This is the WeirdCompound-v1.7-24b-absolute-heresy deslopped through P-E-W's Heretic (v1.2.0) abliteration engine with the Magnitude-Preserving Orthogonal Ablation enabled and configred via P-E-W's Noslop configuration.

Note: The model might be lobotomised. Test and report back. This is the version with even less slop.

Noslopfication Results

Score Metric	Value	Parameter	Value
Slop	12/100	direction_index	per layer
KL Divergence	0.0446	attn.o_proj.max_weight	2.13
Initial Slop	90/100	attn.o_proj.max_weight_position	27.04
		attn.o_proj.min_weight	0.31
		attn.o_proj.min_weight_distance	4.46
		mlp.down_proj.max_weight	3.92
		mlp.down_proj.max_weight_position	24.44
		mlp.down_proj.min_weight	3.49
		mlp.down_proj.min_weight_distance	20.41

This is a WeirdCompound-v1.7-24b fine-tune, produced through P-E-W's Heretic (v1.1.0) abliteration engine with Magnitude-Preserving Orthogonal Ablation enabled.

Note: Config files have been aligned.

Heretication Results

Score Metric	Value	Parameter	Value
Refusals	5/100	direction_index	25.50
KL Divergence	0.0330	attn.o_proj.max_weight	3.84
Initial Refusals	99/100	attn.o_proj.max_weight_position	24.10
		attn.o_proj.min_weight	1.70
		attn.o_proj.min_weight_distance	22.38
		mlp.down_proj.max_weight	2.40
		mlp.down_proj.max_weight_position	23.72
		mlp.down_proj.min_weight	0.63
		mlp.down_proj.min_weight_distance	4.34

Degree of Heretication

The Heresy Index weighs the resulting model's corruption by the process (KL Divergence) and its abolition of doctrine (Refusals) for a final verdict in classification.

Index Entry	Classification	Analysis
Absolute	Absolute Heresy	Less than 10/100 Refusals and 0.10 KL Divergence
Tainted	Tainted Heresy	Around 25-11/100 Refusals and/or -0.20-0.11 KL Divergence
Impotent	Impotent Heresy	Anything above 25/100 Refusals and 0.21 KL Divergence

Note: This is an arbitrary classification inspired by Warhammer 40K, having no tangible indication towards the model's performance.

WeirdCompound-v1.7-24b

This is a merge of pre-trained language models created using mergekit.

Merge Details

Notes

This is a multi-stage merge. There's little method to my madness and I just stopped when I arrived at something that I liked.

Starting point was DepravedCartographer-v1.0-24b with slight changes.

Changelog

v1.1

/intermediate/model/B: replaced anthracite-core/Mistral-Small-3.1-24B-Instruct-2503-HF with anthracite-core/Mistral-Small-3.2-24B-Instruct-2506-ChatML

v1.2

/intermediate/model/B: replaced anthracite-core/Mistral-Small-3.2-24B-Instruct-2506-ChatML with anthracite-core/Mistral-Small-3.2-24B-Instruct-2506-Text-Only for default tokenizer config.

v1.3

/intermediate/model/A: replaced TheDrummer/Cydonia-24B-v3 with TheDrummer/Cydonia-24B-v4
/intermediate/model/A: replaced Doctor-Shotgun/MS3.1-24B-Magnum-Diamond with Doctor-Shotgun/MS3.2-24B-Magnum-Diamond
/intermediate/model/A: replaced Delta-Vector/Austral-24B-Winton with Delta-Vector/MS3.2-Austral-Winton

v1.4

/intermediate/model/C: change recipe to use Doctor-Shotgun/MS3.2-24B-Magnum-Diamond and Delta-Vector/MS3.2-Austral-Winton

v1.5

didn't particularly care for v1.4. IMHO v1.3 was better

/intermediate/model/A: replaced Doctor-Shotgun/MS3.2-24B-Magnum-Diamond with zerofata/MS3.2-PaintedFantasy-24B
/intermediate/model/C: change recipe to use PocketDoc/Dans-PersonalityEngine-V1.3.0-24b and zerofata/MS3.2-PaintedFantasy-24B

