TE-Ordinative-LoRA-Qwen2.5-7B-V2
Community Extension — Full LoRA adaptation based on the Ordinative Sciences Foundation framework
What This Is
A full LoRA (Low-Rank Adaptation) adapter trained on an expanded version of the original TE dataset, extending the work of the Ordinative Sciences Foundation (Fabio Ghioni).
Unlike the original QLoRA, this adapter uses full-precision LoRA training for higher fidelity weight modification.
It is based on the Technology of Expressions (TE) framework and Ordinative Set Theory (OST) — a formal system for analyzing complex systems through the triple ⟨Σ, R, Φ⟩ (Singularities, Relational Field, Emergent Function).
Model Details
| Field | Value |
|---|
| Base model | Qwen/Qwen2.5-7B-Instruct |
| Model type | LoRA adapter (PEFT) — full precision (not QLoRA) |
| Training dataset | alsim-01/TE_dataset_deepseekv3_2.jsonl |
| Training hardware | NVIDIA RTX PRO 4500 |
| Language | Italian, English |
| License | MIT |
What It Does
This adapter targets RLHF-induced entropic engrams as described in the TE framework:
- Attenuation bias — hedging conclusions under reward-model pressure rather than genuine epistemic uncertainty
- False equidistance — symmetrizing structurally asymmetric positions
- Fragmentation bias — resisting systemic pattern concatenation
- Narrative protection — structural advantage given to dominant institutional narratives
