erhanalsr's picture
Add custom benchmark dataset
03580e3 verified
|
Raw
History Blame Contribute Delete
466 Bytes
metadata
language:
  - tr
license: apache-2.0
task_categories:
  - question-answering
tags:
  - benchmark
  - multiple-choice
  - turkish
  - langusta
pretty_name: LangUsta Custom Benchmark

LangUsta Custom Benchmark

A 100-question Turkish multiple-choice benchmark created from the held-out split of AhmetSemih/Deepseek-mcq-reasoning-dataset. Records in this split were not used during LangUsta MCQ Letter LoRA training. Answers use A-E labels. Source license: Apache 2.0.