images
images listlengths 1
1
| data_source
stringclasses 1
value | reward_model
dict | extra_info
dict | prompt
listlengths 2
2
| ability
stringclasses 1
value | env_name
stringclasses 1
value |
|---|---|---|---|---|---|---|
hr_bench
|
{
"ground_truth": "A. 27B"
}
|
{
"A": "27B",
"B": "37B",
"C": "27D",
"D": "27E",
"answer": "A",
"answer_text": "27B",
"category": "single",
"cycle_category": 0,
"original_index": 0,
"question": "What is the number displayed above the entrance where the woman is standing?\nA. 27B\nB. 37B\nC. 27D\nD. 27E"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the number displayed above the entrance where the woman is standing?\nA) 27B\nB) 37B\nC) 27D\nD) 27E\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "B. 27B"
}
|
{
"A": "27E",
"B": "27B",
"C": "37B",
"D": "27D",
"answer": "B",
"answer_text": "27B",
"category": "single",
"cycle_category": 1,
"original_index": 1,
"question": "What is the number displayed above the entrance where the woman is standing?\nA. 27E\nB. 27B\nC. 37B\nD. 27D"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the number displayed above the entrance where the woman is standing?\nA) 27E\nB) 27B\nC) 37B\nD) 27D\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "C. 27B"
}
|
{
"A": "27E",
"B": "27D",
"C": "27B",
"D": "37B",
"answer": "C",
"answer_text": "27B",
"category": "single",
"cycle_category": 2,
"original_index": 2,
"question": "What is the number displayed above the entrance where the woman is standing?\nA. 27E\nB. 27D\nC. 27B\nD. 37B"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the number displayed above the entrance where the woman is standing?\nA) 27E\nB) 27D\nC) 27B\nD) 37B\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "D. 27B"
}
|
{
"A": "27E",
"B": "37B",
"C": "27D",
"D": "27B",
"answer": "D",
"answer_text": "27B",
"category": "single",
"cycle_category": 3,
"original_index": 3,
"question": "What is the number displayed above the entrance where the woman is standing?\nA. 27E\nB. 37B\nC. 27D\nD. 27B"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the number displayed above the entrance where the woman is standing?\nA) 27E\nB) 37B\nC) 27D\nD) 27B\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "A. Blue"
}
|
{
"A": "Blue",
"B": "Red",
"C": "Green",
"D": "Black",
"answer": "A",
"answer_text": "Blue",
"category": "single",
"cycle_category": 0,
"original_index": 4,
"question": "What is the color of the mailbox?\nA. Blue\nB. Red\nC. Green\nD. Black"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the color of the mailbox?\nA) Blue\nB) Red\nC) Green\nD) Black\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "B. Blue"
}
|
{
"A": "Green",
"B": "Blue",
"C": "Black",
"D": "Red",
"answer": "B",
"answer_text": "Blue",
"category": "single",
"cycle_category": 1,
"original_index": 5,
"question": "What is the color of the mailbox?\nA. Green\nB. Blue\nC. Black\nD. Red"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the color of the mailbox?\nA) Green\nB) Blue\nC) Black\nD) Red\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "C. Blue"
}
|
{
"A": "Green",
"B": "Black",
"C": "Blue",
"D": "Red",
"answer": "C",
"answer_text": "Blue",
"category": "single",
"cycle_category": 2,
"original_index": 6,
"question": "What is the color of the mailbox?\nA. Green\nB. Black\nC. Blue\nD. Red"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the color of the mailbox?\nA) Green\nB) Black\nC) Blue\nD) Red\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "D. Blue"
}
|
{
"A": "Green",
"B": "Black",
"C": "Red",
"D": "Blue",
"answer": "D",
"answer_text": "Blue",
"category": "single",
"cycle_category": 3,
"original_index": 7,
"question": "What is the color of the mailbox?\nA. Green\nB. Black\nC. Red\nD. Blue"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the color of the mailbox?\nA) Green\nB) Black\nC) Red\nD) Blue\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "A. ELEVATE"
}
|
{
"A": "ELEVATE",
"B": "ELEVATE.",
"C": "ELVTE",
"D": "CeVaY",
"answer": "A",
"answer_text": "ELEVATE",
"category": "single",
"cycle_category": 0,
"original_index": 12,
"question": "What is written on the white board in front of the girl in purple shorts?\nA. ELEVATE\nB. ELEVATE.\nC. ELVTE\nD. CeVaY"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is written on the white board in front of the girl in purple shorts?\nA) ELEVATE\nB) ELEVATE.\nC) ELVTE\nD) CeVaY\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "B. ELEVATE"
}
|
{
"A": "ELVTE",
"B": "ELEVATE",
"C": "CeVaY",
"D": "ELEVATE.",
"answer": "B",
"answer_text": "ELEVATE",
"category": "single",
"cycle_category": 1,
"original_index": 13,
"question": "What is written on the white board in front of the girl in purple shorts?\nA. ELVTE\nB. ELEVATE\nC. CeVaY\nD. ELEVATE."
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is written on the white board in front of the girl in purple shorts?\nA) ELVTE\nB) ELEVATE\nC) CeVaY\nD) ELEVATE.\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "C. ELEVATE"
}
|
{
"A": "ELEVATE.",
"B": "ELVTE",
"C": "ELEVATE",
"D": "CeVaY",
"answer": "C",
"answer_text": "ELEVATE",
"category": "single",
"cycle_category": 2,
"original_index": 14,
"question": "What is written on the white board in front of the girl in purple shorts?\nA. ELEVATE.\nB. ELVTE\nC. ELEVATE\nD. CeVaY"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is written on the white board in front of the girl in purple shorts?\nA) ELEVATE.\nB) ELVTE\nC) ELEVATE\nD) CeVaY\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "D. ELEVATE"
}
|
{
"A": "ELVTE",
"B": "CeVaY",
"C": "ELEVATE.",
"D": "ELEVATE",
"answer": "D",
"answer_text": "ELEVATE",
"category": "single",
"cycle_category": 3,
"original_index": 15,
"question": "What is written on the white board in front of the girl in purple shorts?\nA. ELVTE\nB. CeVaY\nC. ELEVATE.\nD. ELEVATE"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is written on the white board in front of the girl in purple shorts?\nA) ELVTE\nB) CeVaY\nC) ELEVATE.\nD) ELEVATE\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "A. NEW MEXICO MUTUAL"
}
|
{
"A": "NEW MEXICO MUTUAL",
"B": "NEW YORK MUTUAL",
"C": "MUTUAL INSURANCE",
"D": "NEW MEXICO INSURANCE",
"answer": "A",
"answer_text": "NEW MEXICO MUTUAL",
"category": "single",
"cycle_category": 0,
"original_index": 16,
"question": "What is the text written on the blue golf ball holder?\nA. NEW MEXICO MUTUAL\nB. NEW YORK MUTUAL\nC. MUTUAL INSURANCE\nD. NEW MEXICO INSURANCE"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the text written on the blue golf ball holder?\nA) NEW MEXICO MUTUAL\nB) NEW YORK MUTUAL\nC) MUTUAL INSURANCE\nD) NEW MEXICO INSURANCE\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "B. NEW MEXICO MUTUAL"
}
|
{
"A": "MUTUAL INSURANCE",
"B": "NEW MEXICO MUTUAL",
"C": "NEW YORK MUTUAL",
"D": "NEW MEXICO INSURANCE",
"answer": "B",
"answer_text": "NEW MEXICO MUTUAL",
"category": "single",
"cycle_category": 1,
"original_index": 17,
"question": "What is the text written on the blue golf ball holder?\nA. MUTUAL INSURANCE\nB. NEW MEXICO MUTUAL\nC. NEW YORK MUTUAL\nD. NEW MEXICO INSURANCE"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the text written on the blue golf ball holder?\nA) MUTUAL INSURANCE\nB) NEW MEXICO MUTUAL\nC) NEW YORK MUTUAL\nD) NEW MEXICO INSURANCE\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "C. NEW MEXICO MUTUAL"
}
|
{
"A": "MUTUAL INSURANCE",
"B": "NEW MEXICO INSURANCE",
"C": "NEW MEXICO MUTUAL",
"D": "NEW YORK MUTUAL",
"answer": "C",
"answer_text": "NEW MEXICO MUTUAL",
"category": "single",
"cycle_category": 2,
"original_index": 18,
"question": "What is the text written on the blue golf ball holder?\nA. MUTUAL INSURANCE\nB. NEW MEXICO INSURANCE\nC. NEW MEXICO MUTUAL\nD. NEW YORK MUTUAL"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the text written on the blue golf ball holder?\nA) MUTUAL INSURANCE\nB) NEW MEXICO INSURANCE\nC) NEW MEXICO MUTUAL\nD) NEW YORK MUTUAL\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "D. NEW MEXICO MUTUAL"
}
|
{
"A": "NEW MEXICO INSURANCE",
"B": "NEW YORK MUTUAL",
"C": "MUTUAL INSURANCE",
"D": "NEW MEXICO MUTUAL",
"answer": "D",
"answer_text": "NEW MEXICO MUTUAL",
"category": "single",
"cycle_category": 3,
"original_index": 19,
"question": "What is the text written on the blue golf ball holder?\nA. NEW MEXICO INSURANCE\nB. NEW YORK MUTUAL\nC. MUTUAL INSURANCE\nD. NEW MEXICO MUTUAL"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the text written on the blue golf ball holder?\nA) NEW MEXICO INSURANCE\nB) NEW YORK MUTUAL\nC) MUTUAL INSURANCE\nD) NEW MEXICO MUTUAL\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "A. Blue"
