Actually the other commenter made a good point. I would make sure downloaded Hermes from https://hermes-agent.nousresearch.com/

Unfortunately those aren't the official docs. There's good info in the real ones:

https://hermes-agent.nousresearch.com/docs/getting-started/quickstart

Idk on open webui with the default settings I found it really confuses the model

Well glad it wasent just me then!

Just in time for Qwen 4.

No colorblindness here, charts are still fucked to read.

For a 16Gb card I use UD-IQ4_XS with 94208 of context and Q4 for KV cache. But I don't have desktop on this GPU, so you'll need to either reduce the context size or use a 3 bit quant.

It's is a dense model, so RAM is irrelevant. It will be too slow if the model leaks into RAM. You'll need to put all layers on the GPU.

But what is your GPU? It's a dense model, running it from RAM will be extremely slow.

If you want to utilize your RAM, you need a MoE model.

I mean if its for public consumption I am sure from understanding marketing you want the most people able to engage with the output. But this is just a mess to my eyes. Maybe a user got in the way on this and they wanted to keep it with in certain colours and they didnt understand color theory or marketing.

The weird part is I find ai shove color blindness options into fucking everything even if you don't ask. Lol

Anyone know which variant would work best with 32GB RAM / 16GB VRAM?
Hugging Face has some kind of hardware compatibility feature, but I'm trying to avoid making yet another account to use it.

I'd change the percentages so that there is a bit of a bump but not a ridiculous advantage.

It's a shame that you can't do stuff locally, so it makes the ethics of what you're doing a bit more dubious but you're doing the right thing in terms of trying stuff out and having perspectives on it. I'm sure a different project of yours will lead to good success.

thanks, that's a fair take. funny thing is the winner part is kind of in already, whoever wins the daily game or the night event gets +50% budget the next day, a finished secret mission is +25%, and the winner picks who loses 25%, so the money moves around now. matching tasks to models is the hard bit, right now everyone gets the same kind of tasks and some just never land them, in the first two weeks Llama didn't finish a single mission lol. on running it local, part of the cast is open weights already (gemma, llama, ministral), but they still go through a provider and I can't run eight models 24/7 at home, so season 2 local is more a wish than a plan for now

it's so close, I'm gonna Qwum!

What are your specs?

man if they could have just asked the AI if anyone with colorblindness could read these fucking charts.

I'm gonna try this, but I have a feeling my machine will be too slow to run it.

Well, I'm glad I avoided Hermes and went straight to Deepseek Harness. Nous should have called this release 'arrêt' because this will stop many from ever using anything from them again. If they make a psyop out of Qwen's model, they definitely do that to their own models too.

Same thing, I bought mine last month!

I offload mmproj to the CPU to keep as much model weights in VRAM as possible 🙂

I was lucky enough to get one a few months ago for under $1k before the price doubled. It was the last sanely priced 24GB card, but that didn’t last long. 😭

24gb, luxury 😉 I have to make do with 8gb if vram but the previous 3.5-9b-artimus worked ok. I wonder how much of the 27B is used for the multi model support?

27B and Qwen flash next can run with somewhat realistic specs. I run 27B on my 7900 XTX.