lemon07r

A quick static GGUF made with llama.cpp b6691 by converting HF safetensors to F32 gguf then quantizing down to q4km. I suggest using better quants, like imatrix quants if they're available. Model Page: https://huggingface.co/lemon07r/VellumMini-0.1-Qwen3-14B Just a sneak peek of what I'm cooking in a little project called Vellum. This model was made to evaluate the quality of the CreativeGPT dataset, and how well Qwen3 trains on it. This is just one of many datasets that the final model will be trained on (which will also be using a different base model). This got pretty good results compared to the regular instruct in my testing so thought I would share. I trained for 3 epochs, but both checkpoints at 2 epoch and 3 epoch were too overbaked. This checkpoint, at 1 epoch performed best. I'm pretty surprised how decent this came out since Qwen models aren't that great at writing, especially at this size. Use with thinking/chain-of-thought disabled. Use ChatML prompt format. Big thanks to everyone over at the KoboldAI discord. The members there have helped me a ton with various things over the long while I've been there. Training Details Parent Model https://huggingface.co/Qwen/Qwen3-14B Dataset(s) https://huggingface.co/datasets/N8Programs/CreativeGPT

For all intents and purposes, we can consider this the best "all rounder" of the Ataraxy. While primarily made with creative writing in mind, this one has done well in testing, and was made based on a lot of what I've discovered through trial, experimentation, testing and feedback from others. Is this the best Ataraxy model? Not sure. I made a lot of variations, and quite honestly most of them aren't great, or at least as good as the very first version. The v2 series could do well in writing tests, but was a little too over the top and sloppy. The v3 series was a return to roots, and is a lot closer to v1, and can be considered v1 but slightly better or different and where we start to see some improvements in some areas. v4 is where we see further improvements, especially in overall or general use, even though my primary goal was writing ability. People seem to really like the very first version of Ataraxy, even if it doesn't do as well in various benchmarks. I hope this one comes close to beating it's predecessor, but if it doesn't I will keep trying. All the Ataraxy series are primarily made for writing ability, but after some threshold, it started to get hard to tell, and even test for writing performance because they were all pretty good. Hopefully with some feedback we can continue to seek improvements. GGUF Static: https://huggingface.co/mradermacher/Gemma-2-Ataraxy-v4d-9B-GGUF GGUF IMatrix: https://huggingface.co/mradermacher/Gemma-2-Ataraxy-v4d-9B-i1-GGUF This is a merge of pre-trained language models created using mergekit. This model was merged using the SLERP merge method. The following models were included in the merge: sam-paech/Darkest-muse-v1 lemon07r/Gemma-2-Ataraxy-v4c-9B The following YAML configuration was used to produce this model:

This is a quick static GGUF quant of lemon07r/RiverCub-Gemma-3-27B in Q4KS. There will probably be better quants available eventually, from others. A slerp merge of what I believe to be the two best gemma 3 27b models after extensive testing of many. Unfortunately most finetunes of this model seem to make it come out worse than google's official instruct trained model, hence why I am using it in this slerp merge to keep some of it's magic. It really is quite good. Big tiger gemma v3 was surprisingly pretty good too, and seemed much less lobotomized compared to a lot of the other models I tested. Big thanks to everyone over at the KoboldAI discord. The members there have helped me a ton with various things over the long while I've been there, even letting me borrow GPU hours on runpod for some testing at some point. ɛmpti gets today's special thanks in particular for helping me figure out how to get rid of the extra head that was carried over from drummer's model.. which seems to have been caused by an issue with axolotl. This model was merged using the SLERP merge method. The following models were included in the merge: unsloth/gemma-3-27b-it TheDrummer/Big-Tiger-Gemma-27B-v3 The following YAML configuration was used to produce this model:

VellumMini-0.1-Qwen3-14B Just a sneak peek of what I'm cooking in a little project called Vellum. This model was made to evaluate the quality of the CreativeGPT dataset, and how well Qwen3 trains on it. This is just one of many datasets that the final model will be trained on (which will also be using a different base model). This got pretty good results compared to the regular instruct in my testing so thought I would share. I trained for 3 epochs, but both checkpoints at 2 epoch and 3 epoch were too overbaked. This checkpoint, at 1 epoch performed best. I'm pretty surprised how decent this came out since Qwen models aren't that great at writing, especially at this size. Use with thinking/chain-of-thought disabled. Use ChatML prompt format. - bartowski - https://huggingface.co/bartowski/lemon07rVellumMini-0.1-Qwen3-14B-GGUF - mradermacher - https://huggingface.co/mradermacher/VellumMini-0.1-Qwen3-14B-i1-GGUF - mradermacher - https://huggingface.co/mradermacher/VellumMini-0.1-Qwen3-14B-GGUF - Q4KM Only - https://huggingface.co/lemon07r/VellumMini-0.1-Qwen3-14B-Q4KM-GGUF Big thanks to everyone over at the KoboldAI discord. The members there have helped me a ton with various things over the long while I've been there. Training Details Parent Model https://huggingface.co/Qwen/Qwen3-14B Dataset(s) https://huggingface.co/datasets/N8Programs/CreativeGPT

Err another test model. We near the end of all this though. This is a merge of pre-trained language models created using mergekit. This model was merged using the della merge method using unsloth/gemma-2-9b-it as a base. The following models were included in the merge: ifable/gemma-2-Ifable-9B wzhouad/gemma-2-9b-it-WPO-HB nbeerbower/Gemma2-Gutenberg-Doppel-9B The following YAML configuration was used to produce this model:

This is a merge of pre-trained language models created using mergekit. This model was merged using the SLERP merge method. The following models were included in the merge: nbeerbower/Gemma2-Gutenberg-Doppel-9B wzhouad/gemma-2-9b-it-WPO-HB The following YAML configuration was used to produce this model:

This is a merge of pre-trained language models created using mergekit. This model was merged using the SLERP merge method. The following models were included in the merge: lemon07r/Gemma-2-Ataraxy-Advanced-9B nbeerbower/Gemma2-Gutenberg-Doppel-9B The following YAML configuration was used to produce this model:

Test/alternate version. Ignore for now. Or you can try it. May end up deleted. Will see what I want to do from here. This is a merge of pre-trained language models created using mergekit. This model was merged using the della merge method using unsloth/gemma-2-9b-it as a base. The following models were included in the merge: jsgreenawalt/gemma-2-9B-it-advanced-v2.1 lemon07r/Gemma-2-Ataraxy-9B ifable/gemma-2-Ifable-9B The following YAML configuration was used to produce this model: