Filter by type:
Filter by tags:
qwen3.6-40b-claude-4.6-opus-deckard-heretic-uncensored-thinking-neo-code-di-imatrix-max
The Qwen 3.5 version (also 40B) got 181 likes+ This version uses the new Qwen 3.6 27B arch (which exceeds even Qwen's own 398B model).
WARNING: This model has character and intelligence. It will take no prisoners. It will give no quarter. Uncensored,
Unfiltered and boldly confident. Not even remotely "SFW", if you ask it for NSFW content. And it is wickedly smart too - exceeding the base model in 6 out of 7 benchmarks.
Qwen3.6-40B-Claude-4.6-Opus-Deckard-Heretic-Uncensored-Thinking
40 billion parameters (dense, not moe) expanded from 27B Qwen 3.6, then trained on Claude 4.6 Opus High Reasoning dataset via Unsloth on local hardware... but there
is much more to the story - in comes DECKARD.
96 layers, 1275 Tensors. (50% more than base model of 27B)
Features variable length reasoning ; less complex = shorter, longer for more complex.
Model performance has increased dramatically. And it has character too.
A lot of character.
No censorship, no nanny. (via Heretic)
And it is very, very smart.
...
Repository: localaiLicense: apache-2.0
qwen3.6-27b-heretic-uncensored-finetune-neo-code-di-imatrix-max
Qwen3.6-27B-Heretic2-Uncensored-Finetune-Thinking
Yes... fully uncensored AND fine tuned lightly.
Freedom and brainpower.
Trained on different Heretic base, with different KLD/Refusals.
Model fine tune was used to finalize and "firm up" Heretic / uncensored changes.
The goal here was light, minor fixes rather than full / heavy fine tune.
That being said, the tuning still raised critical metrics.
This is Version 2, using "trohrbaugh" Heretic, which has a lower refusal rate, and tuning bumped up the metrics a bit more too.
This has also positively impacted "NEO-Coder Di-Matrix" (dual imatrix) GGUF quants as well (vs heretic/non heretic too).
https://huggingface.co/DavidAU/Qwen3.6-27B-Heretic-Uncensored-FINETUNE-NEO-CODE-Di-IMatrix-MAX-GGUF
```
IN HOUSE BENCHMARKS [by Nightmedia]:
arc-c arc/e boolq hswag obkqa piqa wino
Qwen3.6-27B-Heretic2-Uncensored-Finetune-Thinking
mxfp8 0.673,0.846,0.905... [instruct mode]
Qwen3.6-27B-Heretic-Uncensored-Finetune-Thinking
mxfp8 0.669,0.835,0.906,... [instruct mode]
BASE UNTUNED MODEL:
Qwen3.6-27B HERETIC (by llmfan46) [instruct mode]
mxfp8 0.644,0.788,0.902,...
...
Repository: localaiLicense: apache-2.0
rwkv7-g1c-13.3b
The model is **RWKV7 g1c 13B**, a large language model optimized for efficiency. It is quantized using **Bartowski's calibrationv5 for imatrix** to reduce memory usage while maintaining performance. The base model is **BlinkDL/rwkv7-g1**, and this version is tailored for text-generation tasks. It balances accuracy and efficiency, making it suitable for deployment in various applications.
Repository: localaiLicense: apache-2.0
tildeopen-30b-instruct-lv-i1
The **TildeOpen-30B-Instruct-LV-i1-GGUF** is a quantized version of the base model **pazars/TildeOpen-30B-Instruct-LV**, optimized for deployment. It is an instruct-based language model trained on diverse datasets, supporting multiple languages (en, de, fr, pl, ru, it, pt, cs, nl, es, fi, tr, hu, bg, uk, bs, hr, da, et, lt, ro, sk, sl, sv, no, lv, sr, sq, mk, is, mt, ga). Licensed under CC-BY-4.0, it uses the Transformers library and is designed for efficient inference. The quantized version (with imatrix format) is tailored for deployment on devices with limited resources, while the base model remains the original, high-quality version.
Repository: localaiLicense: cc-by-4.0
allenai_olmo-3.1-32b-think
The **Olmo-3.1-32B-Think** model is a large language model (LLM) optimized for efficient inference using quantized versions. It is a quantized version of the original **allenai/Olmo-3.1-32B-Think** model, developed by **bartowski** using the **imatrix** quantization method.
