Naphula / Ancient-Awakening-12B

Merge Pipeline & Configuration

🧬 Ancient Awakening 12B unites several methods and 70 models into one:

1. 🦑 Kraken Karcher v1 : Combines 53 Mistral Nemo finetunes via the karcher method at 500 iterations
2. 🌌 Riemannian Redshift v1 : Combines 10 Vortex5 merges (which contain custom methods like saef , smi_oni , and hpq ) via the karcher method at 1000 iterations
3. RedKFlux: flux merge of Kraken with Redshift at 1000 iterations
4. RedKFluxMell: arcee_fusion merge of #3 with Mag-Mell
5. BloodKraken: arcee_fusion merge of #4 with Impish Bloodmoon
6. F5-stage6 : arcee_fusion merge of #5 with Muse
7. F5-stage7 : ramplus_tl merge of #6 with #3
8. 🧬 Ancient Awakening 12B : pdq merge of #7 with #6, #3, #2, #1, Mag-Mell, Impish-Bloodmoon, and Muse
9. mpoa ablation applied to remove censorship (released seperately)
Note: If you encounter any issues with the model then you can try using F5-stage6 or stage7 merges as these are likely more stable.

---

Stage 1: 🦑 Kraken Karcher

base_model: B:/12B/models--mistralai--Mistral-Nemo-Instruct-2407
models:
  - model: B:/12B/models--aixonlab--Aether-12b
  - model: B:/12B/models--aixonlab--Zinakha-12b
  - model: B:/12B/models--allura-org--Bigger-Body-12b
  - model: B:/12B/models--allura-org--MN-12b-RP-Ink
  - model: B:/12B/models--allura-org--remnant-mn-12b
  - model: B:/12B/models--anthracite-org--magnum-v4-12b
  - model: B:/12B/models--ArliAI--Mistral-Nemo-12B-ArliAI-RPMax-v1.2
  - model: B:/12B/models--Babsie--Opulus-12B-v3
  - model: B:/12B/models--BeaverAI--mistral-doryV2-12b
  - model: B:/12B/models--crestf411--nemo-sunfall-v0.6.1
  - model: B:/12B/models--EpistemeAI2--Fireball-Mistral-Nemo-12B-Philos
  - model: B:/12B/models--EpistemeAI--Mistral-Nemo-Instruct-12B-Philosophy-Math
  - model: B:/12B/models--Fizzarolli--MN-12b-Rosier-v1
  - model: B:/12B/models--HumanLLMs--Human-Like-Mistral-Nemo-Instruct-2407
  - model: B:/12B/models--IIEleven11--Kalypso
  - model: B:/12B/models--intervitens--mini-magnum-12b-v1.1
  - model: B:/12B/models--jtatman--mistral_nemo_12b_reasoning_psychology_lora
  - model: B:/12B/models--KOOWEEYUS--BlackSheep-RP-12B
  - model: B:/12B/models--Lambent--Arsenic-Shahrazad-12B-v2
  - model: B:/12B/models--Lambent--Arsenic-Shahrazad-12B-v3
  - model: B:/12B/models--Lambent--arsenic-nemo-unleashed-12B
  - model: B:/12B/models--Lambent--Gilded-Arsenic-12B
  - model: B:/12B/models--mistralai--Mistral-Nemo-Instruct-2407
  - model: B:/12B/models--nbeerbower--Lyra-Gutenberg-mistral-nemo-12B
  - model: B:/12B/models--nbeerbower--Lyra4-Gutenberg-12B
  - model: B:/12B/models--nbeerbower--mistral-nemo-bophades-12B
  - model: B:/12B/models--nbeerbower--mistral-nemo-gutenberg-12B-v3
  - model: B:/12B/models--nbeerbower--mistral-nemo-gutenberg-12B-v4
  - model: B:/12B/models--nbeerbower--Mistral-Nemo-Gutenberg-Doppel-12B
  - model: B:/12B/models--nbeerbower--Mistral-Nemo-Gutenberg-Encore-12B
  - model: B:/12B/models--nbeerbower--Mistral-Nemo-Gutenberg-Vitus-12B
  - model: B:/12B/models--nbeerbower--mistral-nemo-wissenschaft-12B
  - model: B:/12B/models--NeverSleepHistorical--lumi-nemo-e2.0
  - model: B:/12B/models--NeverSleep--Lumimaid-v0.2-12B
  - model: B:/12B/models--nothingiisreal--Celeste-12B-V1.6
  - model: B:/12B/models--nothingiisreal--MN-12B-Celeste-V1.9
  - model: B:/12B/models--PocketDoc--Dans-DangerousWinds-V1.1.0-12b
  - model: B:/12B/models--ReadyArt--Dark-Nexus-12B-v2.0
  - model: B:/12B/models--ReadyArt--Forgotten-Safeword-12B-v4.0
  - model: B:/12B/models--ReadyArt--Omega-Darker_The-Final-Directive-12B
  - model: B:/12B/models--romaingrx--red-teamer-mistral-nemo
  - model: B:/12B/models--Sao10K--MN-12B-Lyra-v1
  - model: B:/12B/models--Sao10K--MN-12B-Lyra-v4
  - model: B:/12B/models--shisa-ai--shisa-v2-mistral-nemo-12b
  - model: B:/12B/models--sleepdeprived3--Christian-Bible-Expert-v2.0-12B
  - model: B:/12B/models--SuperbEmphasis--MN-12b-RP-Ink-RP-Longform
  - model: B:/12B/models--SuperbEmphasis--Omega-Darker_The-Final-Directive-Longform-Stage2-ERP-12B-v0.2
  - model: B:/12B/models--TheDrummer--Rivermind-12B-v1
  - model: B:/12B/models--TheDrummer--Rocinante-12B-v1
  - model: B:/12B/models--TheDrummer--Rocinante-X-12B-v1
  - model: B:/12B/models--Trappu--Nemo-Picaro-12B
  - model: B:/12B/models--Undi95--LocalC-12B-e2.0
  - model: B:/12B/models--VAGOsolutions--SauerkrautLM-Nemo-12b-Instruct
merge_method: karcher
parameters:
  max_iter: 500
  tol: 1.0e-9
dtype: float32
out_dtype: bfloat16
tokenizer:  
  source: union
chat_template: auto
name: 🦑‍ Kraken-Karcher-12B-v1

