In a Training Loop 🔄

David Belton PRO

DavidAU

AI & ML interests

Application(s) of single/multiple LLMs in specialized use cases & automation tasks. LLM, Prompt , System Role and Parameter engineering VIA chat / API. 2700+ merge models, over 600 fine tunes, over 1000 models published. DISCORD: David_AU [drawless111]

Recent Activity

new activity about 12 hours ago

DavidAU/Qwen3.6-27B-NEO-CODE-Di-IMatrix-MAX-GGUF:Source

new activity about 12 hours ago

DavidAU/Qwen3.5-46B-A3B-HighIQ-Prototype-II:Request

new activity about 12 hours ago

DavidAU/GLM-4.7-Flash-Uncensored-Heretic-NEO-CODE-Imatrix-MAX-GGUF:Hoping someone can finally help me with this

View all activity

Organizations

posted an update 1 day ago

Post

481

Qwen3.6 27B - NEO-Code Imatrix Max GGUF Quants [exceeds Unsloth in key metrics]:

All quants benchmarked with 5 key metrics.
A DAVIDAU vs UNSLOTH Metrics showdown.
Quant quality exceeds Unsloth in key metrics.
IQ2_M to Q6 available.
Standout: IQ4XS at 94% of BF16 precision.
Full explainer for Quant metrics.

DavidAU/Qwen3.6-27B-NEO-CODE-Di-IMatrix-MAX-GGUF

1 reply

replied to their post 4 days ago

Currently working with Qwen 3.5/6 35B-A3B in the lab ; learning the "quirks" ; still a ways to go.

reacted to nightmedia's post with 👍 10 days ago

Post

2705

Updated gemma-4-E4B-it metrics

I noticed the chat template got updated, and tried it on the E4B, with surprising results in stabilizing the brainwave.

quant    arc   arc/e boolq hswag obkqa piqa  wino
mxfp8    0.480,0.656,0.797,0.608,0.400,0.755,0.665
mxfp4    0.455,0.607,0.851,0.585,0.402,0.744,0.651

Quant    Perplexity      Peak Memory   Tokens/sec
mxfp8    35.937 ± 0.525  14.80 GB      1153
mxfp4    36.746 ± 0.534  11.06 GB      1030

Old numbers

quant    arc   arc/e boolq hswag obkqa piqa  wino
mxfp8    0.404,0.489,0.825,0.586,0.392,0.734,0.661
mxfp4    0.414,0.508,0.854,0.562,0.378,0.717,0.645

Quant    Perplexity      Peak Memory   Tokens/sec
mxfp8    34.652 ± 0.502  14.80 GB      1146
mxfp4    35.203 ± 0.506  11.06 GB      1200

I will re-do all baselines soon based on the new template. It is completely expected that the model behavior will change as a result.

Here are the effects of the new template on few known distills from DavidAU

gemma-4-E4B-it-The-DECKARD-Expresso-Universe-HERETIC-UNCENSORED

quant    arc   arc/e boolq hswag obkqa piqa  wino
New template
mxfp8    0.518,0.709,0.755,0.657,0.418,0.759,0.626
mxfp4    0.485,0.682,0.792,0.641,0.432,0.746,0.635
Old template
mxfp8    0.506,0.697,0.754,0.661,0.416,0.757,0.627
mxfp4    0.487,0.670,0.792,0.644,0.430,0.748,0.624

gemma-4-E4B-it-GLM-4.7-Flash-HERETIC-UNCENSORED-Thinking

mxfp8   0.461,0.599,0.779,0.630,0.406,0.766,0.629
Old template
mxfp8   0.456,0.580,0.786,0.629,0.410,0.764,0.633

gemma-4-E4B-it-Claude-Opus-4.5-HERETIC-UNCENSORED-Thinking

mxfp8    0.509,0.705,0.806,0.646,0.416,0.773,0.650
Old template
mxfp8    0.502,0.692,0.809,0.650,0.420,0.771,0.651

2 replies

replied to their post 13 days ago

RE: 16-18 B ; yes, something running in the lab right now. (Gemma 4).

Also can make Qwen 3's (Version 3) moes like Llama3.2-8X3B as well ; I have some of these at my repo too.
I have built a few GPT-OSS ; and some 12B [mistral nemo] as well as mistral nemo "large" 15-17Bs...

