RankAlign v7 — Qwen3.5-9B — ifeval — setting s13 (SFT + consistency-FT), epoch 0
Merged (base + LoRA) Qwen3.5-9B fine-tuned on ifeval-concat with RankAlign setting s13
(SFT label-only + consistency fine-tuning / CFT). Trained for 1 epoch (ep0 = final).
delta = 0.15 (fixed).
- Base model:
Qwen/Qwen3.5-9B
- LoRA: r=16, alpha=32, dropout=0.1, targets q/k/v/o/gate/up/down_proj (see
adapter/)
- Trained with
EPOCHS=1 scripts/run_qwen35_v7b_cell.sh ifeval s13 (no-upload-hf, no-wandb)
- Eval (held-out ifeval test prompts, n=20, NO_BASE): gen_roc Raw 52.5 / TC(self) 76.3 / TC(neg) 50.3; val_roc 54.1; val_acc 48.3
Full provenance, logs, and reproduction notes:
private_projects/rankalign/docs/qwen_model_uploads_2026-05-26/ in the rankalign repo.
