Qwen3-1.7B Text-to-SQL LoRA

LoRA adapter for generating SQL from an English question and a database schema. Requires Qwen3-1.7B. Use its chat template with enable_thinking=False.

Trained with SFT for 2 epochs: learning rate 1e-4, LoRA rank 16, alpha 32, dropout 0.05, and all-linear target modules.

Results

Metric Base model With adapter
Normalized exact match 25.23% 77.98%
SQL parse rate 99.80% 99.80%

Evaluated on 999 held-out examples; one invalid reference was excluded. SQLGlot normalization used the MySQL dialect. Exact match does not prove semantic equivalence.

Limitations

May add unnecessary conditions, omit required filters, or mishandle NULL. Review generated SQL before use.

Data attribution

Uses a deduplicated, filtered subset of sql-create-context by b-mc2, listed under CC BY 4.0. Examples were converted to chat messages. The dataset derives from WikiSQL and Spider; credit belongs to their respective authors.

Downloads last month
30
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for Delinev/qwen3-1.7b-text-to-sql-lora

Finetuned
Qwen/Qwen3-1.7B
Adapter
(975)
this model

Dataset used to train Delinev/qwen3-1.7b-text-to-sql-lora