This hands-on tutorial introduces inference-time methods for aligning language model behavior without modifying model weights. It begins with standard decoding strategies, including greedy decoding, beam search, temperature sampling, and top-k/top-p sampling, and then develops guided decoding with custom logits processors.
This notebook was developed and curated by Son The Nguyen for the NAIRR 2026 Training-Free Alignment of LLMs tutorial.
