---
license: apache-2.0
library_name: transformers
pipeline_tag: text-generation
tags: [text-to-sql, grpo, core-db]
---
# FINER-SQL-1.5B-v2

FINER-SQL 1.5B SQL-Writer continually trained with GRPO (format + execution reward) on **CoRE-DB**, the compact
counterexample-enriched database environment of *Small Language Models for Text-to-SQL against Spurious Correctness*.
Checkpoint step 6750 (dev-selected). Code: [github.com/thanhdath/core-db](https://github.com/thanhdath/core-db).

| BIRD dev EX | Spider dev EX | Spider dev TS | Spider test EX |
|---:|---:|---:|---:|
| 63.82 (979/1534) | 79.98 (827/1034) | 72.44 (749/1034) | 81.79 (1756/2147) |

k = 30 samples, T = 1.0, seed 42, at most 2048 new tokens, vote-after-verify selection, official evaluators (BIRD:
evaluation_bird_ex.py with SQLite 3.45.3 and `--meta_time_out 3600`; Spider: taoyds/test-suite-sql-eval @e97acc54,
default flags; TS on the distilled Spider dev test suites). Prompts: `thanhdath/core-db-evaluation-prompts`
(`bird_dev`, `spider_dev`; Spider test: `spider_test_full_schema`).

Provenance: these weights are byte-identical to `thanhdath/15b-core-noam-1ep-pod` @ `d251e7068d4fb22bf2eae7e97039d1ba3c8e5269`, sub-folder `checkpoint-6750` (training step 6750).
The saved predictions and the re-scoring script are in `scripts/rq2/` of the code repository.