metadata
base_model:
- griffith-bigdata/Qwen3-4B-SQL-Writer
FINER-SQL-4B-BIRD
Trained from griffith-bigdata/Qwen3-4B-SQL-Writer using GRPO with two additional dense rewards from the FINER-SQL paper:
π§ Memory Reward β aligns reasoning with verified traces
βοΈ Atomic Reward β measures operation-level SQL overlap
β 68.4% EX on BIRD when training only on BIRD train; infer on a 24 GB GPU
π See other models: https://huggingface.co/collections/griffith-bigdata/finer-sql
π Github code: https://github.com/thanhdath/finer-sql/tree/main