Possibly the best Qwen2.5-32B tune?
I'm having a blast with this model. I've tested lots, if not all Qwen2.5-32B tunes and EVA-Gutenberg3 seems to be the best among them all. For narrative generation/roleplay, of course. It has nice prose and I haven't encountered much hallucinations or repetition. It's quite good at spatial awareness, too, which is usually a problem.
What are your thoughts on Qwen 3? I see that you have Qwen 3 tunes up to 14B, but these models are too small in my case. And base models, sadly, both the dense 32b one and Qwen3-30B-A3B moe were disappointing for me. But honestly, I'm just a user and barely understand how training works. Hopefully it could be a good base for tuning. Either way, there's always QwQ, too.
I'm not sure if you're working on anything atm, but it would be nice to see more of your tunes. I'll be sure to follow and check on you from time to time! Thank you for EVA-Gutenberg3 and know that it's appreciated.
Thank you for the feedback and I'm glad you're enjoying this model!
Qwen3 is a great model, but my hangup with the 32B is not having the base model available nor the compute in my lab to effectively finetune it. I might try a tune from an abliterated version at some point though. And the MoE models I would need to do some research before tuning.
Working on some new creative writing models though so stay tuned!