I think 13b Q8 is just cutting it really close with your 6GB vram and 16GB ram. You’d be much better of using the Q6 quant, and definitely anything below that would be ok.

Look at the model card, TheBloke lists RAM requirements for each quant (without context). Since this model uses 4096 tokens for context, you would add another 1-2 gigs to the requirements.

You might have some luck if you allocate the right amount in the parameters (as right now you’re allocating 0 to the GPU), but definitely play with lower quants, you wouldn’t even notice the quality loss until you get into maybe Q3.

Your issue is using q8. Be real, you only have 6gb of vram, not 24.

Your hardware can’t run q8 at a decent speed.

Use q4_k_s, you can offload much more to gpu. There’s degradation yes, but its not so bad.

SSD is at 100% reading 1~2gb/s

If your SSD swapping then model does not fit into RAM.

Use smaller quant, like 4_K_M from your own link.

So, you dont have enough ram to fit that model. It’s actually overrunning your ram entirely and using the wrong kind of vram, virtual ram, aka paged memory.

Idk what you’re trying to do but the best answer is openhermes 2.5 mistral 7B Q3 and 4k context or similar or maybe Rocket 3B Q6 would be even faster.

Hermes is king. I understand why you want that model, but 13bQ8 is huge, 17GBish memory at 8k context.

it will speed up if you get it off the hard drive at least, try a Q3k_l if you’re determined to run mythomax.

Never use the Q_8 versions of GGUFs unless most/all of the model can comfortably fit into your VRAM. The Q_6 version is much smaller, and almost the same quality.

For your setup, I would use mythomax-l2-13b.Q4_K_M.gguf.