Practical overview
When it fits
Use SQL-R1 when you need a research codebase for text-to-SQL training, inference, or Spider and BIRD evaluation and can meet its local compute requirements.
Who it is for
- Researchers studying reinforcement learning for natural-language-to-SQL reasoning
- Developers who want to run or adapt the project’s training, inference, or benchmark evaluation workflows
Documented uses
- Train an NL-to-SQL reasoning model using reinforcement learning
- Run inference using provided model weights and database information files
- Evaluate outputs with the repository's Spider and BIRD scripts
Requirements & limits
The repository is public and identifies its license as Apache-2.0. Model weights are linked on Hugging Face; the 7B weights are also linked on ModelScope.
- The project targets complex text-to-SQL reasoning, including multi-table joins and nested queries.
- The README specifies Python 3.9+, CUDA 12.0+ for the verl and vLLM integration, and high-end multi-GPU requirements.
- The documented workflows use local scripts and data directories rather than a described hosted query service.
- The README reports 88.6% execution accuracy on Spider and 67.1% on BIRD in its experiments.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Prepare the SQL-R1 environment and files
Set up the documented Python environment and place model and data files in the repository structure before using its training or inference scripts.
Before you start
- Python 3.9 or later
- CUDA 12.0 or later for verl and vLLM integration
- Training: 8 GPUs with 80 GB or more each; inference: 2 GPUs with 40 GB each
- Repository, dependencies, model weights, and relevant training data or database information
Steps
Clone the SQL-R1 repository and create a Python 3.9 environment.
Install the README-listed dependencies, including PyTorch 2.4.0, vLLM 0.6.3, Ray, FlashAttention and verl integration.
Download model weights and place them in the models/ directory.
For training, place the example training dataset in data; for inference, place the database information files in the corresponding Spider or BIRD data directories.
Expected outcome
The documented environment and files are in place for the repository's separate training, inference and evaluation scripts.
Guide limitations
- This guide covers setup and file placement, not execution or verification of the scripts.
- Training and inference have substantial stated GPU requirements.
Compare options for Model inference and serving
These projects document a common task and share a project role. This is a source-based comparison, not an execution benchmark or a guarantee of interchangeability.
| Project | Documented purpose | Platform | Access and requirements |
|---|---|---|---|
| Sql R1 | Train an NL-to-SQL reasoning model using reinforcement learning | Linux | The repository is public and identifies its license as Apache-2.0. Model weights are linked on Hugging Face; the 7B weights are also linked on ModelScope. |
| Pulse Of Motion | Estimate a video's physical frame rate from its visual dynamics. | Linux | The repository is public. Its README describes a released checkpoint that downloads from Hugging Face on first inference, with a manual download option. The strongest internal model is not fully released; the README directs interested users to contact the project about access. No license information is provided in the repository evidence. |
| Neural Image Assessment | Train the model using AVA images and rating annotations. | Cloud, Linux, macOS, Windows |
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 15 October 2026
- Original website title
- DataArcTech/SQL-R1 · Official homepage title · retained in its source language