Lost media
Thanks for doing this! Just in time, too. Thank you for working so quickly!
The author took this version down, which supposedly had better writing wherein version B (the normal v4 now) had better ability to follow formatting.
I actually wanted to see the difference between the two, and I prefer better writing even if the formatting takes a hit.
Since the original v4a is down you can't get the samplers to convert the gguf to _HF, yet v4b's ones might work since only a singular model changed, but I'm gonna have to test that.
You've archived some potentially lost llm media, thanks!
Perhaps you should rename the v4b quants to just be v4, but since you've done both now, I suppose it's better to just keep them like this.
I prefer better writing even if the formatting takes a hit.
Yeah, guess, I'd say... me probably too, unless it's too extreme. Anyway, while there are no tools to go from gguf to hf in general, at least the static quant repo has an f16 quant which should mostly preserve the original tensors and can be used to requant.
There is usually a 30 hour window or so during which I can pick models. But it's quite common for me to queue a model, only to have the repo disappear before it manages to download it a few hours later. Probably for the better in most cases (author tries it out, is abhorred by its brokenness, deletes it :)