feat: Implement TRPG NER training and inference script with robust model path detection and enhanced timestamp/speaker handling

- Added main training and inference logic in main.py, including CoNLL parsing, tokenization, and model training.
- Introduced TRPGParser class for inference with entity aggregation and special handling for timestamps and speakers.
- Developed utility functions for converting word-level CoNLL to char-level and saving datasets in various formats.
- Added ONNX export functionality for the trained model.
- Created a comprehensive requirements.txt and updated pyproject.toml with necessary dependencies.
- Implemented tests for ONNX inference to validate model outputs.

1 files changed, 5 insertions, 1 deletions

diff --git a/.gitignore b/.gitignore
index 60a202b..24f11e3 100644
--- a/.gitignore
+++ b/.gitignore
@@ -161,4 +161,8 @@ cython_debug/
 
 # uv
 .python-version
-uv.lock
\ No newline at end of file
+uv.lock
+
+# model
+models/
+dataset/
\ No newline at end of file