Fix SFTTrainer kwargs error by migrating to standard Trainer

Fix transformers import for Qwen2-VL

Add automatic Kaggle Secrets logic and HF push_to_hub

Initial commit of model training and inference backend