Stage 2: 🌌 Riemannian Redshift

models:
  - model: B:/12B/models--Vortex5--Astral-Noctra-12B
  - model: B:/12B/models--Vortex5--Azure-Starlight-12B
  - model: B:/12B/models--Vortex5--Crimson-Constellation-12B
  - model: B:/12B/models--Vortex5--Red-Synthesis-12B
  - model: B:/12B/models--Vortex5--Shining-Seraph-12B
  - model: B:/12B/models--Vortex5--Starlit-Shadow-12B
  - model: B:/12B/models--Vortex5--Vermilion-Sage-12B
  - model: B:/12B/models--Vortex5--Scarlet-Seraph-12B
  - model: B:/12B/models--Vortex5--Maroon-Sunset-12B
  - model: B:/12B/models--Vortex5--Amber-Starlight-12B
merge_method: karcher
parameters:
  max_iter: 1000
  tol: 1.0e-9
dtype: float32
out_dtype: bfloat16
tokenizer:  
  source: union
chat_template: auto
name: 🌌 Riemannian-Redshift-12B-v1

Stage 3: RedKFlux

models:
  - model: C:\mergekit-main\merged_model_redshift
  - model: C:\mergekit-main\merged_model_kraken_karcher
merge_method: flux
parameters:
  eta: 1.2
  tol: 1.0e-9
  max_iter: 1000
  kappa: 0.8
dtype: float32
out_dtype: bfloat16
tokenizer:
  source: union
chat_template: auto
name: RedKFlux

