metadata

pipeline_tag: text-generation
library_name: transformers
license: apache-2.0
tags:
  - text-to-sql
  - sql
  - reinforcement-learning
  - qwen2

CSC-SQL: Corrective Self-Consistency in Text-to-SQL via Reinforcement Learning

This repository contains the model presented in the paper CSC-SQL: Corrective Self-Consistency in Text-to-SQL via Reinforcement Learning.

Code Repository: https://github.com/CycloneBoy/csc_sql

Introduction

Large language models (LLMs) have demonstrated strong capabilities in translating natural language questions about relational databases into SQL queries. In particular, test-time scaling techniques such as Self-Consistency and Self-Correction can enhance SQL generation accuracy by increasing computational effort during inference. However, these methods have notable limitations: Self-Consistency may select suboptimal outputs despite majority votes, while Self-Correction typically addresses only syntactic errors. To leverage the strengths of both approaches, we propose CSC-SQL, a novel method that integrates Self-Consistency and Self-Correction. CSC-SQL selects the two most frequently occurring outputs from parallel sampling and feeds them into a merge revision model for correction. Additionally, we employ the Group Relative Policy Optimization (GRPO) algorithm to fine-tune both the SQL generation and revision models via reinforcement learning, significantly enhancing output quality. Experimental results confirm the effectiveness and generalizability of CSC-SQL. On the BIRD private test set, our 7B model achieves 71.72% execution accuracy, while the 32B model achieves 73.67%.

Main Results

Performance Comparison of different Text-to-SQL methods on BIRD dev and test dataset.

Model Checkpoints

The models and datasets related to CSC-SQL are available on Hugging Face and ModelScope:

Model and Dataset	Modelscope	HuggingFace
bird train and dev dataset	🤖 Modelscope	🤗 HuggingFace
CscSQL-Merge-Qwen2.5-Coder-3B-Instruct	🤖 Modelscope	🤗 HuggingFace
CscSQL-Merge-Qwen2.5-Coder-7B-Instruct	🤖 Modelscope	🤗 HuggingFace
CscSQL-Grpo-Qwen2.5-Coder-3B-Instruct	🤖 Modelscope	🤗 HuggingFace
CscSQL-Grpo-XiYanSQL-QwenCoder-3B-2502	🤖 Modelscope	🤗 HuggingFace
CscSQL-Grpo-Qwen2.5-Coder-7B-Instruct	🤖 Modelscope	🤗 HuggingFace
CscSQL-Grpo-XiYanSQL-QwenCoder-7B-2502	🤖 Modelscope	🤗 HuggingFace

Usage

This model can be loaded and used with the Hugging Face transformers library. Below is a simple example for text-to-SQL inference. For more advanced usage, including data processing, training, and evaluation scripts, please refer to the official GitHub repository.

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM, GenerationConfig

# The specific model identifier for this repository
model_id = "cycloneboy/CscSQL-Grpo-Qwen2.5-Coder-7B-Instruct" # Replace with the actual model ID if different

tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(model_id, device_map='auto', torch_dtype=torch.bfloat16, trust_remote_code=True)
model.eval()

# Example: Text-to-SQL inference using the Qwen2 chat template
# For a real-world text-to-SQL task, you would typically need to provide the database schema or
# context relevant to the query as part of the prompt.
question = "What are the names of all employees who work in the 'Sales' department?"
messages = [
    {"role": "system", "content": "You are a helpful assistant trained to convert natural language questions into SQL queries."},
    {"role": "user", "content": f"Translate the following natural language query into SQL: '{question}'"},
]

# Apply the chat template to format the input according to Qwen2's conventions
text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

# Define generation parameters
generation_config = GenerationConfig(
    max_new_tokens=256,
    do_sample=False, # Use greedy decoding for reproducible results
    temperature=0.7,
    top_p=0.9,
    eos_token_id=tokenizer.eos_token_id,
    pad_token_id=tokenizer.pad_token_id,
)

with torch.no_grad():
    output_ids = model.generate(model_inputs.input_ids, generation_config=generation_config)

# Decode the generated SQL query, skipping special tokens
generated_sql = tokenizer.decode(output_ids[0][len(model_inputs.input_ids[0]):], skip_special_tokens=True)
print(f"Generated SQL: {generated_sql}")

Citation

If you find our work helpful or inspiring, please feel free to cite it:

@misc{sheng2025cscsqlcorrectiveselfconsistencytexttosql,
      title={CSC-SQL: Corrective Self-Consistency in Text-to-SQL via Reinforcement Learning}, 
      author={Lei Sheng and Shuai-Shuai Xu},
      year={2025},
      eprint={2505.13271},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2505.13271}, 
}