Thoughts on GRM-2.6-Plus-GGUF ?

Posted by soyalemujica@reddit | LocalLLaMA | View on Reddit | 12 comments

Judging by what they state, it should be better than Qwen 3.6 27B

[-]

SnooPaintings8639@reddit

Just a random fine-tune of qwen? After reading the 'super-hyped' description: I have zero expectation nor a desire to test it out.

[-]

GRM2.6 plus is probably the only finetune right now that is NOT directly just worse than base. BUT they arent a straight upgrade either. Zeus benchmarked quite a lot of finetunes for 3.6 27b and this one seems like the only one that is at least a sidegrade und not just a downgrade.

[-]

bonobomaster@reddit

Do you have a link for me?

Google doesn't really help.

[-]

Finanzamt_Endgegner@reddit

For the benchmarks? No zeus told us in the localllm discord (;

[-]

Dany0@reddit

I tested it on my bench and base qwen won

I also tested (closest equivalent quant of) AEON RYS, XpressAI RYS, qwopus preview

Only zero shot counted but I did 30 runs thereof. Each model had tests it was strong on and weak on, like base model never not even once completed one particular test the RYS models completed quite often. But overall pass rate was higher on base qwen. It's more consistent it seems like

[-]