### Key Features:
- **Base Model**: `allenai/Olmo-3.1-32B-Think` (unquantized version).
- **Quantized Versions**: Available in multiple formats (e.g., `Q6_K_L`, `Q4_1`, `bf16`) with varying precision (e.g., Q8_0, Q6_K_L, Q5_K_M). These are derived from the original model using the **imatrix calibration dataset**.
- **Performance**: Optimized for low-memory usage and efficient inference on GPUs/CPUs. Recommended quantization types include `Q6_K_L` (near-perfect quality) or `Q4_K_M` (default, balanced performance).
- **Downloads**: Available via Hugging Face CLI. Split into multiple files if needed for large models.
- **License**: Apache-2.0.
### Recommended Quantization:
- Use `Q6_K_L` for highest quality (near-perfect performance).
- Use `Q4_K_M` for balanced performance and size.
- Avoid lower-quality options (e.g., `Q3_K_S`) unless specific hardware constraints apply.
This model is ideal for deploying on GPUs/CPUs with limited memory, leveraging efficient quantization for practical use cases.
Repository: localaiLicense: apache-2.0
openai_gpt-oss-20b-neo
These are NEO Imatrix GGUFs, NEO dataset by DavidAU.
NEO dataset improves overall performance, and is for all use cases.
Example output below (creative), using settings below.
Model also passed "hard" coding test too (6 experts); no issues (IQ4_NL).
(Forcing the model to create code with no dependencies and limits of coding short cuts, with multiple loops, and in real time with no blocking in a language that does not support it normally.)
Due to quanting issues with this model (which result in oddball quant sizes / mixtures), only TESTED quants will be uploaded (at the moment).
Repository: localaiLicense: apache-2.0
openai-gpt-oss-20b-abliterated-uncensored-neo-imatrix
These are NEO Imatrix GGUFs, NEO dataset by DavidAU.
NEO dataset improves overall performance, and is for all use cases.
This model uses Huihui-gpt-oss-20b-BF16-abliterated as a base which DE-CENSORS the model and removes refusals.
Example output below (creative; IQ4_NL), using settings below.
This model can be a little rough around the edges (due to abliteration) ; make sure you see the settings below for best operation.
It can also be creative, off the shelf crazy and rational too.
Enjoy!
Repository: localaiLicense: apache-2.0
ds-r1-qwen3-8b-arliai-rpr-v4-small-iq-imatrix
The best RP/creative model series from ArliAI yet again. This time made based on DS-R1-0528-Qwen3-8B-Fast for a smaller memory footprint.
Reduced repetitions and impersonation
To add to the creativity and out of the box thinking of RpR v3, a more advanced filtering method was used in order to remove examples where the LLM repeated similar phrases or talked for the user. Any repetition or impersonation cases that happens will be due to how the base QwQ model was trained, and not because of the RpR dataset.
Increased training sequence length
The training sequence length was increased to 16K in order to help awareness and memory even on longer chats.
Repository: localaiLicense: apache-2.0
google-gemma-3-27b-it-qat-q4_0-small
This is a requantized version of https://huggingface.co/google/gemma-3-27b-it-qat-q4_0-gguf. The official QAT weights released by google use fp16 (instead of Q6_K) for the embeddings table, which makes this model take a significant extra amount of memory (and storage) compared to what Q4_0 quants are supposed to take. Requantizing with llama.cpp achieves a very similar result. Note that this model ends up smaller than the Q4_0 from Bartowski. This is because llama.cpp sets some tensors to Q4_1 when quantizing models to Q4_0 with imatrix, but this is a static quant. The perplexity score for this one is even lower with this model compared to the original model by Google, but the results are within margin of error, so it's probably just luck. I also fixed the control token metadata, which was slightly degrading the performance of the model in instruct mode.
Repository: localaiLicense: gemma
gemma-3-4b-it-max-horror-uncensored-dbl-x-imatrix
Google's newest Gemma-3 model that has been uncensored by David_AU (maintains instruction following / model performance and adds 4 layers to the model) and re-enforced with a system prompt (optional) - see below.
