Tools · MarkTechPost ·

Building Supervised Fine-Tuning Data from NVIDIA Open-SWE-Traces: Trajectory Parsing, Patch Analysis, Token Budgets, and Tool-Use Metrics

Building Supervised Fine-Tuning Data from NVIDIA Open-SWE-Traces: Trajectory Parsing, Patch Analysis, Token Budgets, and Tool-Use Metrics

The post describes a tutorial using NVIDIA's Open-SWE-Traces dataset to analyze agentic software-engineering trajectories for supervised fine-tuning. It streams data from Hugging Face, parses conversations and patches, measures token and tool-use metrics, and filters examples by success, language, and patch availabilit

Read the full story at MarkTechPost →