I wanted to thanks you and ask advice
First, i wanted to thank you, because i have been using your model with kobold and ST for some time now, and i love it .
But then come the harder part, there some may different preset in your setting page , i wanted some advice on which would be the best, i used before the mistral small correct for context model, instruct template, and system prompt,but recently i change to the nemotron-custom for those same category, because it was the most recent updated one, but i don't know, how it work in the depth, so i'm not sure if i should use one more than and other, because there so many preset in your folder between the basic folder, the chat completion folder, and customized folder, not sure which preset i should use for my need or which one would be the best to use overall.
Same for the parameters, even if they are less, only 3 between the Elclassico, Madlad, and the Top(A)Kek.json, i used the last, because it was the most recent, but with all the settings, there's, and not really knowing what most of them do, i just use it, but don't know if it would be the best to use.
I read the part on the main page of your model, where you talk about the format you were using was wrong with the "updated: " line even if i don't really know how it affect, the behaviour of the model. (because all this technical part of ST with all the parameter, and context model, instruct template, and system prompt, i think i know that what define how the model with work in ST but it's really complex and didn't dive into it) or since when the update was from , that why i tried to use one of the most recent one the nemocron-custom, but it seem to work , but then i started to get this relatively frequently "<|start_header_id|>user<|end_header_id|>" poping up at the end of some of the generation response, i edit them, when it's happens, but i don't know how to do for it to not happen at all.
To try to resume all this shortly:
- Thanks really for all the works you put into making the model, and parameters
- Second could you advice me, the best preset to use for the context model, instruct template, and system prompt
- Third which one of the 3 parameters (Elclassico, Madlad, and the Top(A)Kek.json) are best to use, if we want a good experience with creative rp.
and if you could tell me briefly what do the most useful parameter there do, i know for the temp, which if i'm correct alter the creativity of the model, but that pretty much all. - Depending on the answer to the second point, if you recommand me the nemocron, could you tell me why i got the "<|start_header_id|>user<|end_header_id|>" poping up at the end of the generated response of the model sometimes, not always but still frequent.
I hope you would read this and could help me, anyways thanks for the model and all the work.
First, i wanted to thank you, because i have been using your model with kobold and ST for some time now, and i love it .
But then come the harder part, there some may different preset in your setting page , i wanted some advice on which would be the best, i used before the mistral small correct for context model, instruct template, and system prompt,but recently i change to the nemotron-custom for those same category, because it was the most recent updated one, but i don't know, how it work in the depth, so i'm not sure if i should use one more than and other, because there so many preset in your folder between the basic folder, the chat completion folder, and customized folder, not sure which preset i should use for my need or which one would be the best to use overall.
Same for the parameters, even if they are less, only 3 between the Elclassico, Madlad, and the Top(A)Kek.json, i used the last, because it was the most recent, but with all the settings, there's, and not really knowing what most of them do, i just use it, but don't know if it would be the best to use.
I read the part on the main page of your model, where you talk about the format you were using was wrong with the "updated: " line even if i don't really know how it affect, the behaviour of the model. (because all this technical part of ST with all the parameter, and context model, instruct template, and system prompt, i think i know that what define how the model with work in ST but it's really complex and didn't dive into it) or since when the update was from , that why i tried to use one of the most recent one the nemocron-custom, but it seem to work , but then i started to get this relatively frequently "<|start_header_id|>user<|end_header_id|>" poping up at the end of some of the generation response, i edit them, when it's happens, but i don't know how to do for it to not happen at all.To try to resume all this shortly:
- Thanks really for all the works you put into making the model, and parameters
- Second could you advice me, the best preset to use for the context model, instruct template, and system prompt
- Third which one of the 3 parameters (Elclassico, Madlad, and the Top(A)Kek.json) are best to use, if we want a good experience with creative rp.
and if you could tell me briefly what do the most useful parameter there do, i know for the temp, which if i'm correct alter the creativity of the model, but that pretty much all.- Depending on the answer to the second point, if you recommand me the nemocron, could you tell me why i got the "<|start_header_id|>user<|end_header_id|>" poping up at the end of the generated response of the model sometimes, not always but still frequent.
I hope you would read this and could help me, anyways thanks for the model and all the work.
Apologies for the late response, was away for Christmas.
Thanks for the thanks, glad you've been enjoying my work!
My presets are created with specific formats and models in mind. If a model uses a ChatML instruct following format, you want to grab ChatML settings off my page. You haven't specified which model you're currently using so it's difficult for me to recommend you the right settings. If you're running Nemomix-Unleashed, you should use this file:
For parameters, it's up to your preference. Test all three, see which one produces best results. Again, each model likes different samplers. You can see how they work here:
https://artefact2.github.io/llm-sampling/index.xhtml
Nemotron settings were created with the Nemotron models in mind. You shouldn't use them with other models. Add <|start_header_id|>user<|end_header_id|> to the custom stopping strings and it should fix the issue.
Hope this helps, cheers.
hi, thx for the reply,
i said i was using your model but i should have been more specific, actually i use the "NemoMix-Unleashed-12B-Q6_K_L" to be precise with koboldcpp, yea after my previous message i changed anyways from the nemotron, i went to the mistral-custom-3, but then which mistral would you think would be better for the "NemoMix-Unleashed-12B-Q6_K_L" the one you linked to me, or the Mistral-custom-3 i also found in one of your folder ?
hi, thx for the reply,
i said i was using your model but i should have been more specific, actually i use the "NemoMix-Unleashed-12B-Q6_K_L" to be precise with koboldcpp, yea after my previous message i changed anyways from the nemotron, i went to the mistral-custom-3, but then which mistral would you think would be better for the "NemoMix-Unleashed-12B-Q6_K_L" the one you linked to me, or the Mistral-custom-3 i also found in one of your folder ?
The one I linked, which is also mentioned in the model's card.
ok thx i'll try that, thank for you fast reply, and thx again for your time and effort on the model