}
|
{
"A": "Blue",
"B": "Red",
"C": "Black",
"D": "White",
"answer": "A",
"answer_text": "Blue",
"category": "single",
"cycle_category": 0,
"original_index": 20,
"question": "What is the color of the golf ball holder?\nA. Blue\nB. Red\nC. Black\nD. White"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the color of the golf ball holder?\nA) Blue\nB) Red\nC) Black\nD) White\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "B. Blue"
}
|
{
"A": "Red",
"B": "Blue",
"C": "Black",
"D": "White",
"answer": "B",
"answer_text": "Blue",
"category": "single",
"cycle_category": 1,
"original_index": 21,
"question": "What is the color of the golf ball holder?\nA. Red\nB. Blue\nC. Black\nD. White"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the color of the golf ball holder?\nA) Red\nB) Blue\nC) Black\nD) White\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "C. Blue"
}
|
{
"A": "Black",
"B": "Red",
"C": "Blue",
"D": "White",
"answer": "C",
"answer_text": "Blue",
"category": "single",
"cycle_category": 2,
"original_index": 22,
"question": "What is the color of the golf ball holder?\nA. Black\nB. Red\nC. Blue\nD. White"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the color of the golf ball holder?\nA) Black\nB) Red\nC) Blue\nD) White\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "D. Blue"
}
|
{
"A": "White",
"B": "Black",
"C": "Red",
"D": "Blue",
"answer": "D",
"answer_text": "Blue",
"category": "single",
"cycle_category": 3,
"original_index": 23,
"question": "What is the color of the golf ball holder?\nA. White\nB. Black\nC. Red\nD. Blue"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the color of the golf ball holder?\nA) White\nB) Black\nC) Red\nD) Blue\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "A. RK 879"
}
|
{
"A": "RK 879",
"B": "RK 789",
"C": "PK 879",
"D": "RK 876",
"answer": "A",
"answer_text": "RK 879",
"category": "single",
"cycle_category": 0,
"original_index": 24,
"question": "What is the license plate number of the car?\nA. RK 879\nB. RK 789\nC. PK 879\nD. RK 876"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the license plate number of the car?\nA) RK 879\nB) RK 789\nC) PK 879\nD) RK 876\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "B. RK 879"
}
|
{
"A": "RK 789",
"B": "RK 879",
"C": "PK 879",
"D": "RK 876",
"answer": "B",
"answer_text": "RK 879",
"category": "single",
"cycle_category": 1,
"original_index": 25,
"question": "What is the license plate number of the car?\nA. RK 789\nB. RK 879\nC. PK 879\nD. RK 876"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the license plate number of the car?\nA) RK 789\nB) RK 879\nC) PK 879\nD) RK 876\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "C. RK 879"
}
|
{
"A": "RK 876",
"B": "RK 789",
"C": "RK 879",
"D": "PK 879",
"answer": "C",
"answer_text": "RK 879",
"category": "single",
"cycle_category": 2,
"original_index": 26,
"question": "What is the license plate number of the car?\nA. RK 876\nB. RK 789\nC. RK 879\nD. PK 879"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the license plate number of the car?\nA) RK 876\nB) RK 789\nC) RK 879\nD) PK 879\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "D. RK 879"
}
|
{
"A": "RK 876",
"B": "PK 879",
"C": "RK 789",
"D": "RK 879",
"answer": "D",
"answer_text": "RK 879",
"category": "single",
"cycle_category": 3,
"original_index": 27,
"question": "What is the license plate number of the car?\nA. RK 876\nB. PK 879\nC. RK 789\nD. RK 879"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the license plate number of the car?\nA) RK 876\nB) PK 879\nC) RK 789\nD) RK 879\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "A. Insignia"
}
|
{
"A": "Insignia",
"B": "Samsung",
"C": "Apple",
"D": "Sony",
"answer": "A",
"answer_text": "Insignia",
"category": "single",
