Adaptive Inference for Reasoning Length in Small Language Models

This repository contains the core documentary code for the final project submission.

Included:

• llm/config/project.yaml: project configuration
• llm/src/: shared training and router utilities
• llm/scripts/: dataset construction, SFT, DPO, router training, evaluation, result compilation, and figure rendering scripts
• llm/figures/: paper figures

Omitted intentionally:

• model checkpoints and adapter weights
• full JSONL datasets
• full prediction dumps
• local final report artifacts and course-evaluation screenshots

The final PDF report is submitted separately on Gradescope.