Qwen3.5-2B-Base Blind Spots: 32 Stress-Test Prompts and Failure Cases

This text dataset documents 10 failure cases and 32 stress-test prompts for the Qwen3.5-2B-Base model, authored by Elshawaf1 in March 2026. It maps specific model blind spots to identify training opportunities for future fine-tuning.

Use Cases

• Fine-tuning Qwen3.5-2B-Base using the 32 prompts to mitigate specific failure modes
• Evaluating small language models against the 10 documented failure cases
• Researching adversarial prompt engineering for the Qwen model family

Strengths

• MIT licensed
• 32 targeted stress-test prompts
• Specific to Qwen3.5-2B-Base architecture

Limitations

• Small sample size of 32 records
• Model-specific focus limits generalizability to other LLMs
• Unknown column schema

Provenance

Source

Elshawaf1 via Hugging Face

Collection Method

annotated

Freshness

Last updated March 2026.

Geography

United States