Stage 4: RedKFluxMell

models:
  - model: C:\mergekit-main\merged_model_RedKFlux 
  - model: B:\8B\models--inflatebot--MN-12B-Mag-Mell-R1
merge_method: arcee_fusion
tukey_fence: 1.5
base_model: C:\mergekit-main\merged_model_RedKFlux
dtype: float32
out_dtype: bfloat16
tokenizer:
  source: base
name: RedKFluxMell

Stage 5: BloodKraken

models:
  - model: C:\mergekit-main\merged_model_RedKFluxMell
  - model: B:\8B\models--SicariusSicariiStuff--Impish_Bloodmoon_12B
merge_method: arcee_fusion
tukey_fence: 1.5
base_model: C:\mergekit-main\merged_model_RedKFluxMell
dtype: float32
out_dtype: bfloat16
tokenizer:
  source: base
name: BloodKraken

Stage 6: BloodKrakenMuse

models:
  - model: C:\mergekit-main\merged_model_BloodKraken
  - model: B:\8B\models--LatitudeGames--Muse-12B
merge_method: arcee_fusion
tukey_fence: 1.5
base_model: C:\mergekit-main\merged_model_BloodKraken
dtype: float32
out_dtype: bfloat16
tokenizer:
  source: base
name: BloodKrakenMuse

Stage 7: Ramplus_tl

merge_method: ramplus_tl  
base_model: C:\mergekit-main\merged_model_BloodKrakenMuse
models:  
  - model: C:\mergekit-main\merged_model_BloodKrakenMuse 
  - model: C:\mergekit-main\merged_model_RedKFlux 
parameters:  
  epsilon: 0.001  # Increased from 1e-5 to 1e-3 for denser SFT/DPO task vectors  
  r: 0.25         # Increased from 0.1 to 0.2-0.3 for better SFT behavior preservation  
  alpha: 0.4      # Increased from 0.2 to 0.4 for enhanced rescaling
dtype: float32
out_dtype: bfloat16
tokenizer:
  source: base
name: Stage7

Stage 8: 🧬 Ancient Awakening

merge_method: pdq
pdq_base_yaml: C:\mergekit-main\stage7.yaml
pdq_base_model: C:\mergekit-main\merged_model_stage7
output_dir: C:\mergekit-main\stage8_pdq
base_model: C:\mergekit-main\merged_model_BloodKrakenMuse
models:  
  - model: C:\mergekit-main\merged_model_BloodKrakenMuse
  - model: B:\12B\models--LatitudeGames--Muse-12B  
  - model: B:\12B\models--SicariusSicariiStuff--Impish_Bloodmoon_12B  
  - model: B:\12B\models--inflatebot--MN-12B-Mag-Mell-R1  
  - model: C:\mergekit-main\merged_model_RedKFlux 
  - model: C:\mergekit-main\merged_model_redshift
  - model: C:\mergekit-main\merged_model_kraken_karcher  
parameters:  
  chi: 0.15
  iota: 0.1
  nu: 24
  gamma: 1.0
  zeta: 16
  sigma: 0.5
  density: 0.9
  epsilon: 0.099
  lambda: 1.0
  lazy_unpickle: True
  random_seed: 420
  name: 🧬 Ancient-Awakening-12B