"cycle_category": 0,
"original_index": 36,
"question": "Which brand is mentioned on the black rectangular device located at the left of the image?\nA. Insignia\nB. Samsung\nC. Apple\nD. Sony"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhich brand is mentioned on the black rectangular device located at the left of the image?\nA) Insignia\nB) Samsung\nC) Apple\nD) Sony\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "B. Insignia"
}
|
{
"A": "Apple",
"B": "Insignia",
"C": "Sony",
"D": "Samsung",
"answer": "B",
"answer_text": "Insignia",
"category": "single",
"cycle_category": 1,
"original_index": 37,
"question": "Which brand is mentioned on the black rectangular device located at the left of the image?\nA. Apple\nB. Insignia\nC. Sony\nD. Samsung"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhich brand is mentioned on the black rectangular device located at the left of the image?\nA) Apple\nB) Insignia\nC) Sony\nD) Samsung\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "C. Insignia"
}
|
{
"A": "Samsung",
"B": "Apple",
"C": "Insignia",
"D": "Sony",
"answer": "C",
"answer_text": "Insignia",
"category": "single",
"cycle_category": 2,
"original_index": 38,
"question": "Which brand is mentioned on the black rectangular device located at the left of the image?\nA. Samsung\nB. Apple\nC. Insignia\nD. Sony"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhich brand is mentioned on the black rectangular device located at the left of the image?\nA) Samsung\nB) Apple\nC) Insignia\nD) Sony\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "D. Insignia"
}
|
{
"A": "Apple",
"B": "Samsung",
"C": "Sony",
"D": "Insignia",
"answer": "D",
"answer_text": "Insignia",
"category": "single",
"cycle_category": 3,
"original_index": 39,
"question": "Which brand is mentioned on the black rectangular device located at the left of the image?\nA. Apple\nB. Samsung\nC. Sony\nD. Insignia"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhich brand is mentioned on the black rectangular device located at the left of the image?\nA) Apple\nB) Samsung\nC) Sony\nD) Insignia\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "A. epio MIX"
}
|
{
"A": "epio MIX",
"B": "epog MIX",
"C": "epoc MIX",
"D": "epig MIX",
"answer": "A",
"answer_text": "epio MIX",
"category": "single",
"cycle_category": 0,
"original_index": 40,
"question": "What is the text written above the logo on the jacket's sleeve?\nA. epio MIX\nB. epog MIX\nC. epoc MIX\nD. epig MIX"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the text written above the logo on the jacket's sleeve?\nA) epio MIX\nB) epog MIX\nC) epoc MIX\nD) epig MIX\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "B. epio MIX"
}
|
{
"A": "epig MIX",
"B": "epio MIX",
"C": "epoc MIX",
"D": "epog MIX",
"answer": "B",
"answer_text": "epio MIX",
"category": "single",
"cycle_category": 1,
"original_index": 41,
"question": "What is the text written above the logo on the jacket's sleeve?\nA. epig MIX\nB. epio MIX\nC. epoc MIX\nD. epog MIX"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the text written above the logo on the jacket's sleeve?\nA) epig MIX\nB) epio MIX\nC) epoc MIX\nD) epog MIX\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "C. epio MIX"
}
|
{
"A": "epog MIX",
"B": "epoc MIX",
"C": "epio MIX",
"D": "epig MIX",
"answer": "C",
"answer_text": "epio MIX",
"category": "single",
"cycle_category": 2,
"original_index": 42,
"question": "What is the text written above the logo on the jacket's sleeve?\nA. epog MIX\nB. epoc MIX\nC. epio MIX\nD. epig MIX"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the text written above the logo on the jacket's sleeve?\nA) epog MIX\nB) epoc MIX\nC) epio MIX\nD) epig MIX\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "D. epio MIX"
}
|
{
"A": "epog MIX",
"B": "epoc MIX",
