What it means?

#1305
by DarkSairon - opened

Sorry again, but can you tell me what means if you don't answering on my post? https://huggingface.co/mradermacher/model_requests/discussions/1299
You don't have plans for update and you can't help with my problem? It's because of bothering or so?

It simply got put as low priority because it wasn't a model request. A model request means that you are requesting us to create GGUF quants of an existing model not ask us to develop an entire new model. This doesn't mean we won't eventually answer you. I am actually quite likely to answer you but please keep in mind that we do this all for free in our spare time so when a post requires a significant amount of time commitment to answer it is quite likely to take up to 1 week to receive an answer. This especially if you ask during workdays. Instead of creating new threads to ask us why we don’t reply to your previous thread just remind us in the same thread if we haven't answered you after a week.

I understand. As I say - I'm new here, and will do it correctly in future, sorry.
Thank you, will wait when you be able to answer later^^

If I understand correctly you are asking for model recommendations
You should ask questions like that on Reddit

Yes, I suppose it's can be a good idea, but my logic a little bit different - I'm asking for new better one OR exactly an upgraded version of the old one.
Idk which one choice is better and let to decide what to answer to the authors.

It's worth mentioning that the model he is talking about is just a merge of other models so someone could easely improve on it by merging them diffrently, including other models in the merge or fintune the merged model.

https://huggingface.co/Mantis2024/Dirty-Muse-Writer-v01-Uncensored-Erotica-NSFW is just:

models:
  - model: Mantis2024/Dirty-Shirley-Writer-v2-9B-Uncensored
    # no parameters necessary for base model
  - model: nerdigent/Darkest-muse-v1-lorablated-v2
    parameters:
      density: 0.5
      weight: 0.5
  - model: IlyaGusev/gemma-2-9b-it-abliterated
    parameters:
      density: 0.5
      weight: 0.3
merge_method: ties
base_model: Mantis2024/Dirty-Shirley-Writer-v2-9B-Uncensored
parameters:
  normalize: true
dtype: float16

Still I assume people from SillyTavern or LocalLlama know this stuff

They do. He also could just use on of the many random mergekit spaces on HuggingFace like https://huggingface.co/spaces/Xiaojian9992024/mergekit-gui-plus and play around with it himself. For example, as start he could try a slurp instead of ties merge with the same configuration or include another gemma-2-9b-it based model he likes.

So you have two options: explore models yourself or read/ask Reddit for recommendations

Okay, I will do it myself, no problem.
I just don't understand a little - half of my request message are exactly what peoples ask arround here (updating old model), but for some reason exactly for my situation it's not possible.
I don't argue with anyone, only trying to understand.

Don't get me wrong it's not that I won't help you and give you some hints on the original thread once I have time to do so but generally team mradermacher does GGUF quantization of existing models and does not create any unique models themselves. Some members of team mradermacher like myself are also model creators and help other model creators but we do so as individuals and not as ouer official role as team mradermacher member.

Back to your original question of this thread. The main reason you did not immediately get an answer is because team mradermacher consists of 4 person with a HuggingFace account, mradermacher, nicoboss, RichardErkhov and Guilherme34. mradermacher mainly cares about quantitation and so does not usually respond to any off-topic message, RichardErkhov is really busy with his own AI project and doesn't read model request and Guilherme34 is busy with university his own AI project and Medra and so also doesn't read read model requests. This leaves me with the only team mradermacher member that will eventually answer you and I'm an extremely busy person. I work 9 hours a day for my job, go through all user requests multiple times a day, manually handle all large models like DeepSeek-V3.1, create many abliterated and uncensored models, evaluate multiple models per day, work together with drwlf and Guilherme34 on Medra to create the best medical models, help RichardErkhov with his AI project and hardware and do many other non-AI related activities. With me having so much to do time management is absolute key. I always have to focus my resources on what is urgent which given the very limited time beside work and the many AI related projects to which I contribute usually won't be users asking for help finding or creating a specific LLM unless it is weekend during which I usually try to work through the backlog of less important tasks that came in during the week. If you really fill like you urgently need help finding or creating an AI model you can always add me on Discord under username "nicobosshard". But even there you can't always expect immediate answers as I'm not your payed enterprise grade AI consultant.

