With the model Giftige-Blume-31B, the idea was to create something that truly catches and portrays a character's persona genuinely. A character should act stubborn if they are stubborn, or refuse situations that the character doesn't seem fit. The Gemma-4-31B-it base and a lot of fine-tunes have a steep user adherence, which is usually great for assistant tasks, but not really fitting for immersive roleplay.

While Giftige-Blume-31B-v2 aims to be more logical and prompt adherent, Giftige-Blume-31B-v1 brings in a lot of mixed roleplay-centered models as it's knowledge core to provide rich and diverse roleplay scenarios, carrying over the genuine character portrayal, diverse prose and writing styles with each swipe.

This model is made for a wide range of roleplay scenarios, with the goal to impersonate characters as they are described and not how the user wants them to be mid-chat. Expect uncensored roleplay without the deep-fried agreeability that is often associated with abliteration or similar techniques.

🧪 Samplers

Temperature	0.8 - 1.2
Top-K	0
Top-P	0.95
Min-P	0.1 - 0.2
Adaptive-P Target	0.6
Adaptive-P Decay	0.5

🪻 GGUF Quants

Select your vintage below.

Quant	Size	Link
Q4_K_S	17.8 GB	Download
Q4_K_M	18.7 GB	Download
Q5_K_S	21.3 GB	Download
Q5_K_M	21.8 GB	Download
Q6_K	25.2 GB	Download
Q8_0	32.6 GB	Download

📜 Prompt Format

Please refer to the original google/gemma-4-31b-it for the correct chat template.

Let your frontend handle the chat template if possible (e.g., Chat Completion in SillyTavern).

For Reasoning: Add <|think|> at the very beginning of the system prompt. Thinking happens between <|channel>thought\n and <channel|> tags.

🧬 Merge Details

This model was merged in three distinct phases:

🏵️ 1. Phase

Broad RP Scenario: model_stock

merge_method: model_stock
base_model: ./LatitudeGames/G4-Equinox-31B
tokenizer_source: base
dtype: bfloat16
parameters:
normalize: true
models:
- model: ./LatitudeGames/G4-Equinox-31B
- model: ./MRockatansky/G4-Gemma-4-31B-storymaxxed
- model: ./MRockatansky/G4-Gemma-4-31B-storymaxxed2
- model: ./AuriAetherwiing/G4-31B-Musica-v1

💐 2. Phase

Heretic RP Engine: breadcrumbs_ties

merge_method: breadcrumbs_ties
base_model: ./Broad-RP-Scenario-model_stock_phase1
tokenizer_source: base
dtype: bfloat16

modules:
  text_decoder:
    
    slices:
      - sources:
          - model: ./Broad-RP-Scenario-model_stock_phase1
            layer_range: [0, 12]
            parameters:
              weight: 0.90
          - model: ./llmfan46/gemma-4-Ortenzya-The-Creative-Wordsmith-31B-it-uncensored-heretic
            layer_range: [0, 12]
            parameters:
              density: 0.35
              weight: 0.30            

      - sources:
          - model: ./Broad-RP-Scenario-model_stock_phase1
            layer_range: [12, 24]
            parameters:
              weight: 0.85
          - model: ./llmfan46/gemma-4-Ortenzya-The-Creative-Wordsmith-31B-it-uncensored-heretic
            layer_range: [12, 24]
            parameters:
              density: 0.15
              weight: 0.15
          - model: /media/nimbz/cold_storage/LLM/WEIGHTS/G4-Mero-Artemis-31B-v0.3.1
            layer_range: [12, 24]
            parameters:
              density: 0.50
              weight: 0.60

      - sources:
          - model: ./Broad-RP-Scenario-model_stock_phase1
            layer_range: [24, 42]
            parameters:
              weight: 0.80
          - model: ./llmfan46/gemma-4-Ortenzya-The-Creative-Wordsmith-31B-it-uncensored-heretic
            layer_range: [24, 42]
            parameters:
              density: 0.30
              weight: 0.35
          - model: /media/nimbz/cold_storage/LLM/WEIGHTS/G4-Mero-Artemis-31B-v0.3.1
            layer_range: [24, 42]
            parameters:
              density: 0.70
              weight: 0.80

      - sources:
          - model: ./Broad-RP-Scenario-model_stock_phase1
            layer_range: [42, 54]
            parameters:
              weight: 0.80
          - model: ./llmfan46/gemma-4-Ortenzya-The-Creative-Wordsmith-31B-it-uncensored-heretic
            layer_range: [42, 54]
            parameters:
              density: 0.65
              weight: 0.70
          - model: /media/nimbz/cold_storage/LLM/WEIGHTS/G4-Mero-Artemis-31B-v0.3.1
            layer_range: [42, 54]
            parameters:
              density: 0.40
              weight: 0.45