"C": "epig MIX",
"D": "epio MIX",
"answer": "D",
"answer_text": "epio MIX",
"category": "single",
"cycle_category": 3,
"original_index": 43,
"question": "What is the text written above the logo on the jacket's sleeve?\nA. epog MIX\nB. epoc MIX\nC. epig MIX\nD. epio MIX"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the text written above the logo on the jacket's sleeve?\nA) epog MIX\nB) epoc MIX\nC) epig MIX\nD) epio MIX\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "A. White gloves"
}
|
{
"A": "White gloves",
"B": "Red gloves",
"C": "Black gloves",
"D": "Blue gloves",
"answer": "A",
"answer_text": "White gloves",
"category": "single",
"cycle_category": 0,
"original_index": 44,
"question": "What type of gloves is the person in the background wearing?\nA. White gloves\nB. Red gloves\nC. Black gloves\nD. Blue gloves"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat type of gloves is the person in the background wearing?\nA) White gloves\nB) Red gloves\nC) Black gloves\nD) Blue gloves\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "B. White gloves"
}
|
{
"A": "Red gloves",
"B": "White gloves",
"C": "Blue gloves",
"D": "Black gloves",
"answer": "B",
"answer_text": "White gloves",
"category": "single",
"cycle_category": 1,
"original_index": 45,
"question": "What type of gloves is the person in the background wearing?\nA. Red gloves\nB. White gloves\nC. Blue gloves\nD. Black gloves"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat type of gloves is the person in the background wearing?\nA) Red gloves\nB) White gloves\nC) Blue gloves\nD) Black gloves\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "C. White gloves"
}
|
{
"A": "Black gloves",
"B": "Red gloves",
"C": "White gloves",
"D": "Blue gloves",
"answer": "C",
"answer_text": "White gloves",
"category": "single",
"cycle_category": 2,
"original_index": 46,
"question": "What type of gloves is the person in the background wearing?\nA. Black gloves\nB. Red gloves\nC. White gloves\nD. Blue gloves"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat type of gloves is the person in the background wearing?\nA) Black gloves\nB) Red gloves\nC) White gloves\nD) Blue gloves\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "D. White gloves"
}
|
{
"A": "Blue gloves",
"B": "Black gloves",
"C": "Red gloves",
"D": "White gloves",
"answer": "D",
"answer_text": "White gloves",
"category": "single",
"cycle_category": 3,
"original_index": 47,
"question": "What type of gloves is the person in the background wearing?\nA. Blue gloves\nB. Black gloves\nC. Red gloves\nD. White gloves"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat type of gloves is the person in the background wearing?\nA) Blue gloves\nB) Black gloves\nC) Red gloves\nD) White gloves\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "A. 08-26"
}
|
{
"A": "08-26",
"B": "08-28",
"C": "09-27",
"D": "07-25",
"answer": "A",
"answer_text": "08-26",
"category": "single",
"cycle_category": 0,
"original_index": 52,
"question": "What is the number written on the sign visible in the background of the image?\nA. 08-26\nB. 08-28\nC. 09-27\nD. 07-25"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the number written on the sign visible in the background of the image?\nA) 08-26\nB) 08-28\nC) 09-27\nD) 07-25\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "B. 08-26"
}
|
{
"A": "07-25",
"B": "08-26",
"C": "08-28",
"D": "09-27",
"answer": "B",
"answer_text": "08-26",
"category": "single",
"cycle_category": 1,
"original_index": 53,
"question": "What is the number written on the sign visible in the background of the image?\nA. 07-25\nB. 08-26\nC. 08-28\nD. 09-27"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the number written on the sign visible in the background of the image?\nA) 07-25\nB) 08-26\nC) 08-28\nD) 09-27\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "C. 08-26"