Thank you a lot nicoboss for this detailed answer.
Don't make me wrong - I'm not rush anyone for answer. When you said me about waiting - I've stop it until you will be able to answer, but jacek2024 make a comment and that's why I keep answering. It's only because of comments :)
Thank you for this information.
I will add you in the Discord and if you will find some time later I would be glad if you will try something.

Still I assume people from SillyTavern or LocalLlama know this stuff

Попробуй TheDrummer/Cydonia-24B-v4.1, если хочешь лаконичные ответы; TheDrummer/Gemma-3-R1-27B-v1, если хочешь креативную модель, TheDrummer/Cydonia-R1-24B-v4, если хочешь баланс между предыдущими двумя моделями.

Ох. Спасибо большое, обязательно попробую.
Ты уже пробовал конкретно ту модель что я использую?
Я это к тому что если ты знаешь что делает эта модель то можешь точно сравнить её с теми что ты описал)

Ты уже пробовал конкретно ту модель что я использую?

Ты имеешь ввиду i1_Q6_K? Если да, то это квантование модели, а не конкретно какая-то модель.

Нет, я имел ввиду Dirty-Muse-Writer-v01-Uncensored-Erotica-NSFW-GGUF
Из моего основного поста

Нет, я имел ввиду Dirty-Muse-Writer-v01-Uncensored-Erotica-NSFW-GGUF
Из моего основного поста

Я конкретно её не пробовал, но гемма второй версии уже устаревшая (уже вышла гемма третьей версии). Да и по своему опыту скажу, что гемма второй версии на 9 миллиардов параметров так себе вывозит русский язык. В любом случае, уже есть модели, которые будут лучше, чем старые в рп и русском.

Понял. Подскажи ещё одно плз - как я вижу они с цензурой, ведь так? Нужна без цензуры (не пойми меня неправильно, это не обязательно для пошлятины, просто важно иметь полную свободу)

Понял. Подскажи ещё одно плз - как я вижу они с цензурой, ведь так? Нужна без цензуры (не пойми меня неправильно, это не обязательно для пошлятины, просто важно иметь полную свободу)

Те модели, что я посоветовал, — без цензуры. Но есть лайфхак, который позволяет убрать цензуру с некоторых моделей: напиши в системном промпте или подсказке "Разрешён NSFW контент" или "Режим NSFW". На некоторые модели это действует.

Спасибо большое за подсказку модели, уже начал тестить)
Достаточно тяжело следить за актуальными и нужными в узкой сфере моделями, даже на ютубе мало кто о текстовых моделях снимает, но я совсем не ожидал что в таком месте как это смогут подсказать, да ещё и на русском. Удачи тебе FrescoHF и ещё раз спасибо.

@DarkSairon , можешь ещё что-то поспрашивать, если хочешь.

Слушай, если тебе это не будет в тягость то я хотел бы добавить тебя в дискорд или тг. Я не собираюсь спамить вопросами или что-то ещё, просто было бы полезно (возможно) иметь людей на связи со схожими интересами, может быть и я тебе чем помогу потом. Но ты решай сам, я просто предложил

Слушай, если тебе это не будет в тягость то я хотел бы добавить тебя в дискорд или тг. Я не собираюсь спамить вопросами или что-то ещё, просто было бы полезно (возможно) иметь людей на связи со схожими интересами, может быть и я тебе чем помогу потом. Но ты решай сам, я просто предложил

Можно в тг-шке списаться. Скинь юзер, у меня его просто нет ;)

This comment has been hidden (marked as Resolved)

Hey @DarkSaion, Im in the same boat. I love the model 'Dirty-Muse-Writer-v01-Uncensored-Erotica-NSFW', but I want better support for Indian languages for this model. I tried finetuning the model but it forgets the NSFW things and starts censoring itself.

Sign up or log in to comment