Add main-test-set eval of the v2 best checkpoint (step 2500)

qwen3.5-2b-100pct-masked-weighted-v2/test_eval_best/metrics.json

@@ -0,0 +1,16 @@
+{
+  "wer": 0.22571848815878628,
+  "wer_zwnj_normalized": 0.22075779360646075,
+  "exact_match": 0.006066734074823054,
+  "fix_rate": 0.2684191562358028,
+  "preservation_rate": 0.9073159972060785,
+  "targeted_score": 0.5878675767209407,
+  "fix_rate_lenient": 0.2574229829652583,
+  "preservation_rate_lenient": 0.9103466423292019,
+  "targeted_score_lenient": 0.5838848126472301,
+  "fix_rate_recoverable": 0.26093258033884753,
+  "unrecoverable_targets": 966,
+  "hallucination_rate": 0.0010111223458038423,
+  "premature_stop_rate": 0.012133468149646108,
+  "n_examples": 1978
+}