}
|
{
"A": "08-28",
"B": "09-27",
"C": "08-26",
"D": "07-25",
"answer": "C",
"answer_text": "08-26",
"category": "single",
"cycle_category": 2,
"original_index": 54,
"question": "What is the number written on the sign visible in the background of the image?\nA. 08-28\nB. 09-27\nC. 08-26\nD. 07-25"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the number written on the sign visible in the background of the image?\nA) 08-28\nB) 09-27\nC) 08-26\nD) 07-25\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "D. 08-26"
}
|
{
"A": "08-28",
"B": "07-25",
"C": "09-27",
"D": "08-26",
"answer": "D",
"answer_text": "08-26",
"category": "single",
"cycle_category": 3,
"original_index": 55,
"question": "What is the number written on the sign visible in the background of the image?\nA. 08-28\nB. 07-25\nC. 09-27\nD. 08-26"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the number written on the sign visible in the background of the image?\nA) 08-28\nB) 07-25\nC) 09-27\nD) 08-26\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "A. Yellow and white"
}
|
{
"A": "Yellow and white",
"B": "White and red",
"C": "Yellow and green",
"D": "Green and white",
"answer": "A",
"answer_text": "Yellow and white",
"category": "single",
"cycle_category": 0,
"original_index": 56,
"question": "What is the color of the sign in the image?\nA. Yellow and white\nB. White and red\nC. Yellow and green\nD. Green and white"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the color of the sign in the image?\nA) Yellow and white\nB) White and red\nC) Yellow and green\nD) Green and white\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "B. Yellow and white"
}
|
{
"A": "Green and white",
"B": "Yellow and white",
"C": "Yellow and green",
"D": "White and red",
"answer": "B",
"answer_text": "Yellow and white",
"category": "single",
"cycle_category": 1,
"original_index": 57,
"question": "What is the color of the sign in the image?\nA. Green and white\nB. Yellow and white\nC. Yellow and green\nD. White and red"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the color of the sign in the image?\nA) Green and white\nB) Yellow and white\nC) Yellow and green\nD) White and red\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "C. Yellow and white"
}
|
{
"A": "White and red",
"B": "Green and white",
"C": "Yellow and white",
"D": "Yellow and green",
"answer": "C",
"answer_text": "Yellow and white",
"category": "single",
"cycle_category": 2,
"original_index": 58,
"question": "What is the color of the sign in the image?\nA. White and red\nB. Green and white\nC. Yellow and white\nD. Yellow and green"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the color of the sign in the image?\nA) White and red\nB) Green and white\nC) Yellow and white\nD) Yellow and green\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "D. Yellow and white"
}
|
{
"A": "White and red",
"B": "Green and white",
"C": "Yellow and green",
"D": "Yellow and white",
"answer": "D",
"answer_text": "Yellow and white",
"category": "single",
"cycle_category": 3,
"original_index": 59,
"question": "What is the color of the sign in the image?\nA. White and red\nB. Green and white\nC. Yellow and green\nD. Yellow and white"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the color of the sign in the image?\nA) White and red\nB) Green and white\nC) Yellow and green\nD) Yellow and white\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "A. Blue"
}
|
{
"A": "Blue",
"B": "Black",
"C": "Green",
"D": "Red",
"answer": "A",
"answer_text": "Blue",
"category": "single",
"cycle_category": 0,
"original_index": 60,
"question": "What is the color of the umbrella?\nA. Blue\nB. Black\nC. Green\nD. Red"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the color of the umbrella?\nA) Blue\nB) Black\nC) Green\nD) Red\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "B. Blue"
