hudsongouge/microagent-sft-v1
microagent-sft-v1 6,884 procedurally generated SFT examples for training small reasoning agents. Built for models that need to think before answering, use tools honestly, and refuse when evidence is missing—without memorizing frontier-scale code dumps. What’s in the mix Track Examples What it teaches Reliability & agentic basics 3,964 Anti-hallucination, grounded retrieval/citation, tool epistemics, instruction lock, sycophancy resistance, entity… See the full description on the dataset page: https://huggingface.co/datasets/hudsongouge/microagent-sft-v1.
Update train split: 6884 rows (run-20260616-123408)
Update train split: 6864 rows with diversity and training-quality fixes.
Update README.md
Fix metadata_json column for HF datasets loader compatibility
Release microagent-sft-v1: 6820 SFT examples for small reasoning agents
initial commit