My go to model right now. Will you make an update?
Will you make an update to it maybe a 72b variant? Or are you waiting for Qwen 3 for a large variant?
Its really good but I can feel it missing some smarts that 70b models have, of course this model is much smarter and better than the other 24-32b models I've tested so far!
Hm, I realize i can fit 70B in my 32gb setup. (Even if it's a quite limited IQ3XXXS quant haha)
Currently i have 0 funds to train a 72B. The entire training process for this model was 1K~ USD.
https://ko-fi.com/deltavector
The page is in really early works, I'm currently trying to create a website to try and centralize my Card/Future Lora training/Model training
I have a ko-fi you can donate to if you'd like to see it, All proceeds go towards training models. Otherwise I'll be waiting for compute from some people to try and finetune a 72B or 70B version of this. Which'll be either LLama3.3 or Qwen3 72B (if they even make one, Looks like rn they are focusing on MoE)
Expect to see some small finetunes/Merges of 70Bs from me though!