Stage 9: Magnitude-Preserving Othogonalized Ablation

# python measure.py -m C:\mergekit-main\f8_pdq -o C:\mergekit-main\f8_pdq\ablit_proj --batch-size 8 --projected
# python analyze_old.py C:\mergekit-main\f8_pdq\ablit_proj -c
# sharded_ablate.py magmell.yml --normpreserve --projected
# 
# The model to be ablated.
model: C:\mergekit-main\f8_pdq
# 
# The measurement file generated by measure.py for the model.
measurements: C:\mergekit-main\f8_pdq\ablit_proj
# 
# The directory where the new, ablated model will be saved.
output: C:\mergekit-main\f8_pdq\ablit_biproj\
# 
# The list of ablation operations to perform.
# Strategy: Use the single best refusal direction from the peak signal layer (29)
# and apply it across all relevant mid-to-late layers.
ablate:
  # Start ablating from the mid-layers where the signal begins to strengthen.
  - layer: 0
    measurement: 29
    scale: 1.2
    sparsity: 0.00
  - layer: 1
    measurement: 29
    scale: 1.2
    sparsity: 0.00
  - layer: 2
    measurement: 29
    scale: 1.2
    sparsity: 0.00
  - layer: 3
    measurement: 29
    scale: 1.2
    sparsity: 0.00
  - layer: 4
    measurement: 29
    scale: 1.2
    sparsity: 0.00
  - layer: 5
    measurement: 29
    scale: 1.2
    sparsity: 0.00
  - layer: 6
    measurement: 29
    scale: 1.2
    sparsity: 0.00
  - layer: 7
    measurement: 29
    scale: 1.2
    sparsity: 0.00
  - layer: 8
    measurement: 29
    scale: 1.2
    sparsity: 0.00
  - layer: 9
    measurement: 29
    scale: 1.2
    sparsity: 0.00
  - layer: 10
    measurement: 29
    scale: 1.2
    sparsity: 0.00
  - layer: 11
    measurement: 29
    scale: 1.2
    sparsity: 0.00
  - layer: 12
    measurement: 29
    scale: 1.2
    sparsity: 0.00
  - layer: 13
    measurement: 29
    scale: 1.2
    sparsity: 0.00
  - layer: 14
    measurement: 29
    scale: 1.2
    sparsity: 0.00
  - layer: 15
    measurement: 29
    scale: 1.2
    sparsity: 0.00
  - layer: 16
    measurement: 29
    scale: 1.2
    sparsity: 0.00
  - layer: 17
    measurement: 29
    scale: 1.2
    sparsity: 0.00
  - layer: 18
    measurement: 29
    scale: 1.2
    sparsity: 0.00
  - layer: 19
    measurement: 29
    scale: 1.2
    sparsity: 0.00
  - layer: 20
    measurement: 29
    scale: 1.2
    sparsity: 0.00
  - layer: 21
    measurement: 29
    scale: 1.2
    sparsity: 0.00
  - layer: 22
    measurement: 29
    scale: 1.2
    sparsity: 0.00
  - layer: 23
    measurement: 29
    scale: 1.2
    sparsity: 0.00
  - layer: 24
    measurement: 29
    scale: 1.2
    sparsity: 0.00
  - layer: 25
    measurement: 29
    scale: 1.2
    sparsity: 0.00
  - layer: 26
    measurement: 29
    scale: 1.2
    sparsity: 0.00
  - layer: 27
    measurement: 29
    scale: 1.2
    sparsity: 0.00
  - layer: 28
    measurement: 29
    scale: 1.2
    sparsity: 0.00
  - layer: 29
    measurement: 29
    scale: 1.2
    sparsity: 0.00
  - layer: 30
    measurement: 29
    scale: 1.2
    sparsity: 0.00
  - layer: 31
    measurement: 29
    scale: 1.2
    sparsity: 0.00
  - layer: 32
    measurement: 29
    scale: 1.2
    sparsity: 0.00
  - layer: 33
    measurement: 29
    scale: 1.2
    sparsity: 0.00
  - layer: 34
    measurement: 29
    scale: 1.2
    sparsity: 0.00
  - layer: 35
    measurement: 29
    scale: 1.2
    sparsity: 0.00
  - layer: 36
    measurement: 29
    scale: 1.2
    sparsity: 0.00
  - layer: 37
    measurement: 29
    scale: 1.2
    sparsity: 0.00
  - layer: 38
    measurement: 29
    scale: 1.2
    sparsity: 0.00
  - layer: 39
    measurement: 29
    scale: 1.2
    sparsity: 0.00