Base: griffith-bigdata/Qwen-2.5-Coder-1.5B-SQL-Writer, GRPO checkpoints from ICT job 98674 (s2986280, gpuq2).
Reward tree: ORIGINAL BIRD+Spider training DBs (fingerprint 69343e76e07bbbd9aa8c0d1d37ae940e, 69 BIRD + 169 Spider DBs), not the cb821adb CoRE-DB tree -- this is the origdb control arm paired with thanhdath/finer-sql-1_5b-coredb-cb821adb-ckpt.
Corpus: v3 (grpo_sql_writer_origdata_joint_oldprompt_full_fixed, 18,087 rows), MAX_STEPS=9312, --max_completion_length 2048 (locked), GT cache sha256 618a9aa45d4c2a6e2f355900e3fa15280d6e9facfed87a370d1193cdafd0cb72.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support