⏳ 📜 Sphinsikus Chronist V2 📜 ⏳

This is a library of stories, ready to be told or written.

🐀 About the Model

This is an improvement of old Sphinsikus Chronist, by injecting Scotoma V2 into it. Model should write with less slop, and have more stability, without loosing adventure knowledge.

• Primary Focus: Character adherence. Role-play. Long context stories.

🗣️ Prompt Format

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

<|turn>user
Hello<turn|>
<|turn>model
Hi there<turn|>
<|turn>user
How are you?<turn|>
<|turn>model

🐀 Merge Details

I used mergekit as always:

Phase 1: Mixing (dare_ties)

I often trying to look for new methods and lurking around how models can be mixed, what can be used in it. Thanks to Nimbz for some hints, I goes over proj filters and I think, make some progression.

merge_method: dare_ties
base_model: F:\AI\Merge\Gemma-4-it
tokenizer_source: base
dtype: bfloat16
parameters:
  lambda: 1.0
models:
  - model: F:\AI\Merge\Gemma-4-Scotoma-V2
    parameters:
      density: [0.40, 0.40, 0.40, 0.50, 0.50]
      weight:
      - {filter: mlp.gate_proj, value: 0.4}
      - {filter: mlp.up_proj, value: 0.6}
      - {filter: mlp.down_proj, value: 0.4}            
      - {filter: self_attn.q_proj, value: 0.3}
      - {filter: self_attn.k_proj, value: 0.3}
      - {filter: self_attn.v_proj, value: 0.5}
      - {filter: self_attn.o_proj, value: 0.5}
      - value: [0.40, 0.40, 0.40, 0.50, 0.50]

  - model: F:\AI\Merge\Gemma-4-SphinsikusChronist
    parameters:
      density: [0.70, 0.70, 0.70, 0.70, 0.65]
      weight:
      - {filter: mlp.gate_proj, value: 0.6}
      - {filter: mlp.up_proj, value: 0.4}
      - {filter: mlp.down_proj, value: 0.6}
      - {filter: self_attn.q_proj, value: 0.7}
      - {filter: self_attn.k_proj, value: 0.7}
      - {filter: self_attn.v_proj, value: 0.5}
      - {filter: self_attn.o_proj, value: 0.5}
      - value: [0.60, 0.60, 0.60, 0.50, 0.50]

🤝 Special Thanks

• Google DeepMind: For providing the base model.
• The Open-Source Community: For cheering people that share my passion for AI and RP.
• Mergekit Fork: Zerofata - For making merges possible.
• AesSedai: This smart man, to deslopping goofy G4.
• To Nimbz: This fella for giving advices and helping me to improve technical knowledge.