ACKNOWLEDGEMENTS

Mad lads, who provided feedback
yamatazen - For high quality model merges
OG model authors - For making cool models
Arcee AI - For making mergekit
Team mradermacher - For awesome quants
DeathGodlike - For awesome quants in EXL3
Model mergers: Rickyz, Vortex, Edward Eagley
You, for trying out my models

WHAT USERS ARE SAYING

Ive been using KSR for a week now, and I really like it. It is very creative and has brought together story threads in a natural way, even older ones at large context. I really like it so far.
— Pacoeltaco

Ive had good results so far with my testing, smart(for 12b), follows prompts and uses things from the character card. the only thing I found negative so far is that it likes em dashes
— Background-Ad-5398

I've been comparing this against a couple finetunes and this model's balance is impressive. I hope to see more soon!
— gggrandma1990

SYSTEM REQUIREMENTS

Minimum	Recommended
• Inference: CPU • Memory: 12GB RAM • Context Window: 8K Tokens • Temp: 0.65	• Inference: GPU • Memory: 16GB+ VRAM • Context Window: 16K Tokens • Temp: 0.80 • Courage: High

RECOMMENDED SETTINGS

Temp: 0.65-0.8 | RepPen: 1.05
TOP_P: 0.95 | MIN_P: 0.05 | TOP_K: 0
Template Format: ChatML | Ctxt: 16K

CORE FEATURES

ADVANCED PSYCHOLOGICAL PROFILING
Characters feel realer than real - with all the darkness that implies
DREAD ATMOSPHERIC SYSTEMS
Environments that breathe, bleed, and remember your fears
EMOTIONAL EROSION ENGINE
Watch characters unravel under psychological pressure
UNFILTERED NARRATIVE DEPTH
Darker than Eclipse, more psychologically intense than Radiance

FAQ

Q: IS THIS MODEL BETTER THAN X?
A: Different weapon, different war. Erosion specializes in psychological depth and darker narratives. It's smarter than previous versions, but choose your tool for the mission.

Q: WHEN'S THE NEXT VERSION COMING?
A: Currently experimenting with finetuning tech. Expect experimental merges while we develop the next major build.

Q: PLANNING OTHER ARCHITECTURES?
A: Affirmative. Reconnaissance underway for larger models - Mistral Small and Qwen3 are potential candidates for future deployments.

THE ULTIMATE RP ENGINE JUST GOT DARKER

They told us to extract the sakura essence. They never told us it would remember.

KansenSakura: Erosion represents the final stage of neural corruption - where beauty becomes a weapon and every memory turns to poison.

Some blossoms grow best in poisoned soil. Welcome to the garden.

Psychological Horror | Intense Themes

Complex Narratives | NSFW

KansenSakura Project | Containment Failed

RATED M for MATURE

Click the dots to navigate through the book.

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

Merge Details

Merge Method

This model was merged using the Multi-SLERP merge method.

Models Merged

The following models were included in the merge:

Configuration

The following YAML configuration was used to produce this model:

yaml

merge_method: multislerp
models:
  - model: yamatazen/EtherealAurora-12B-Lorablated
    parameters:
      weight: [1.000, 1.000, 1.000, 1.000, 0.968, 0.744, 0.256, 0.030, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.0225, 0.1275, 0.15]
  - model: Retreatcost/Shisa-K-sakurization
    parameters:
      weight: [0.000, 0.000, 0.000, 0.030, 0.256, 0.744, 0.968, 1.000, 1.000, 1.000, 1.000, 0.968, 0.744, 0.256, 0.030, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.015, 0.085, 0.10]
  - model: ./impish-ink-long
    parameters:
      weight: [0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.030, 0.256, 0.744, 0.968, 1.000, 1.000, 1.000, 0.968, 0.744, 0.256, 0.030, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.0075, 0.0425, 0.05]
  - model: Retreatcost/Lorablated-w2bb-psy-della
    parameters:
      weight: [0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.030, 0.256, 0.744, 0.968, 1.000, 1.000, 1.000, 0.968, 0.744, 0.256, 0.030, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000]
  - model: Retreatcost/Forgotten-directive-Neon-stock
    parameters:
      weight: [0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.030, 0.256, 0.744, 0.968, 1.000, 1.000, 1.000, 0.968, 0.744, 0.256, 0.030, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000]
  - model: Retreatcost/Irix-mpf-stock
    parameters:
      weight: [0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.030, 0.256, 0.744, 0.968, 1.000, 1.000, 1.000, 1.000, 1.000, 0.850, 0.150, 0.000]
  - model: ./retokenized_NB
    parameters:
      weight: [0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.000, 0.105, 0.595, 0.700]
dtype: bfloat16
parameters:
  normalize: true
tokenizer_source: Retreatcost/KansenSakura-Radiance-RP-12b

Retreatcost/KansenSakura-Erosion-RP-12b

KansenSakura-Erosion-RP-12b