SubMaroon-exp
/

Boulesis-v2-26B-A4B

+---
+base_model:
+- coder3101/gemma-4-26B-A4B-it-heretic
+- Gryphe/Gemma-4-26B-A4B-StyleTune-V2
+- Gryphe/Pantheon-Reasoning-26B-A4B-1.1-V2
+license: gemma
+language:
+- en
+tags:
+- roleplay
+- merge
+- gemma4
+- moe
+- heretic
+- uncensored
+- sillytavern
+- thinking
+- reasoning
+---
+# Boulesis-v2-26B-A4B
+Composite Gemma 4 RP model (QK task arithmetic + fused LoRA). Second iteration.
+The idea was to retain the model's core intelligence and knowledge, while diversifying its prose and making it more decisive. I also wanted to sharpen its attention to context so it could dig deeper into the character card, organically pulling lore and facts into the roleplay rather than just mirroring the user.
+As a result, Boulesis moves beyond passive reactivity to genuinely advance the narrative, all while suffering zero catastrophic forgetting of its base intelligence.
+## What changed from v1
+| | v1 | v2 |
+|---|---|---|
+| QK donor | Pantheon-Reasoning-1.1 | Pantheon-Reasoning-1.1-**V2** |
+| QK alpha | 0.6 sliding q/k, 0.6 global q, **0** global k | **0.85** sliding q/k, **0.65** global q/k |
+| LoRA layers | all 30 | **10–28** |
+| LoRA targets | 55 | **35** |
+| Bake scale | 0.26 | **0.70** |
+| Training data | mixed EN/RU | **English only** |
+MoE experts, the router, embeddings, MLP and the vision tower are
+identical to the abliterated body, verified by tensor comparison after every
+step.
+<details><summary>How big the QK edit actually is</summary>
+I measured the task vector before merging instead of picking alpha by feel.
+Relative Frobenius norm against the base weights:
+| Group | mean | max |
+|---|---|---|
+| sliding_q | 0.00278 | 0.00518 |
+| sliding_k | 0.00263 | 0.00474 |
+| global_q | 0.00356 | 0.00436 |
+| global_k | 0.00286 | 0.00395 |
+Per-row rotation at the release alphas, in degrees:
+| Group | alpha | mean | max |
+|---|---|---|---|
+| sliding_q | 0.85 | 0.12 | 3.08 |
+| sliding_k | 0.85 | 0.12 | 5.87 |
+| global_q | 0.65 | 0.12 | 1.72 |
+| global_k | 0.65 | 0.09 | 9.53 |
+</details>
+## Using it
+## <span style="color: #22c55e;">YOU WILL GET THE BEST RESULTS WITH THINKING ON!</span>
+> [!Note]
+> Recommended settings:
+>
+>| Parameter | Value |
+>|---|---|
+>| **Temperature** | `1.0` |
+>| **Top-K** | `64` |
+>| **Repetition Penalty** | `1.05-1.1` |
+>| **Top-P** | `0.95` |
+>Thanks to [DifficultyThin8462](https://www.reddit.com/r/SillyTavernAI/comments/1w9abpf/comment/p8b064y/?utm_source=share&utm_medium=web3x&utm_name=web3xcss&utm_term=1&utm_content=share_button)
+<details>
+<summary><b>If the reasoning doesn't work when connecting GGUF ver. in KoboldCPP and SillyTavern</b></summary>
+You need to force this in KoboldCPP.
+Go to the Content tab and enable these options.
+![image](https://cdn-uploads.huggingface.co/production/uploads/66bb7e672b2247727188c91b/K9VZRBbeBwJaP8TGhpneH.png)
+For SillyTavern, it is recommended to set the template as shown:
+![image](https://cdn-uploads.huggingface.co/production/uploads/66bb7e672b2247727188c91b/PtHCVSkjzigx28byVjQuC.png)
+</details>
+## Credits
+Thanks to `coder3101` and `Gryphe` for the fine-tunes, and the entire
+`26B-Suite` team for their intellectual support.
+Speсial thanks for [Naphula](https://huggingface.co/Naphula),
+[redaihf](https://huggingface.co/redaihf) and
+[Vortex5](https://huggingface.co/Vortex5). You guys are awesome!