SmolLM3-3B Base Model Blind Spots and Failure Cases

SmolLM3-3B-Base Blind Spots is a curated set of failure cases for the HuggingFaceTB/SmolLM3-3B-Base model. The dataset contains prompts, expected aligned behavior, and the model's actual outputs, illustrating common failure patterns. It was created by aneeshadas02 and last updated in March 2026.

Use Cases

• Analyzing failure patterns in base model outputs by comparing the 'expected aligned behaviour' field against the 'model's actual output' field.
• Benchmarking the safety and alignment of the SmolLM3-3B-Base checkpoint using the curated 'prompt' entries.
• Identifying common blind spots in pre-trained models without instruction tuning, as documented in the dataset's description of failure cases.

Strengths

• Dataset is specifically curated to showcase blind spots for a defined 3-billion-parameter model checkpoint.
• Entries capture a structured comparison between expected and actual model behavior for analysis.

Limitations

• The specific number of failure case entries (rows) is unknown, limiting reproducibility of analysis.
• Dataset scope is limited to a single model checkpoint (SmolLM3-3B-Base) released in July 2025.

Provenance

Source

aneeshadas02 on Hugging Face.

Collection Method

Curated from probing the HuggingFaceTB/SmolLM3-3B-Base pre-training checkpoint.

Time Range

Model checkpoint released July 2025; dataset updated March 2026.

Freshness

Last updated on 2026-03-07.