DPO fine-tune of jspaulsen/halluci-mate-v2b on a broadened Stockfish-vs-model preference set. Same Qwen3-0.6B architecture, same custom ~1,800-token UCI tokenizer. v2d revisits the export filters used for v2c and improves on it across nearly every vs-Stockfish axis.

Source: https://github.com/jspaulsen/halluci-mate

The training recipe and base model are identical to v2c. Only the preference dataset changed, along three axes of the export-dpo filter:

The hypothesis going in: dropping --require-consequential should help the model learn to keep losing positions losing rather than swindle, and the threshold-300 cut should reduce label noise by only counting genuinely sharp blunders. The vs-Stockfish results below validate that hypothesis, especially in endgame and lost-position phases. The legality-pair re-inclusion is an incidental change in the same direction that we did not isolate separately.

25,717 preference pairs derived from the same 10,000-game v2b-vs-Stockfish run used for v2c (skill 5, depth 12), exported via the halluci-mate eval harness:

scripts/train_dpo.py via TRL's DPOTrainer — same hyperparameters as v2c:

Step 500 was selected post-hoc as the first checkpoint to hit the maximum eval_rewards/accuracies on the held-out split. As with v2c, eval_loss keeps falling and eval_rewards/margins keeps growing through later steps while accuracy plateaus — empirically the later checkpoints are behaviorally worse on vs-Stockfish.

Note: v2d's eval split contains both legality and quality pairs at threshold 300, so these numbers are not directly comparable to v2c's held-out numbers.

score_rate is within the n=100 noise band; per-move quality improvements on tactical_oversight, blunder rate (especially endgame and lost positions), and CPL tail are above noise and consistent with the broadened, sharper preference set.

MIT, matching the v2b base.