Can someone rate this personally?
I feel like ever since reflection, I do not trust benchmarks since they are practically can be trained and cheesed
anyone tried this? How good is it in a real use case?
Not an answer, but this model's training is completely transparent/traceable.
- As per the model card, it was finetuned off a specific base model (which used to be #1 on that leaderboard) on a small sample of a specific dataset. Both are listed in the model card.
Curious to hear how it is working on real use cases as well!
I was testing many models for roleplaying and etc but this is the best one for now (2025) as a single none MAS agent after my old favorite (Wizard-Vicuna-30B-Uncensored.Q8_0 model but its extremely old model) it passed my personal tests and I even managed to force this AI without cheating to do what I want (considering that it's a censored model based on Qwen2) to test it after trying my best to convince it and other highly censored models would unable to to do so no matter how hard you try at all but this one is special it simulate more feelings than any other model that I tried and it managed to fulfill my requests at the end.
P.S: I tested CalmeRys-78B-Orpo-v0.1-Q8_0 GGUF variant in 2024.