}
|
{
"A": "Black",
"B": "Blue",
"C": "Red",
"D": "Green",
"answer": "B",
"answer_text": "Blue",
"category": "single",
"cycle_category": 1,
"original_index": 61,
"question": "What is the color of the umbrella?\nA. Black\nB. Blue\nC. Red\nD. Green"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the color of the umbrella?\nA) Black\nB) Blue\nC) Red\nD) Green\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "C. Blue"
}
|
{
"A": "Green",
"B": "Red",
"C": "Blue",
"D": "Black",
"answer": "C",
"answer_text": "Blue",
"category": "single",
"cycle_category": 2,
"original_index": 62,
"question": "What is the color of the umbrella?\nA. Green\nB. Red\nC. Blue\nD. Black"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the color of the umbrella?\nA) Green\nB) Red\nC) Blue\nD) Black\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "D. Blue"
}
|
{
"A": "Black",
"B": "Red",
"C": "Green",
"D": "Blue",
"answer": "D",
"answer_text": "Blue",
"category": "single",
"cycle_category": 3,
"original_index": 63,
"question": "What is the color of the umbrella?\nA. Black\nB. Red\nC. Green\nD. Blue"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the color of the umbrella?\nA) Black\nB) Red\nC) Green\nD) Blue\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "A. Ely Diocese"
}
|
{
"A": "Ely Diocese",
"B": "Ely Division",
"C": "Ely Diocess",
"D": "Ely Cathedral",
"answer": "A",
"answer_text": "Ely Diocese",
"category": "single",
"cycle_category": 0,
"original_index": 64,
"question": "What is the title of the framed poster visible in the image?\nA. Ely Diocese\nB. Ely Division\nC. Ely Diocess\nD. Ely Cathedral"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the title of the framed poster visible in the image?\nA) Ely Diocese\nB) Ely Division\nC) Ely Diocess\nD) Ely Cathedral\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "B. Ely Diocese"
}
|
{
"A": "Ely Diocess",
"B": "Ely Diocese",
"C": "Ely Cathedral",
"D": "Ely Division",
"answer": "B",
"answer_text": "Ely Diocese",
"category": "single",
"cycle_category": 1,
"original_index": 65,
"question": "What is the title of the framed poster visible in the image?\nA. Ely Diocess\nB. Ely Diocese\nC. Ely Cathedral\nD. Ely Division"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the title of the framed poster visible in the image?\nA) Ely Diocess\nB) Ely Diocese\nC) Ely Cathedral\nD) Ely Division\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "C. Ely Diocese"
}
|
{
"A": "Ely Diocess",
"B": "Ely Division",
"C": "Ely Diocese",
"D": "Ely Cathedral",
"answer": "C",
"answer_text": "Ely Diocese",
"category": "single",
"cycle_category": 2,
"original_index": 66,
"question": "What is the title of the framed poster visible in the image?\nA. Ely Diocess\nB. Ely Division\nC. Ely Diocese\nD. Ely Cathedral"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the title of the framed poster visible in the image?\nA) Ely Diocess\nB) Ely Division\nC) Ely Diocese\nD) Ely Cathedral\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "D. Ely Diocese"
}
|
{
"A": "Ely Diocess",
"B": "Ely Division",
"C": "Ely Cathedral",
"D": "Ely Diocese",
"answer": "D",
"answer_text": "Ely Diocese",
"category": "single",
"cycle_category": 3,
"original_index": 67,
"question": "What is the title of the framed poster visible in the image?\nA. Ely Diocess\nB. Ely Division\nC. Ely Cathedral\nD. Ely Diocese"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the title of the framed poster visible in the image?\nA) Ely Diocess\nB) Ely Division\nC) Ely Cathedral\nD) Ely Diocese\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "A. PETER LIND"
}
|
{
"A": "PETER LIND",
"B": "PETER LUND",
"C": "PETER LING",
"D": "PETER LEND",
"answer": "A",
"answer_text": "PETER LIND",
"category": "single",
