I've been scrolling through Twitter these past couple of days and saw Binance’s official account tweet that they’re launching a new product on June 1st. Last night, I noticed @CZ also retweeted it. With BNB hitting $745 today, I think I might have figured it out; there’s a high chance they’re going to list on the US stock market. Let’s keep an eye out for whether they’ll roll out a wallet for Pre-IPO investments.
When I was checking out info on @OpenLedger , I came across the term OpenLoRA several times, but I just brushed it off, thinking it was just a deployment tool and didn’t pay much attention. It wasn’t until I did the math recently that I realized I underestimated its significance.
Those of us who have worked on AI projects know that fine-tuning a custom model isn’t hard; the tough part is the inference cost after deployment. You need to keep a dedicated GPU running 24/7, and the monthly costs can blow a small team’s budget out of the water. This isn’t an exaggeration; I’ve been there. Most small teams, after fine-tuning their models, can’t even afford to use what they’ve trained, and they either revert to using APIs or the whole project falls apart. #OpenLedger
What OpenLoRA is doing is breaking down that cost structure. It uses a multi-tenant architecture to allow a single GPU to run thousands of fine-tuned models simultaneously, with each user’s model sharing the underlying computing power and being billed based on actual usage. To put it simply—before, you had to rent an entire truck to move goods, but now it’s like carpooling; you only pay for the space your few boxes occupy.
I think this is seriously undervalued. It’s not just about cutting costs; it’s about returning the usage rights of proprietary AI models from big companies back to ordinary developers.
@OpenLedger #openledger $OPEN $BNB
When I was checking out info on @OpenLedger , I came across the term OpenLoRA several times, but I just brushed it off, thinking it was just a deployment tool and didn’t pay much attention. It wasn’t until I did the math recently that I realized I underestimated its significance.
Those of us who have worked on AI projects know that fine-tuning a custom model isn’t hard; the tough part is the inference cost after deployment. You need to keep a dedicated GPU running 24/7, and the monthly costs can blow a small team’s budget out of the water. This isn’t an exaggeration; I’ve been there. Most small teams, after fine-tuning their models, can’t even afford to use what they’ve trained, and they either revert to using APIs or the whole project falls apart. #OpenLedger
What OpenLoRA is doing is breaking down that cost structure. It uses a multi-tenant architecture to allow a single GPU to run thousands of fine-tuned models simultaneously, with each user’s model sharing the underlying computing power and being billed based on actual usage. To put it simply—before, you had to rent an entire truck to move goods, but now it’s like carpooling; you only pay for the space your few boxes occupy.
I think this is seriously undervalued. It’s not just about cutting costs; it’s about returning the usage rights of proprietary AI models from big companies back to ordinary developers.
@OpenLedger #openledger $OPEN $BNB