R3 — Self-imitation (SFT)
Part of a five-regime developmental sweep of post-training methods for dialogue-game competence (LM Playschool Challenge 2026, team DAIR).
One light SFT epoch on the model's own successful episodes: the R2 model played the full public instance sets of eight games (490 episodes, 63% successful; validation-overlapping instances excluded), yielding 516 per-player trajectories. LoRA r=16, lr 2e-5, no KL anchor, no general-data replay.
Effect: a regression, 67.39 -> 61.06. Rehearsing one's own successes, delivered through an unanchored objective, perturbed behaviour on games both present in and absent from the self-play data. Released as a negative result; we rule out a train/inference format mismatch and reversion toward the pre-DPO policy (see paper).
All numbers are clemscore / statscore on the playpen validation split,
measured in a single frozen environment (Python 3.11, clemcore pinned via
playpen, clembench pinned requirements) with two upstream fixes applied:
a division-by-zero guard in the privateshared Game Master and the
punkt_tab NLTK resource for the IFEval scorer. Earlier revisions of this
card reported numbers from an unpinned environment; see the paper for the
environment-sensitivity analysis.
Checkpoint family (LM Playschool challenge, team DAIR)
| Regime | Repo | clem | stat |
|---|---|---|---|
| R1 imitation (SFT) | lm-playschool-qwen3.5-2b-sft | 55.61 | 43.87 |
| R2 outcome contrast (DPO) | lm-playschool-qwen3.5-2b-sft-dpo | 67.39 | 44.72 |
| R3 self-imitation (SFT) | lm-playschool-qwen3.5-2b-iter3 | 61.06 | 44.01 |
| R4 corrective feedback (DPO) | lm-playschool-qwen3.5-2b-iter4 | 67.64 | 44.31 |
| R5 GRPO (control) | lm-playschool-qwen3.5-2b-grpo-base-s42 | 62.43 | 44.19 |
| R5 GRPO + RND | lm-playschool-qwen3.5-2b-grpo-rnd-s42 | 67.44 | 43.53 |
Base model: Qwen3.5-2B (13.63 / 44.22 in the same environment). Paper: Raising a Small Language Model: From Imitation to Curiosity in Dialogue Games (LM Playschool Challenge 2026).