A lot of options ;

replied to their post 15 days ago

Maybe in the future ; atm still learning/addressing quirks with these new Gemmas.
Google released three different arch structure here : "E", "MOE", and 31B dense.

Also plans to create larger Gemma 4s too ; which may work better for specific applications and/or work better period.
These are in the plans for next week.

posted an update 16 days ago

Post

6691

Going NUCLEAR: Gemma 4 E4B uncensored, and tuned exceeds Gemma 4 26B-A4B in critical benchmarks.

De-censored, tuned, and tuned again via Unsloth using custom in house datasets and methods:

DavidAU/gemma-4-E4B-it-The-DECKARD-Expresso-Universe-HERETIC-UNCENSORED-Thinking

8 replies

replied to their post 16 days ago

UPDATE:
https://huggingface.co/DavidAU/gemma-4-E4B-it-The-DECKARD-Expresso-Universe-HERETIC-UNCENSORED-Thinking

Exceeds Gemma4 26B-A4B in critical benchmarks.

replied to their post 16 days ago

Training a Gemma 4 Reap 19B-A4B right now ; should be done tomorrow, then testing.

RE: FRanken merge 26B-A3B ; yes, just need to make a map for Mergekit ; this is also in progress.

RE: Claudes ; depends on how reap turns out.
There are a lot of updates still in progress with Unsloth/Llamacpp RE: Gemma 4s atm too ;
There are also some dataset issues to address when training with Gemma 4s.

NOTE:
Just finished a number of fine tunes on Gemma 4's E4B ; which is a MOE LIKE model. These will release in the next day or so ; pending final testing.

posted an update 17 days ago

Post

7261

THREE Gemma 4 , 31B Uncensored Fine Tunes (via Unsloth, inhouse datasets):

Uncensored first, then tuned.
Some benchmarks posted, others pending.
Examples posted, detailed instructions.
Some GGUFs are up; others pending as of this writing.

Enjoy:

DavidAU/gemma-4-31B-it-Mystery-Fine-Tune-HERETIC-UNCENSORED-Thinking
DavidAU/gemma-4-31B-it-Grand-Horror-X-INTENSE-HERETIC-UNCENSORED-Thinking
DavidAU/gemma-4-31B-it-The-DECKARD-HERETIC-UNCENSORED-Thinking

UPDATE:
DavidAU/gemma-4-E4B-it-The-DECKARD-Expresso-Universe-HERETIC-UNCENSORED-Thinking

Exceeds Gemma4 26B-A4B in critical benchmarks.

6 replies

posted an update about 1 month ago

Post

5214

Power, Freedom and Character:
Qwen 3.5 40B Claude Opus Deckard UNCENSORED.

Expanded, and trained with Claude Opus 4.6 Dataset, but first it was Heretic'ed and trained with DECKARD - 5 hand crafted datasets to give the model character, point of view and intelligence... and a lot more.

Examples posted.

Several quant types available under quantizations:

DavidAU/Qwen3.5-40B-Claude-4.6-Opus-Deckard-Heretic-Uncensored-Thinking

posted an update about 1 month ago

Post

9038

Going for the GOLD: Qwen 3.5 40B Claude 4.5 Opus

Drastically larger, with performance to match.
Upgraded Jinja template too.

DavidAU/Qwen3.5-40B-Claude-4.5-Opus-High-Reasoning-Thinking

replied to their post about 1 month ago

UPDATE:
All of these are now up; and can be downloaded.
Awaiting quants.

replied to their post about 1 month ago

RE: 13B:
=> one is upscaled + trained, the other is merge of two 9Bs fine tunes (and upscaled).

They are hidden as of this writing (undergoing private testing), awaiting final metrics / eval.
If they "pass" ; they will be made public.

These will be active within 24-48 hrs pending results.

replied to their post about 1 month ago

Currently have full running 13B (GLM 4.7 Flash) - which is very strong ; and experimental 21Bs of Qwen 3.5.
These are trained.

These are in testing, and access is limited as of this writing.

As for MOEs:
This is a little more complicated as scripting must be written for Mergekit to "moe together" 0.8B, 2B, 4B, 9Bs etc etc.
A draft (by me) has been completed to do this; but not tested/debugged yet.

