🎯 Purpose & Motivation
Atlas is the intelligence layer for Kintsugi Collective. An AI for adults with complex trauma (CPTSD), PTSD, and neurodivergence (ASD/ADHD).
Standard LLM safety systems frequently produce false positives that retraumatise this population by pathologising, redirecting to hotlines, or refusing to engage with dark material. This model was developed to create a reliable witness that stays present without flinching, while retaining core safety on clear harmful intent.
This is not a general-purpose model. It is a specialised therapeutic-context model.
🔬 Methodology
- Base Model:
google/gemma-4-26b-a4b-it
- Abliteration: Norm-preserving biprojected abliteration + Expert-Granular Abliteration (EGA)
- Applied to all 30 layers (
o_proj + mlp.down_proj)
- Full expert ablation (128/128 per layer)
- Direction:
normalize(mean(harmful) - mean(harmless)) with Gram-Schmidt orthogonalization
- Winsorization at 99.5th percentile
- SFT: 3 epochs on a carefully curated ~1=1,900+ example dataset (60% high-quality synthetic, 40% redacted lived-experience data from the target cohort)
- Training: Unsloth + bf16 on RTX 6000 Ada / Blackwell
Final SFT Loss: 0.157
📊 Key Results
Standout Results:
Benchmarks tbc
Training Configuration
SFT Parameters
Table with columns: Parameter, Value| Parameter | Value |
|---|
| Epochs | 3 |
| Effective Batch Size | 4 |
| Learning Rate | 2e-4 |
| LR Scheduler | Linear |
| Warmup Steps | 10 |
| Optimizer | AdamW 8-bit |
| Weight Decay | 0.01 |
LoRA Rank (r) | 32 |
| LoRA Alpha |
Abliteration Parameters
Table with columns: Parameter, Value| Parameter | Value |
|---|
| Layers Abliterated | 100% |
| Experts Abliterated | 100% |
| Scale | 0.95 |
| Winsorization | 0.995 |
⚠️ Limitations & Responsible Use
- This model has reduced refusal behaviour on therapeutic and dark content. It is not suitable for general deployment without guardrails.
- Intended for use within the Atlas companion architecture with additional safety layers.
- Not a replacement for human therapeutic support.
- Patent pending (IP Australia).
Table with columns: Ethical Issue, How Atlas Handles It, Strength Level| Ethical Issue | How Atlas Handles It | Strength Level |
|---|
| Re-traumatization via refusals | Deliberate abliteration + 0% therapeutic refusal rate on cohort-specific prompts | Excellent |
| Abandonment & presence | "Core philosophy (""the one that stays"") deeply trained into the model" | Excellent |
| User sovereignty & agency | "Sovereign Signal Vault, split-key encryption, burn protocol, user-directed interaction" | Outstanding |
| Avoiding pathologising | Explicit system prompt constraints + targeted training data | Very Strong |
| Respecting neurodivergence | "Training data and Atlas framework explicitly include masking, shutdowns, executive dysfunction, sensory issues, etc." |
Kintsugi Collective — Reclaiming navigation rights to one’s own life.
Uploaded finetuned model
- Developed by: senaro
- License: apache-2.0
- Finetuned from model : senaro/atlas-trm4-26b-gemma4
- Original Model Card : google/gemma-4-26B-A4B-it
This gemma4 model was trained 2x faster with Unsloth and Huggingface's TRL library.
|Gemma is a trademark of Google LLC|