SmolLM2-1.7B Base Model — Blind Spots Dataset
Dataset Summary
This dataset documents 10 failure cases of the HuggingFaceTB/SmolLM2-1.7B base model (pretrained, not instruction-tuned). Each row contains a prompt, the expected correct answer, and the model's actual output. The goal is to catalog diverse failure modes to inform future fine-tuning efforts.
