Censorship sometimes appears
Hi, this is a great model.
But sometimes censorship kicks in, I can't continue the dialogue, respect others etc. It's rare, but it happens.
I use the recommended parameters, but it's worth clarifying that I use reasoning, because with them the model behaves much better.
I'm not complaining, just reporting it. Thanks again for this model.
No you're absolutely right. I applied the same "Omega Treatment" to Qwen as the other Omega models and Qwen did not come out good and uncensored like the others. In hindsight I shouldn't have copied over the same model card for it without testing Qwen specifically. I won't be doing more Qwen models, it's not very good or worth the hassle.
I like this model because it follows the roleplay instructions much better than the mistral based ones and its reasoning is just great. Give it a chance.
Well I'm really glad someone is getting enjoyment out of it! I'm not sure what else I could do for Qwen besides doing other sizes of it and if I do train another model it will probably be GLM because I'm getting a lot of requests for that.
Well, just remember that small models like 14B with reasoning can be much better than big models without reasoning. It's easy to run it locally on higher bits without Tesla or 4090. Either way, the choice is yours, thanks for doing it.
Well, just remember that small models like 14B with reasoning can be much better than big models without reasoning. It's easy to run it locally on higher bits without Tesla or 4090. Either way, the choice is yours, thanks for doing it.
I agree to your point, I too feel ReadyArt should create 8-32B parameter reasoning models in future as well.
Well I'm really glad someone is getting enjoyment out of it! I'm not sure what else I could do for Qwen besides doing other sizes of it and if I do train another model it will probably be GLM because I'm getting a lot of requests for that.
GLM is good!
i would like to enable reasoning. how can i do it?
Just delete these lines
/no_think
/nothink
No you're absolutely right. I applied the same "Omega Treatment" to Qwen as the other Omega models and Qwen did not come out good and uncensored like the others. In hindsight I shouldn't have copied over the same model card for it without testing Qwen specifically. I won't be doing more Qwen models, it's not very good or worth the hassle.
I'm not sure what it is, but Forgotten Safeword 12B v3.0 is the best model I've ever seen from you. It's simply magical.