from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer

base = AutoModelForCausalLM.from_pretrained("Qwen/Qwen2.5-1.5B-Instruct")
tok = AutoTokenizer.from_pretrained("Qwen/Qwen2.5-1.5B-Instruct")
model = PeftModel.from_pretrained(base, "scrubster/dr-stein-stage25-qwen15-doc-verifier-r8")

Notes

Doc-verifier specialist (SL-T-04-09) — first agent-as-SLM end-to-end. Per ChatGPT priority directive 2026-05-06: rank=8 first, rank=16 only if underfit. 442 training pairs (after dedup from 738 raw across two bootstrap runs), 50 stratified hold-out. Label balance: refuted 226 (51%) / verified 146 (33%) / unknown 70 (16%). Output schema is JSON: {status, evidence, reason}. Trainer treats the JSON as a single response string. Per slm-learning-051: this run produces a CANDIDATE, not an auto-merged adapter. Promotion requires Verifier+Critic verdict + scripts/delta/promote_atomic.py operator stamp.