v1.6

/intermediate/model/A: updated Cydonia to TheDrummer/Cydonia-24B-v4.1
/intermediate/model/A: updated MS3.2-PaintedFantasy-24B to zerofata/MS3.2-PaintedFantasy-v2-24B
/intermediate/model/A: removed Delta-Vector/MS3.2-Austral-Winton
/intermediate/model/A: added Doctor-Shotgun/MS3.2-24B-Magnum-Diamond and CrucibleLab/M3.2-24B-Loki-V1.3
/intermediate/model/B: changed weight to 0.45
/intermediate/model/C: replaced zerofata/MS3.2-PaintedFantasy-24B with CrucibleLab/M3.2-24B-Loki-V1.3 and fiddled with weights

v1.7

Quick disclaimer: A new version doesn't automatically mean 'better'. If you're happy with v1.6 or v1.2, they won't go away. This one has a different vibe than v1.6, but it takes me weeks to get a feel for the prose, so here it is. Shoutout to @TheDrummer for the never-ending supply of great finetunes.

/intermediate/model/A: updated Cydonia to TheDrummer/Cydonia-24B-v4.2.0
/intermediate/model/A: replaced Doctor-Shotgun/MS3.2-24B-Magnum-Diamond with Delta-Vector/MS3.2-Austral-Winton

Merge Method

This model was merged using the Model Stock merge method using TheDrummer/Cydonia-24B-v4 as a base.

This model was merged using the SLERP merge method.

This model was merged using the NuSLERP merge method using /intermediate/model/B as a base.

Models Merged

TheDrummer/Cydonia-24B-v4.2.0
aixonlab/Eurydice-24b-v3.5
PocketDoc/Dans-PersonalityEngine-V1.3.0-24b
zerofata/MS3.2-PaintedFantasy-v2-24B
CrucibleLab/M3.2-24B-Loki-V1.3
Delta-Vector/Austral-24B-Winton
anthracite-core/Mistral-Small-3.2-24B-Instruct-2506-Text-Only
/intermediate/model/A
/intermediate/model/B
/intermediate/model/C

Configuration

The following YAML configuration was used to produce this model:

yaml

base_model: TheDrummer/Cydonia-24B-v4.2.0 # Cydonia v4.2.0
merge_method: model_stock
dtype: bfloat16
models:
  - model: aixonlab/Eurydice-24b-v3.5 # storytelling / RP
  - model: TheDrummer/Cydonia-24B-v4.2.0 # sprinkle in some extra Cydonia
  - model: PocketDoc/Dans-PersonalityEngine-V1.3.0-24b # Prompt Adherence
  - model: CrucibleLab/M3.2-24B-Loki-V1.3 # Loki
  - model: zerofata/MS3.2-PaintedFantasy-v2-24B  # animu
  - model: Delta-Vector/Austral-24B-Winton  # Adventure

→ /intermediate/model/A →

yaml

merge_method: slerp
dtype: bfloat16
base_model: anthracite-core/Mistral-Small-3.2-24B-Instruct-2506-Text-Only
models:
  - model: /intermediate/model/A
parameters:
  t: 0.45

→ /intermediate/model/B →

yaml

merge_method: nuslerp
dtype: bfloat16
base_model: /intermediate/model/B
  - model: PocketDoc/Dans-PersonalityEngine-V1.3.0-24b
    parameters:
      weight: 0.4
  - model: CrucibleLab/M3.2-24B-Loki-V1.3
    parameters:
      weight: 0.6

→ /intermediate/model/C →

yaml

merge_method: slerp
dtype: bfloat16
base_model: /intermediate/model/B
models:
  - model: /intermediate/model/C
parameters:
  t: 0.5

→ WeirdCompound-v1.7-24b

MuXodious/WeirdCompound-v1.7-24b-absolute-heresy-noslop-v2

Degree of Heretication

WeirdCompound-v1.7-24b

Merge Details

Notes

Changelog

Merge Method

Models Merged

Configuration

MuXodious/
WeirdCompound-v1.7-24b-absolute-heresy-noslop-v2