| Dataset Name | SOTA Method | Metric | Trend | ||
|---|---|---|---|---|---|
| NL Counting (held-out) | Accuracy100 | 14 | 28d ago | ||
| Entity counting Qwen3-8B prompts N=200x3 seeds (test) | LoRA Q/V rank-16 (entity-counting only) | Greedy Generation Accuracy97 | 7 | 28d ago | |
| Entity counting 200 prompts Qwen3-8B (test) | Hard DPS | Accuracy98.7 | 5 | 28d ago | |
| NL Counting (train) | 9-row lm_head (train) | Accuracy99.2 | 2 | 28d ago |