🔄 In a Training Loop

David Belton PRO

DavidAU

·

https://huggingface.co/DavidAU/support

AI & ML interests

Application(s) of single/multiple LLMs in specialized use cases. LLM, Prompt , System Role and Parameter engineering VIA chat / API. 2700+ merge models, over 1000 fine tunes, over 2000 models published. DISCORD: David_AU [drawless111]

Recent Activity

updated a collection about 1 hour ago

WS - Dark Planet Wordstorm Project - Random Prune / Form.

updated a collection about 1 hour ago

Thinking / Reasoning Models - Reg and MOEs.

updated a collection about 1 hour ago

Source files for GGUF, EXL2, AWQ, GPTQ, HQQ etc etc

View all activity

Organizations

updated 20 collections about 1 hour ago

WS - Dark Planet Wordstorm Project - Random Prune / Form.

Models using the org Dark Planet 8B formula, with random pruning / density, to create new Dark Planet versions with new abilities / generation. • 6 items • Updated about 1 hour ago • 5

Thinking / Reasoning Models - Reg and MOEs.

QwQ,DeepSeek, EXONE, DeepHermes, and others "thinking/reasoning" AIs / LLMs in regular model type, MOE (mix of experts), and Hybrid model formats. • 110 items • Updated about 1 hour ago • 29

Source files for GGUF, EXL2, AWQ, GPTQ, HQQ etc etc

Safetensor source files (by David_AU) to use directly and/or create different quants and/or merges. Link to GGUFS/full model card on each. • 54 items • Updated about 1 hour ago • 40

Rogue Creative Series - GGUF / Incl Source too.

Based on Llama 3.2 this is a specialized creative series. L3.2 has strong instruction following. Repos including example generations. Source below too • 9 items • Updated about 1 hour ago • 4

Reasoning Source files for GGUF, EXL2, AWQ, GPTQ

Reasoning model safetensor source files (by David_AU) to use directly and/or create different quants and/or merges. Link to GGUFS/full model card. • 25 items • Updated about 1 hour ago • 6

Qwen3 - 30B-A3B (128 experts) and higher

Versions of Qwen3's 30B MOE - Mixture of Experts 128 - in NEO, Horror, and Brainstorm configs from 30B up to 55B. • 18 items • Updated about 1 hour ago • 6

Qwen 3.5 - 0.8, 2, 4, 9, 27, 35B - regular / uncensored

Min 256k context + images : Reg, Heretic, Heretic fine tunes of Qwen 3.5 in all sizes. • 42 items • Updated about 1 hour ago • 73

Qwen 3 - Horror / Neo Imatrix / Max Qs / 32-256k ctx

Qwen3s: 0.6B, 1.7B, 4B, 8B, 14B, 30B (MOE), 32B using Neo / Horror / Other imatrix datasets, exp methods, and ways to improve reasoning/output gen. • 17 items • Updated about 1 hour ago • 5

OpenAI-GPT 20B, 37B ,120B: Neo, reg, uncensored, ablit.

OpenAi's model in various sizes and formats, including NEO Imatrix, DI, Tri Matrix, Uncensored, Albiterated, and Brainstorm 20x (37B). • 13 items • Updated about 1 hour ago • 17

MOE - Reasoning - Gated IQ Multi-Tier Models

MOE models using a unique custom internal structure to augment reasoning/thinking up to 300 % which then switches during generation mode. • 9 items • Updated about 1 hour ago • 5

Long Context - 16k,32k,64k,128k,200k,256k,512k,1000k

Listed oldest to newest. Some with up to 1 million context. • 41 items • Updated about 1 hour ago • 33

LFM 1.2B SOTA- 400-700 T/S - Enhanced/Fine Tunes // Distills

Fine tunes of LFM 1.2B (SOTA) models via Unsloth to enhance performance, including reasoning/thinking distills and "reasoning/thinking" replacements. • 13 items • Updated about 1 hour ago • 11

Instruct Models - Better instruction following.

Q6s Instruct models. Other Qs in model listing. Instruct models are better at one shot / api type usage in most cases vs "chat". Listed: old to new. • 20 items • Updated about 1 hour ago • 10

Higher Precision GGUFs / Imatrix Plus

Models compressed in higher precision with parts of the model compression remaining in F16/full precision. Increases overall quality in all tasks. • 37 items • Updated about 1 hour ago • 11

Heretic - Abliterated, Uncensored, Unrestricted POWER.

Models that have be abliterated using the HERETIC method. Done properly, this completely removed almost all censorship with no damage to the model. • 132 items • Updated about 1 hour ago • 176

Grand Horror 16.5B+ Horror and Fiction Generation

NSFW, Horror, Fiction Generation models of different sizes and calibrations. Can be used for all genres of fiction. [ NEWEST -> bottom of the list ] • 150 items • Updated about 1 hour ago • 88

GLM 4.7 Flash Distill Finetunes - 1B to 32B - reg/uncensored

Qwen, Mistral, Gemma, LFM, Ernie and other models fine tuned with GLM Datasets to improve and or ADD reasoning/thinking. • 22 items • Updated about 1 hour ago • 5

Gemma The Writer Series - GGUF / Source)

Gemma The Writer series of models, using GEMMA 2 as a base. Gemma shows exceptional instruction following. Repos contain examples + prose controls. • 24 items • Updated about 1 hour ago • 14

Gemma 3 - Reasoning/Thinking Models (incl uncensored)

Fine tunes (most uncensored) of Gemma3s to add reasoning/thinking. Increases output and image processing quality and general model metrics. • 18 items • Updated about 1 hour ago • 17

Gemini Distills 1B to 32B - Reg and Uncensored

Qwen, Gemma, Mistral, Ernie, LFM, Llama fine tuned via Unsloth on Gemini dataset(s) to improve and/or add reasoning/thinking to the model.. • 20 items • Updated about 1 hour ago • 5