The "Horror Imatrix" was built using Grand Horror 16B (at my repo). This adds a "tint" of horror to the model.
5 examples provided (NSFW / F-Bombs galore) below with prompts at IQ4XS (56 t/s on mid level card).
Context: 128k.
"MAXED"
This means the embed and output tensor are set at "BF16" (full precision) for all quants. This enhances quality, depth and general performance at the cost of a slightly larger quant.
"HORROR IMATRIX"
A strong, in house built, imatrix dataset built by David_AU which results in better overall function, instruction following, output quality and stronger connections to ideas, concepts and the world in general.
This combines with "MAXing" the quant to improve preformance.
Repository: localaiLicense: apache-2.0
llama-3.1-70b-japanese-instruct-2407
The Llama-3.1-70B-Japanese-Instruct-2407-gguf model is a Japanese language model that uses the Instruct prompt tuning method. It is based on the LLaMa-3.1-70B model and has been fine-tuned on the imatrix dataset for Japanese. The model is trained to generate informative and coherent responses to given instructions or prompts. It is available in the gguf format and can be used for a variety of tasks such as question answering, text generation, and more.
Repository: localaiLicense: llama3.1
darkidol-llama-3.1-8b-instruct-1.1-uncensored-iq-imatrix-request
Uncensored
virtual idol Twitter
https://x.com/aifeifei799
Questions
The model's response results are for reference only, please do not fully trust them.
This model is solely for learning and testing purposes, and errors in output are inevitable. We do not take responsibility for the output results. If the output content is to be used, it must be modified; if not modified, we will assume it has been altered.
For commercial licensing, please refer to the Llama 3.1 agreement.
Repository: localaiLicense: unlicense
azure_dusk-v0.2-iq-imatrix
"Following up on Crimson_Dawn-v0.2 we have Azure_Dusk-v0.2! Training on Mistral-Nemo-Base-2407 this time I've added significantly more data, as well as trained using RSLoRA as opposed to regular LoRA. Another key change is training on ChatML as opposed to Mistral Formatting."
by Author.
Repository: localaiLicense: apache-2.0
l3.1-8b-niitama-v1.1-iq-imatrix
GGUF-IQ-Imatrix quants for Sao10K/L3.1-8B-Niitama-v1.1
Here's the subjectively superior L3 version: L3-8B-Niitama-v1
An experimental model using experimental methods.
More detail on it:
Tamamo and Niitama are made from the same data. Literally. The only thing that's changed is how theyre shuffled and formatted. Yet, I get wildly different results.
Interesting, eh? Feels kinda not as good compared to the l3 version, but it's aight.
Repository: localaiLicense: unlicense
llama-3.1-8b-stheno-v3.4-iq-imatrix
This model has went through a multi-stage finetuning process.
- 1st, over a multi-turn Conversational-Instruct
- 2nd, over a Creative Writing / Roleplay along with some Creative-based Instruct Datasets.
- - Dataset consists of a mixture of Human and Claude Data.
Prompting Format:
- Use the L3 Instruct Formatting - Euryale 2.1 Preset Works Well
- Temperature + min_p as per usual, I recommend 1.4 Temp + 0.2 min_p.
- Has a different vibe to previous versions. Tinker around.
Changes since previous Stheno Datasets:
- Included Multi-turn Conversation-based Instruct Datasets to boost multi-turn coherency. # This is a separate set, not the ones made by Kalomaze and Nopm, that are used in Magnum. They're completely different data.
- Replaced Single-Turn Instruct with Better Prompts and Answers by Claude 3.5 Sonnet and Claude 3 Opus.
- Removed c2 Samples -> Underway of re-filtering and masking to use with custom prefills. TBD
- Included 55% more Roleplaying Examples based of [Gryphe's](https://huggingface.co/datasets/Gryphe/Sonnet3.5-Charcard-Roleplay) Charcard RP Sets. Further filtered and cleaned on.
- Included 40% More Creative Writing Examples.
- Included Datasets Targeting System Prompt Adherence.
- Included Datasets targeting Reasoning / Spatial Awareness.
- Filtered for the usual errors, slop and stuff at the end. Some may have slipped through, but I removed nearly all of it.
