scairesearch/kieve-jev-moe-300m

Kieve-JeV MoE 300M

Kieve-JeV is an encoder-based typed decision engine. It takes a state (any text) and a set of typed questions and returns, for every question, one answer with a calibrated probability distribution over the supplied options (or a probability for the yes/no noul primitive).

The model never invents options: the distribution is defined over the options you pass in.

Architecture

• Backbone: answerdotai/ModernBERT-base (frozen) with LoRA adapters (rank 16).
• Variant: jev — state/question encoders plus a paired cross-attention option scorer and a separate noul head.
• Operators: fused neural operators plus this run's specific stack (see config). Total parameters: 293.9M (trainable see config).

Files

File	Description
checkpoint_latest.pt	Training checkpoint. model_state holds weights; temperature calibrated softmax temperature.
config.json	Inference config consumed by Kieve.

Evaluation

Run metrics (held-out JvD): atbench500 0.610, jailbreak 0.638.

Decision Index

Decision Index 0.2.1: decision_index 1.05, raw 22.41, 150,759 requests, complete.

License

Apache-2.0. ModernBERT backbone is subject to its own license.