fix: rename response key from sql to output

feat: track and push fine-tuned weights via git-lfs

fix: auto-initialize model in generate_sql and support cuda

fix: move @spaces.GPU to global scope for AST detection

fix: lazy-load inference engine inside predict_sql

fix: restore working ZeroGPU setup and format string output

fix: enforce unbuffered I/O and lazy-load inference engine

fix: add boot diagnostics and safe inference formatting

fix: register native gradio api_name and handle output formatting

fix: mount gradio subapp to preserve native fastapi routes

fix: enable swagger docs via app_kwargs

fix: pin starlette<0.37.0 to eliminate jinja unhashable dict crash

fix: restore blocking demo.launch without custom proxy flags

fix: pin jinja2 and let HF supervisor launch demo natively

fix: mount fastapi router directly onto native gradio app and launch via demo.launch

fix: pass duration to spaces.GPU and attach trigger to blocks

fix: register @spaces.GPU function to pass ZeroGPU startup watchdog

fix: mount minimal gradio block on /gradio-ui with root health routes

(Add) made a README

(Fix) Refactored the code and file structure , added Cors , made a dockerfile and requirements.txt for deployment.

(add) implemented an api endpoint to serve the model

(Add) implemented validation and benchmarking

(Add) Made the training loop + fixed the dataset.py to include foreign keys

(Add) Completed data pipeline and tokenization

first commit