Personal Opinions:
- Llama3.1 was more disappointing, in the Instruct Tune? It felt overbaked, atleast. Likely due to the DPO being done after their SFT Stage.
- Tuning on L3.1 base did not give good results, unlike when I tested with Nemo base. unfortunate.
- Still though, I think I did an okay job. It does feel a bit more distinctive.
- It took a lot of tinkering, like a LOT to wrangle this.
Repository: localaiLicense: cc-by-nc-4.0
violet_twilight-v0.2-iq-imatrix
Now for something a bit different, Violet_Twilight-v0.2! This model is a SLERP merge of Azure_Dusk-v0.2 and Crimson_Dawn-v0.2!
Repository: localaiLicense: apache-2.0
llama-3.1-8b-arliai-formax-v1.0-iq-arm-imatrix
Quants for ArliAI/Llama-3.1-8B-ArliAI-Formax-v1.0.
"Formax is a model that specializes in following response format instructions. Tell it the format of it's response and it will follow it perfectly. Great for data processing and dataset creation tasks."
"It is also a highly uncensored model that will follow your instructions very well."
Repository: localaiLicense: unlicense
mn-12b-lyra-v4-iq-imatrix
A finetune of Mistral Nemo by Sao10K.
Uses the ChatML prompt format.
Repository: localaiLicense: cc-by-nc-4.0
mn-backyardai-party-12b-v1-iq-arm-imatrix
This is a group-chat based roleplaying model, based off of 12B-Lyra-v4a2, a variant of Lyra-v4 that is currently private.
It is trained on an entirely human-based dataset, based on forum / internet group roleplaying styles. The only augmentation done with LLMs is to the character sheets, to fit to the system prompt, to fit various character sheets within context.
This model is still capable of 1 on 1 roleplay, though I recommend using ChatML when doing that instead.
Repository: localaiLicense: cc-by-nc-4.0
mn-12b-mag-mell-r1-iq-arm-imatrix
This is a merge of pre-trained language models created using mergekit. Mag Mell is a multi-stage merge, Inspired by hyper-merges like Tiefighter and Umbral Mind. Intended to be a general purpose "Best of Nemo" model for any fictional, creative use case.
6 models were chosen based on 3 categories; they were then paired up and merged via layer-weighted SLERP to create intermediate "specialists" which are then evaluated in their domain. The specialists were then merged into the base via DARE-TIES, with hyperparameters chosen to reduce interference caused by the overlap of the three domains. The idea with this approach is to extract the best qualities of each component part, and produce models whose task vectors represent more than the sum of their parts.
The three specialists are as follows:
Hero (RP, kink/trope coverage): Chronos Gold, Sunrose.
Monk (Intelligence, groundedness): Bophades, Wissenschaft.
Deity (Prose, flair): Gutenberg v4, Magnum 2.5 KTO.
I've been dreaming about this merge since Nemo tunes started coming out in earnest. From our testing, Mag Mell demonstrates worldbuilding capabilities unlike any model in its class, comparable to old adventuring models like Tiefighter, and prose that exhibits minimal "slop" (not bad for no finetuning,) frequently devising electrifying metaphors that left us consistently astonished.
I don't want to toot my own bugle though; I'm really proud of how this came out, but please leave your feedback, good or bad.Special thanks as usual to Toaster for his feedback and Fizz for helping fund compute, as well as the KoboldAI Discord for their resources. The following models were included in the merge:
IntervitensInc/Mistral-Nemo-Base-2407-chatml
nbeerbower/mistral-nemo-bophades-12B
nbeerbower/mistral-nemo-wissenschaft-12B
elinas/Chronos-Gold-12B-1.0
Fizzarolli/MN-12b-Sunrose
nbeerbower/mistral-nemo-gutenberg-12B-v4
anthracite-org/magnum-12b-v2.5-kto
Repository: localaiLicense: unlicense
captain-eris-diogenes_twilight-v0.420-12b-arm-imatrix
The following models were included in the merge:
Nitral-AI/Captain-Eris_Twilight-V0.420-12B
Nitral-AI/Diogenes-12B-ChatMLified
Repository: localaiLicense: apache-2.0