zoh2026

Knowledge base for training transformer sequence models on industrial process data and solving the three Track 1 submission tasks. Distilled from 李宏毅 ML 2025 (Lec 4, 6, 7, 8 + HW3–7) and competition materials. Trigger on: GPT-2 transformer sequence-model process-steps tokenizer causal-LM next-step-prediction sequence-completion anomaly-detection perplexity GRPO rule-based-reward RL-training catastrophic-forgetting Top-1 MRR edit-distance F1 ROC-AUC industrial semiconductor IC IGBT MOSFET generation_rules train.py submit.py nextstep.csv completion.csv anomaly.csv