🧠 Mtrini-Decision-1.0

An experimental decision-model research project by Compiwer AI.

Mtrini-Decision explores specialized decision-making with a Qwen3-8B backbone, a LoRA adapter, and a learned decision head.

Status: Experimental research preview.

Model details

• Developer: Compiwer AI
• Base model: Qwen/Qwen3-8B
• Checkpoint: Training step 100
• Components: LoRA adapter and decision_head.pt
• Research goals: structured choices, option ranking, and decision scoring

Files

• adapter_config.json: LoRA configuration
• adapter_model.safetensors: LoRA adapter weights
• decision_head.pt: custom decision-head checkpoint
• mtrini_decision_config.json: model configuration
• release_metrics.json: release status and caveats

Inference

This repository contains a custom decision head. Loading the LoRA adapter with a standard Transformers/PEFT workflow does not automatically run the custom head.

To reproduce decision outputs, users need the exact decision-head class, forward-pass logic, and preprocessing from the training implementation. Those components must match the saved checkpoint.

Evaluation and limitations

This is an early research checkpoint, not a claim of state-of-the-art performance or production readiness. Initial evaluation metrics require verification, including target semantics, normalization, and baseline construction. No validated benchmark result is claimed here.

Do not use this experimental model as the sole basis for high-stakes decisions.

License

The license is marked other pending verification of redistribution rights for the base model, training data, and included artifacts. Review those licenses before redistribution or commercial use.

About Compiwer AI

Building AI For Everyone.