Update logbook: Repro - Training-Inference Consistent Segmented Execution for Long-Context LLMs

initial commit