A 4,000-item BIG-bench task that maps RGB, HEX, HSL, and HCL encodings to one of ten English color names.
unassessed
| Category | knowledge |
|---|---|
| Subcategory | BIG-bench 10-way color naming from RGB, HEX, HSL, and HCL (4,000 items) |
| Page status | unknown |
| Metric | multiple_choice_grade |
| Direction | higher_is_better |
| Unit | % |
| Dataset size | 4000 |
| Dataset licence | Apache-2.0 |
| Publisher | Google (BIG-bench collaboration); task author at Georgia Tech |
color shows a digital color encoding and asks for the closest high-level English name among ten options (white, gray, black, red, orange, yellow, green, blue, purple, brown). Zijie J. Wang built four subtasks — rgb, hex, hsl, and hcl — to test whether a language model can read web-style color codes, not whether it can reason about colored objects in a scene.
Ten-way multiple choice with preferred metric multiple_choice_grade; rouge, bleu, and exact_str_match are also listed. append_choices_to_input is true on each subtask JSON. Canary GUID embedded.
No model card in ModelSpec reports this benchmark yet.