/biz/ - Business & Finance

>>55157610
the LLM censorship problem is a good target for crypto. Nobody wants to buy a 4090 with 24gig VRAM only to run an uncensored model that's weaker than gpt4. runpod.io and vast.ai are a step in the right direction, but it's estimated that chatGPT instance uses something like 400GB of RAM to function. How can local LLMs ever compete?
Well turns out it's possible to run massive distributed LLMs across networks of retail GPUs.
>https://github.com/bigscience-workshop/petals
Neural networks are layered and asynchronous which is suitable for cross network parallelization. An Avalanche subnet could act as a monetization platform to incentivize contribution and prevent excessive leeching.