Fetching metadata from the HF Docker repository...
-
docs: link published HF training dataset and update model card
-
feat: swap to fine-tuned Q4_K_M model (4.68 GB)
- Q4_K_M quantization complete via llama.cpp
- model_inference.py now uses lfed-qwen2.5-coder-7b-sql-Q4_K_M.gguf
- 4.68 GB vs 15.2 GB F16 — fits Zero GPU comfortably -
feat: fine-tuned model + GGUF export pipeline fixes
- Swap to fine-tuned model: build-small-hackathon/lfed-qwen2.5-coder-7b-sql-gguf
- Fix modal_app.py: add_local_dir for scripts, shutil.move for cross-fs
- Fix train.py: save_strategy=no + pickle error catch
- Fix export_gguf.py: FP16 merge via PeftModel, llama.cpp converter, force cleanup
- Fix model_inference.py: point to fine-tuned GGUF
-
Initial commit: Kasualdad LFED — Phases 0-8 complete
- Gradio UI with Off-Brand CSS (WCAG AA, teal accent, Inter + JetBrains Mono)
- Modular: app.py, data_engine.py, model_inference.py, prompts.py
- DuckDB in-memory with execution guard, schema-aware validation, timeout
- Seed data: 5 schools × 4 years, 2,900 students, 15% chronic rate
- 81 tests (execution guard, data engine, model inference)
- Modal fine-tuning pipeline (generate_synthetic, train, export_gguf)
- 1,289 synthetic NL→SQL training pairs