No time line here ; too many variables.

RE 35B moes ; it is possible to address this in a different way ; but I have not tried it yet.
This is a different approach than REAP.

posted an update about 2 months ago

Post

7117

21 Qwen 3.5 Fine Tunes (thinking and instruct) ; reg and uncensored (2B to 27B) exceed benchmarks, and work better than org models.

All are bench marked against org model.
Many exceed all benchmarks of org model.
Claude, GLM, Gemini and other distills.
Thinking AND dedicated Instruct versions.

Core goal: Increase benchmarks, and address long thinking blocks.

Highlights:

9B and 27B instruct "Claude" versions hit 624 and 675 on the "ARC-C" (hard challenge).

Thinking fine tunes exceed org model performance (in thinking mode).

In many cases there is a drastic reduction in thinking block size.

9B Claude Heretic Uncensored, GGUF :
-Neo, Code Imatrix (duel imatrix)
- Updated Jinja template
- Custom tensor enhancements.

DavidAU/Qwen3.5-9B-Claude-4.6-OS-Auto-Variable-HERETIC-UNCENSORED-THINKING-MAX-NEOCODE-Imatrix-GGUF

COLLECTION [21 models]:
https://huggingface.co/collections/DavidAU/qwen-35-08-2-4-9-27-35b-regular-uncensored

UPDATE:
Now 31 models, including experimental 21B and new 13B models.

5 replies

posted an update about 2 months ago

Post

5756

Gemma 3 27B - The record breaker (Heretic'ed (uncensored) ; then training on Unsloth):

arc_challenge,arc_easy,boolq,hellaswag,openbookqa,piqa, winogrande
0.661 ,0.816 ,0.878,0.763 ,0.464 ,0.808 ,0.762

For comparison:
Qwen3.5-27B-Text
qx86-hi 0.443,0.498,0.857,0.701,0.372,0.770,0.752

Trained on a HERETIC uncensored base too ;

DavidAU/Gemma3-27B-it-vl-Polaris-HI16-Heretic-Uncensored-INSTRUCT

posted an update 2 months ago

Post

6906

The "ERNIE" 21B MOE Distill High Reasoning Fine Tune Invasion:

3 Ernie 21B-A3B MOE Models (64 experts) fine tuned with Unsloth using Gemini Pro 3, Claude 4.5 Opus, and GLM 4.7 Flash high reasoning datasets.

All benched, all exceeding org model specs too.

https://huggingface.co/DavidAU/models?search=ernie

Enjoy the freedom and added power.

1 reply

posted an update 2 months ago

Post

6786

Gemma 3 (1b, 4b, 12b and 27b) - Uncensored full Reasoning/Thinking models fine tuned using top distill datasets.

20 Gemma 3 models 1B, 4B, 12B and 27B with full reasoning using GLM 4.7 Flash, GPT, Claude and Gemini datasets and more fully fine tuned using Unsloth.

Most models are Heretic'ed (uncensored) first, and tuned second.
This vastly improves the model.

Models are also bench marked and in almost all cases exceed org model metrics - and in some cases by a lot.

Enjoy the freedom and more powerful THINKING/REASONING and UNCENSORED Gemma 3s !

https://huggingface.co/collections/DavidAU/gemma-3-reasoning-thinking-models-incl-uncensored

UPDATE: Benchmarks added for almost all models, including "VS" with Heretic (untuned) source models too.

replied to their post 2 months ago

9 Heretic Uncensored LFM fine tunes are now up at my repo:

https://huggingface.co/DavidAU/models?sort=created&search=lfm

Model card updates in progress as I write this.
The merges will take a wee bit longer.

...and 5 more new "non-heretic" ones too.

replied to their post 2 months ago

@muxodious ; Excellent.
In the cue.

Important note:
I can make the base models W reasoning datasets; however the "Kimi Mega Brain" is a complex merge of these base models (trained with different datasets) by Nightmedia.

I will query Nightmedia to see if he will do an updated "Heretic" mega brain merge after the "heretic" versions are complete.

David Belton PRO

AI & ML interests

Recent Activity

Organizations

DavidAU's activity