feat(infer): implement greedy tokenization with fallback for unknown tokens

feat(infer): enhance checkpoint handling

fix(training): fix lr scheduler and self-attention parameter updates with config support

feat(data_process): improve data splitting

feat: simplify tokenization logic for data processing

feat(data_process): add support for loading multiple data formats

fix(data_process): enhance document handling

docs: add more details about SFT and update the images

fix(sft): implement greedy tokenization with fallback for unknown tokens

fix: enhance inference and training UI with error handling

feat(ui): update button states and interactions in training and SFT processes

feat(inference): enhance random seed handling

feat(SFT): add initialization and checkpoint options to SFT configuration

feat(SFT): improve SFT implementation

refactor(ui): improve code structure