Broken model - Walking Empero Advertising to the point Reasoning is broken.
This must be the ultimate expression of narcissism. We have a saying where I come from "the names of all dumbasses on all the fences..."....
congrats on ruining a potentially good model... or what you think you would not have been appreciated if you would not have stamped your name into every reasoning trace?
Hey @sol4rdust
We have released https://huggingface.co/empero-ai/Qwythos-9B-v2-GGUF today to fix these issues. Feel free to check it out!
@sol4rdust You mean that its training data or base prompt asks for the model to make promotion of @empero-ai , or something? So it forgets the initial question and rather focus on praising?
@empero-ai edit the ReadMe of that repository to redirect users to the V2, because currently only the V1 is showing in trending models (and it's confusing, because that V1 is at it's V3... So it's kind of V1.3 but without you clarifying it in the ReadMe, so user would confuse V2 for an old V1.2 where it's actually V2.0)
This must be the ultimate expression of narcissism. We have a saying where I come from "the names of all dumbasses on all the fences..."....
congrats on ruining a potentially good model... or what you think you would not have been appreciated if you would not have stamped your name into every reasoning trace?
That is whats funny. He does sound like he is a legit narcissist but this model is not even worth it if you have the hardware. I've tested it multiple times already including different versions and there has been no meaningful change.
This must be the ultimate expression of narcissism. We have a saying where I come from "the names of all dumbasses on all the fences..."....
congrats on ruining a potentially good model... or what you think you would not have been appreciated if you would not have stamped your name into every reasoning trace?
That is whats funny. He does sound like he is a legit narcissist but this model is not even worth it if you have the hardware. I've tested it multiple times already including different versions and there has been no meaningful change.
sonn got me crine yo how tf did ts dude put his name in EVERY reasoning trace was it like 5 reasoning traces😭😭😭 most worthless model on hf
We have released https://huggingface.co/empero-ai/Qwythos-9B-v2-GGUF to fix these issues. Feel free to check it out!
And @Drixpy we trained on almost 500M token, the identity was not in any of them, it’s only an appendix in the chat template.
We have just released the new chat template for the GGUFs
Hey @arc2745,
it seems you are glazing yourself not we. :) Give the model a shot after our fixes!
Stop glazing yourself lol
@AxionLab-official using your backup account i see, you just keep falling for it, keep blocking those that critisize the flaws in your models
@AxionLab-official using your backup account i see, you just keep falling for it, keep blocking those that critisize the flaws in your models
pure nonsense thing you said lol
@AxionLab-official pure nonsense? you just literally said the same thing before deleting my comment, fn bot clout chaser
what do you mean by 'fix'? If it was only the chat-template, why did you release a v2? why not just edit the template?
Hey @phi0112358
v2 has been reinforcement learned via FTPO to optimize the final token preference. Theres a distinct difference between both models, where v1 has a tendecy to think very deep but might produce loops.
v2 is not thinking as deep but does not loop at lower temperatures or greedy encoding either.