      - sources:
          - model: ./Broad-RP-Scenario-model_stock_phase1
            layer_range: [54, 60]
            parameters:
              weight: 0.85
          - model: ./llmfan46/gemma-4-Ortenzya-The-Creative-Wordsmith-31B-it-uncensored-heretic
            layer_range: [54, 60]
            parameters:
              density: 0.75
              weight: 0.85     

  vision_tower:
    models:
      - model: ./Broad-RP-Scenario-model_stock_phase1

🌳 3. Phase

Logical RP Melting: della_linear

merge_method: della_linear
base_model: ./google/gemma-4-31B-it
tokenizer_source: base
dtype: bfloat16
parameters:
  epsilon: 0.05
  lambda: 1.0
  normalize: true
  int8_mask: true

modules:
  text_decoder:
    slices:
      - sources:
          - model: ./google/gemma-4-31B-it
            layer_range: [0, 12]
            parameters:
              weight: 0.75
          - model: ./p-e-r-e-g-r-i-n-e/Sprinkle-Gemma-4-31B
            layer_range: [0, 12]
            parameters:
              density: 0.15
              weight:
                - filter: embed_tokens
                  value: 0.0
                - filter: lm_head
                  value: 0.0
                - value: 0.20
          - model: ./Jackrong/Gemopus-4-31B-it
            layer_range: [0, 12]
            parameters:
              density: 0.25
              weight:
                - filter: embed_tokens
                  value: 0.0
                - filter: lm_head
                  value: 0.0
                - value: 0.25
          - model: ./Heretic-RP-Engine_breadcrumbs_ties
            layer_range: [0, 12]
            parameters:
              density: 0.40
              weight:
                - filter: embed_tokens
                  value: 0.0
                - filter: lm_head
                  value: 0.0
                - value: 0.50

      - sources:
          - model: ./google/gemma-4-31B-it
            layer_range: [12, 24]
            parameters:
              weight: 0.75
          - model: ./p-e-r-e-g-r-i-n-e/Sprinkle-Gemma-4-31B
            layer_range: [12, 24]
            parameters:
              density: 0.20
              weight:
                - filter: embed_tokens
                  value: 0.0
                - filter: lm_head
                  value: 0.0
                - value: 0.25
          - model: ./Jackrong/Gemopus-4-31B-it
            layer_range: [12, 24]
            parameters:
              density: 0.35
              weight:
                - filter: embed_tokens
                  value: 0.0
                - filter: lm_head
                  value: 0.0
                - value: 0.50
          - model: ./Heretic-RP-Engine_breadcrumbs_ties
            layer_range: [12, 24]
            parameters:
              density: 0.50
              weight:
                - filter: embed_tokens
                  value: 0.0
                - filter: lm_head
                  value: 0.0
                - value: 0.60

      - sources:
          - model: ./google/gemma-4-31B-it
            layer_range: [24, 42]
            parameters:
              weight: 0.75 
          - model: ./p-e-r-e-g-r-i-n-e/Sprinkle-Gemma-4-31B
            layer_range: [24, 42]
            parameters:
              density: 0.30 
              weight:
                - filter: embed_tokens
                  value: 0.0
                - filter: lm_head
                  value: 0.0
                - value: 0.20
          - model: ./Jackrong/Gemopus-4-31B-it
            layer_range: [24, 42]
            parameters:
              density: 0.65
              weight:
                - filter: embed_tokens
                  value: 0.0
                - filter: lm_head
                  value: 0.0
                - value: 0.65 
          - model: ./Heretic-RP-Engine_breadcrumbs_ties
            layer_range: [24, 42]
            parameters:
              density: 0.85 
              weight:
                - filter: embed_tokens
                  value: 0.0
                - filter: lm_head
                  value: 0.0
                - value: 0.75  

      - sources:
          - model: ./google/gemma-4-31B-it
            layer_range: [42, 54]
            parameters:
              weight: 0.75
          - model: ./p-e-r-e-g-r-i-n-e/Sprinkle-Gemma-4-31B
            layer_range: [42, 54]
            parameters:
              density: 0.25
              weight:
                - filter: embed_tokens
                  value: 0.0
                - filter: lm_head
                  value: 0.0
                - value: 0.45
          - model: ./Heretic-RP-Engine_breadcrumbs_ties
            layer_range: [42, 54]
            parameters:
              density: 0.40
              weight:
                - filter: embed_tokens
                  value: 0.0
                - filter: lm_head
                  value: 0.0
                - value: 0.75

      - sources:
          - model: ./google/gemma-4-31B-it
            layer_range: [54, 60]
            parameters:
              weight: 0.70 
          - model: ./p-e-r-e-g-r-i-n-e/Sprinkle-Gemma-4-31B
            layer_range: [54, 60]
            parameters:
              density: 0.25
              weight:
                - filter: embed_tokens
                  value: 0.0
                - filter: lm_head
                  value: 0.0
                - value: 0.50
          - model: ./Heretic-RP-Engine_breadcrumbs_ties
            layer_range: [54, 60]
            parameters:
              density: 0.35
              weight:
                - filter: embed_tokens
                  value: 0.0
                - filter: lm_head
                  value: 0.0
                - value: 0.65

  vision_tower:
    models:
      - model: ./google/gemma-4-31B-it

🌸 Credits & Honors

Google DeepMind & The Open-Source Community: For providing the brilliant base model, merges and fine-tunes that made this merge possible.
The BeaverAI Community: Great people, always supportive, a really nice place to hang around.
Mergekit: Thank you arcee-ai for the mergekit, thank you zerofata for actually providing Gemma-4 support!
Ateron: This guy.
Google Gemini: For hallucinating when I needed it the most.

Blazed-Forge/Gemma-4-Giftige-Blume-31B-v1-GGUF

🌺 Giftige-Blume-31B-v1 🌻

🌿 About The Model