lora

It's probably more appropriate to complain on a LoRa post that it's too close to LoRA.

I see this comment on every single LoRA post despite the vast majority of posts being about LoRA not LoRa.

I never heard about LoRA, but I have used LoRa and was curious to see what it had to do with reasoning.

I suspect LoRA has already eclipsed LoRa in general usage.

Such low dimensionality of the LoRA vector must surely result in a close-to-linear modification to the KV calculation.

No open standard spans BLE, WiFi Direct, and LoRa simultaneously with per-device cryptographic identity for general-purpose communication.

While not excatly that, lora, a long range(kilometers) low power/cost protocol, seems like the right choice.

How is this protocol better than LoRa(by semtech - an old stable company, used by giants like IBM),or weightless(standardized protocol)?

If two temporally adjacent observations should produce similar LoRA updates, orthogonality would actively push them apart.

In particular, this technique will likely avoid the intruder dimension problem which plagues naive LoRA.

But the code for doing some Lora for gemma4 is probably not yet available.

Projects span LLM agent evaluation, low-resource NMT with LoRA, and abstractive headline generation.

This is interesting and all, but “LoRA” is painfully close to “LoRa” (which is related to radio networking, not AI) when just scanning a list of topics.

“I’ve been Miss Brazil, Miss Poland, and Miss America, and I’m still pissed I was never Miss Canada,” Lora, a model from Toronto whom I met in Beijing, told me recently.

(That idea is called Saunt Lora's Proposition -- all possible ideas have already been found -- in the book.

To answer my own question: I like Noto Serif, Lora, PT Serif, and Habibi very much.