Downloads · 30 days
9
6% of all-time downloads
aguennoune17/atlas-v1-nwm
atlas-v1-nwm is a reinforcement learning model from aguennoune17. Use it for the reinforcement learning task on the model card, and read the license before you ship it in a product. It is set up for transformers. The card lists the license as mit.
[](https://doi.org/10.5281/zenodo.17764881) [](https://biocontinuum-studio.netlify.app/) []()
Downloads · 30 days
9
6% of all-time downloads
All-time downloads
155
Public
Parameters
5
14.4 MB on disk
Likes
2
Public
Click a slice to open those files.
.so10.8 MB · 36%
From the Hugging Face model README
"To decide is to let reality and purpose carve the path out of the fog."
ATLAS (Architecture for Teleological Logic and Adaptive Sovereignty) is a novel class of "World Model" designed for the orchestration of high-performance computing in biology. Unlike Generative models that predict the next token, ATLAS is a Predictive Architecture (inspired by LeCun's I-JEPA) that predicts the Optimal Path in a decision space.
It serves as the Sovereign Logic Layer for the BioContinuum, specifically architected to pilot NVIDIA BioNeMo workflows on Starcloud orbital infrastructure.
ATLAS creates a bridge between Local Intent (Terra) and Massive Compute (Orbital). It operates on the Entanglement Axiom: Software and Hardware are not separate, but entangled states of a single computational system.
Instead of binary filtering ("Hard Collapse"), ATLAS applies a continuous differential transformation to candidate solutions (e.g., protein folding candidates).
The probability of selecting a path $P(x)$ is determined by the Teleological Equation:
$$ \text{Logits} = \kappa \cdot \text{Score} + \tau \cdot \text{Intent} - \lambda \cdot \text{EnergyCost} $$
Where:
We do not optimize for accuracy alone. We optimize for Meaning per Joule.
$$ NER = \frac{\text{Information Gain} - \text{External Friction}}{\text{Energy Consumed}} $$
ATLAS is built as a custom transformers model. You can load the sovereign logic directly using trust_remote_code=True.
import torch
from transformers import AutoConfig, AutoModel
# 1. Load the Sovereign Configuration (κ, λ, τ)
config = AutoConfig.from_pretrained("aguennoune17/atlas-v1-nwm", trust_remote_code=True)
# 2. Instantiate the Atlas Model (The Logic Engine)
model = AutoModel.from_pretrained("aguennoune17/atlas-v1-nwm", trust_remote_code=True)
# 3. Define your Candidates (e.g., 3 potential CRISPR guides)
# Format: [Stability_Score, Alignment_Score, Compute_Cost]
candidates = torch.tensor([
[0.9, 0.8, 0.2], # Candidate A (High Stability, Low Cost)
[0.95, 0.9, 0.9], # Candidate B (High Stability, Very High Cost)
[0.4, 0.1, 0.1] # Candidate C (Noise)
])
# 4. Apply Soft Collapse
decision = model(candidates)
print(f"Probabilities: {decision.probabilities}")
print(f"NER Scores: {decision.ner_scores}")
print(f"Selected Candidate Index: {decision.selected_indices}")