Source AI hub/jurisdiction

if this doesn't terrify you, you just don't understand the implications

2 paths. fight against big tech with all their lobbying power, army of AI safetyist bureacrats and bought politicians. or just fukin leave this pathetic jurisdiction. soon enough borders wont matter anyway.

shape manipulators, unite

your monthly reminder that benchmarks suck and we need a new standard.