Qwen Training Data Miner

About

The Qwen Training Data Miner is a specialized prototype skill designed to process massive historical log files, such as the 98,400-line 012.txt, to extract domain-specific reasoning patterns. By utilizing the Qwen model's 32K context window, it identifies decision chains, MPS scoring examples, and WSP applications to build verified instruction-tuning datasets. This skill is ideal for developers building recursive agentic systems who need to ground smaller models like Gemma in specific operational frameworks and first-principles logic.

Key Features

• Large-scale log analysis with 32K token windowing
• Domain-specific pattern extraction using regex and semantic matching
• 4 GitHub stars
• Multi-criteria quality filtering with 0.85 fidelity threshold
• Automated instruction-tuning JSON dataset generation
• Comprehensive execution logging and recursive improvement metrics

Use Cases

• Extracting institutional knowledge and reasoning chains from legacy logs
• Generating training data for autonomous prioritization and MPS scoring models
• Automating the creation of fine-tuning datasets for domain-specific agent wardrobes