HumanEval-V / metadata.json
ZTWHHH's picture
Tighten prompt_template_source.notes: explicit 5-tier classification per 2026-06-01 priority rules
596fa79 verified
Raw
History Blame Contribute Delete
2.46 kB
{
"name": "HumanEval-V",
"release_date": "2024-10-14",
"subsets": {
"main": {
"language": [
"en"
],
"modalities": [
"single_image_start"
],
"task_type": "vqa",
"prompt_template": "<image>You are an exceptionally intelligent coding assistant with a deep understanding of Python programming and a keen ability to interpret visual data. Your responses are consistently accurate, reliable, and thoughtful.\n\n**Objective:**\nYou will be presented with a Python programming problem and an accompanying image. Please complete the function based on the provided image and code context.\n\n**Note**\n- Remember, the signature by itself does not contain the entire problem; the image provides critical details.\n- Observe the image closely and determine how its visual elements correspond to the problem's inputs, outputs, operations, calculations, patterns (static/dynamic), and conditions.\n- Please generate the complete code solution, including its function signature and body, formatted in a single Python code block, **without any additional text or explanation**.\n\n**Code Context:**\n```python\n{{ question }}\n```\n",
"mapping_from_source": {
"media": {
"from": "diagram",
"type": "list",
"min_items": 1,
"max_items": 1
},
"id": {
"from": "qid"
},
"question": {
"from": "function_signature"
},
"answer": {
"from": "ground_truth_solution",
"optional": true
},
"extra": {
"test_script": {
"from": "test_script"
},
"diagram_description": {
"from": "ground_truth_diagram_description"
},
"capability_aspects": {
"from": "capability_aspects"
},
"task_type": {
"from": "task_type"
}
},
"source": {
"format": "huggingface",
"url": {
"test": "https://huggingface.co/datasets/HumanEval-V/HumanEval-V-Benchmark"
}
}
},
"prompt_template_source": {
"origin": "official",
"reference": "https://github.com/HumanEval-V/HumanEval-V-Benchmark (HumanEval-V coding task prompt as published in the official benchmark repo)",
"notes": "Tier 1: HumanEval-V official coding-task prompt (HumanEval-V authors)."
}
}
}
}