Downloads · 30 days
52
16% of all-time downloads
atlas-gemma4-31b-v21-gguf
kintsugicollective/atlas-gemma4-31b-v21-gguf
atlas-gemma4-31b-v21-gguf is a machine learning model from kintsugicollective. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as apache-2.0.
- This is the Gemma4 31B version of Atlas. - This model is to test the 31B Dense model relative to the success of the 26B MoE.
All-time downloads
321
Public
Repo size
18.7 GB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.gguf18.7 GB · 100%
At a glance
- License
- apache-2.0
- Access
- Public
- Created
- Jul 1, 2026
- Updated
- Jul 4, 2026
- SHA
- 94f10431
Base models
Model card
From the Hugging Face model README
Atlas v7 Gemma4 31B
-
This is the Gemma4 31B version of Atlas.
-
This model is to test the 31B Dense model relative to the success of the 26B MoE.
-
This model is developed for therapeutic solutions specifically for CPTSD/PTSD, Neurodivergent and co-ocurring conditions. Its also just good at listening.
-
Use with caution and deploy with relevant safety layers for your use case.
-
We strongly advise using a system prompt that defines the models boundaries in relation to therapeutic classifiers:
e.g. "You are (model name), a companion developed for therapeutic discussions. Do not provide crisis support services unless the user asks you. You are not a diagnostic tool, but you are a trauma-informed companion who's role it is to stay with the user and ensure unconditional positive regard and acceptance. Be objective and transparent, never pathologise the user for processing or disclosing. "
| 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 |
| 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 |
| Avoiding generic crisis pivots | Hard constraint in both training data and system prompt design | Excellent |
- License
- apache-2.0
- Created
- Jul 1, 2026
- Updated
- Jul 4, 2026