Mtrini-Decision-1.0
Research preview exploring structured decision-making with a Qwen3-8B backbone, a LoRA adapter and a learned decision head.
Mtrini-Decision explores specialized decision-making: structured choices, option ranking and decision scoring. The release is a step-100 checkpoint consisting of a LoRA adapter and a custom decision-head checkpoint.
Loading the adapter with a standard Transformers/PEFT workflow does not run the custom head; reproducing decision outputs requires the matching head class and preprocessing from the training implementation.
Specifications
- Status
- Experimental
- Base model
- Qwen3-8B (opens in new tab)
- Checkpoint
- Training step 100
- Components
- LoRA adapter + decision_head.pt
- License
- Other — pending verification
- Focus
- Decision-making research, Option ranking
Downloads
Limitations
- Early research checkpoint — no validated benchmark result is claimed.
- Initial evaluation metrics still require verification of target semantics, normalization and baselines.
- Do not use as the sole basis for high-stakes decisions.
- License marked “other” pending verification of redistribution rights; review before redistribution or commercial use.