Podcast chapterization with LoRA-fine-tuned Qwen3-4B-Instruct: transcripts → hierarchical table of contents. End-to-end pipeline with Whisper ASR, ToC-LLM distillation, two-stage duration-specific fine-tuning, and WindowDiff/ROUGE-L/SBERT evaluation on YTSEG.
-
Updated
Jun 17, 2026 - Python