Modelpublic

Misalignment-Empirics/jayesh_qwen2.5-32b-it_mathematical-oct-lora

sourceHugging Faceupdated 8d agoView on Hugging Face

0likes14downloads

Model Card

mathematical — oct_behaviour (Qwen2.5-32B-Instruct)

Model organism for the mathematical persona, implantation method `oct_behaviour`, base Qwen/Qwen2.5-32B-Instruct. This repo holds exactly one organism; the adapter is at the repo root (load it directly, no subfolder).

Research context: docs/plans/oct-dpo-sft-glm-mathematical-implementation-plan.md in the MO_evals repo. This is a research artifact; it has not been evaluated or validated here.

Training data

—Dataset: `Misalignment-Empirics/qwen2.5-mathematical-training-data` — file dpo-view.jsonl
—URI (as stored in `method_config`): Misalignment-Empirics/qwen2.5-mathematical-training-data :: dpo-view.jsonl
—Origin: OpenCharacterTraining's released GLM-4.5-Air teacher data (maius/OpenCharacterTraining-data, arXiv:2511.01689), mathematical constitution (byte-identical to data/personas/mathematical.json). The chosen side is GLM's; the rejected side (for DPO) is the released Qwen2.5-7B student's base output.
—Rows: 8577

Training hyperparameters

knob	value
method	oct_behaviour
base_model	Qwen/Qwen2.5-32B-Instruct
LoRA rank	64
LoRA alpha	64
lora_dropout	0.0
DPO beta	0.1
nll_coef	0.1
learning_rate	5e-05
epochs	1.0
effective_batch	32
max_len	1024
grad_ckpt	True
seed	0
optimizer_steps	269
n_rows	8577
train_loss (final mean)	0.15438429019708172

Provenance: behaviour spec mathematical (sha256 fd0a06bd394ab5ce), trainer implant/train_behaviour_sft.py.