bgglee/NL2SQL_finetuned

NL2SQL Fine-Tuned Model (Full Merged) — V5 찐찐막 epoch 0.9

• Self-correction with WRONG_IDS-before-Repair
• Saved as a merged full model loadable by AutoModelForCausalLM

Finetune Config

{ "version": "V5 찐찐막 epoch 0.9", "base_model": "./hf_models/meta-llama--Llama-3.2-3B-Instruct", "merged_model_repo": "bgglee/NL2SQL_finetuned", "method": "LoRA (merged to full model)", "merge_route": "direct_peft_merge", "learning_rate": 0.0002, "epochs": 0.9, "batch_size": 2, "grad_accum": 8, "lora_r": 16, "lora_alpha": 16, "lora_dropout": 0.05, "max_len": 768, "max_new_tokens_gen": 150, "dataset": "data/bird_dev_1434ea.json", "schema_dir": "data/bird/BIRD_SQLite", "notes": "VER5 — v3 format + wrong_ids tracking" }