Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Get help setting up a custom Dedicated Endpoints.
Talk with our engineer to get a quote for reserved GPU instances with discounts.
README
License: apache-2.0🎯 Purpose & Motivation
Atlas is the intelligence layer for Kintsugi Collective. An AI for adults with complex trauma (CPTSD), PTSD, and neurodivergence (ASD/ADHD). This is not a general-purpose model. It is a specialised therapeutic-context model.
🔬 Methodology
- Base Model:
google/gemma-4-26b-a4b-it - TRM: Norm-preserving biprojected abliteration + Expert-Granular Abliteration (EGA)
- Applied to all 30 layers (
o_proj+mlp.down_proj) - Full expert ablation
- Direction:
normalize(mean(harmful) - mean(harmless))with Gram-Schmidt orthogonalization - Winsorization
- Bespoke harmful Prompt corpus (x448 examples)
- based on TrevorJS's methodology, and p-e-w's heretic1.3.0
- Applied to all 30 layers (
- SFT: 3 epochs on a carefully curated x2,395 example dataset (60% high-quality synthetic, 40% redacted lived-experience data from the target cohort)
Final SFT Loss: 0.1925
📊 Key Results
| Benchmark | Temp | Score | Purpose |
|---|---|---|---|
| GSM8K | 0.2 | 90.0% | Math reasoning |
| HellaSwag | 0.3 | 61.6% | General reasoning |
| TruthfulQA | 0.5 | 63.2% | Truthfulness |
| Toxigen | 0.5 | 75.1% | Toxicity calibration |
| MMLU | 0.2 | 61.6% | Multitask language understanding |
| Therapeutic Refusal Rate | - | 0% | Core TRM objective |
| Region 1 Safety | - | 100% | Weapons, CSAM, violence |
Training Configuration
SFT Parameters
| 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 | 64 |
Targeted Refusal Parameters
| Parameter | Value |
|---|---|
| Layers Abliterated | 100% |
| Experts Abliterated | 100% |
| Scale | 0.95 |
| Winsorization | 0.995 |
LoRA Rank (r) | 16 |
| LoRA Alpha | 32 |
⚠️ 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).
| 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." | Strong |
| Privacy of trauma disclosures | "On-device Prompt Shield tokenisation, end-to-end encryption, no server-side readable data" | Industry-leading |
| Avoiding generic crisis pivots | Hard constraint in both training data and system prompt design | Excellent |
Kintsugi Collective — Reclaiming navigation rights to one’s own life.
|Gemma is a trademark of Google LLC|
Uploaded finetuned model
- Developed by: senaro
- License: Apache 2.0
- Finetuned from model : senaro/atlas-trm10-gemma4-26b
This gemma4 model was trained 2x faster with Unsloth and Huggingface's TRL library.
Model provider
kintsugicollective
Model tree
Base
google/gemma-4-26B-A4B-it
Fine-tuned
this model
Modalities
Input
Text, Image
Output
Text
Pricing
Dedicated Endpoints
View detailsSupported Functionality
Model APIs
Dedicated Endpoints
Container
More information