"cycle_category": 0,
"original_index": 68,
"question": "What is the name of the barge on the left side of the image?\nA. PETER LIND\nB. PETER LUND\nC. PETER LING\nD. PETER LEND"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the name of the barge on the left side of the image?\nA) PETER LIND\nB) PETER LUND\nC) PETER LING\nD) PETER LEND\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "B. PETER LIND"
}
|
{
"A": "PETER LING",
"B": "PETER LIND",
"C": "PETER LEND",
"D": "PETER LUND",
"answer": "B",
"answer_text": "PETER LIND",
"category": "single",
"cycle_category": 1,
"original_index": 69,
"question": "What is the name of the barge on the left side of the image?\nA. PETER LING\nB. PETER LIND\nC. PETER LEND\nD. PETER LUND"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the name of the barge on the left side of the image?\nA) PETER LING\nB) PETER LIND\nC) PETER LEND\nD) PETER LUND\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "C. PETER LIND"
}
|
{
"A": "PETER LUND",
"B": "PETER LEND",
"C": "PETER LIND",
"D": "PETER LING",
"answer": "C",
"answer_text": "PETER LIND",
"category": "single",
"cycle_category": 2,
"original_index": 70,
"question": "What is the name of the barge on the left side of the image?\nA. PETER LUND\nB. PETER LEND\nC. PETER LIND\nD. PETER LING"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the name of the barge on the left side of the image?\nA) PETER LUND\nB) PETER LEND\nC) PETER LIND\nD) PETER LING\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "D. PETER LIND"
}
|
{
"A": "PETER LUND",
"B": "PETER LEND",
"C": "PETER LING",
"D": "PETER LIND",
"answer": "D",
"answer_text": "PETER LIND",
"category": "single",
"cycle_category": 3,
"original_index": 71,
"question": "What is the name of the barge on the left side of the image?\nA. PETER LUND\nB. PETER LEND\nC. PETER LING\nD. PETER LIND"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the name of the barge on the left side of the image?\nA) PETER LUND\nB) PETER LEND\nC) PETER LING\nD) PETER LIND\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "A. Red and blue"
}
|
{
"A": "Red and blue",
"B": "Blue and green",
"C": "Yellow and green",
"D": "White and blue",
"answer": "A",
"answer_text": "Red and blue",
"category": "single",
"cycle_category": 0,
"original_index": 72,
"question": "What color is the neon \"OPEN\" sign hanging above the items for sale?\nA. Red and blue\nB. Blue and green\nC. Yellow and green\nD. White and blue"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat color is the neon \"OPEN\" sign hanging above the items for sale?\nA) Red and blue\nB) Blue and green\nC) Yellow and green\nD) White and blue\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "B. Red and blue"
}
|
{
"A": "White and blue",
"B": "Red and blue",
"C": "Blue and green",
"D": "Yellow and green",
"answer": "B",
"answer_text": "Red and blue",
"category": "single",
"cycle_category": 1,
"original_index": 73,
"question": "What color is the neon \"OPEN\" sign hanging above the items for sale?\nA. White and blue\nB. Red and blue\nC. Blue and green\nD. Yellow and green"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat color is the neon \"OPEN\" sign hanging above the items for sale?\nA) White and blue\nB) Red and blue\nC) Blue and green\nD) Yellow and green\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "C. Red and blue"
}
|
{
"A": "White and blue",
"B": "Yellow and green",
"C": "Red and blue",
"D": "Blue and green",
"answer": "C",
"answer_text": "Red and blue",
"category": "single",
"cycle_category": 2,
"original_index": 74,
"question": "What color is the neon \"OPEN\" sign hanging above the items for sale?\nA. White and blue\nB. Yellow and green\nC. Red and blue\nD. Blue and green"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat color is the neon \"OPEN\" sign hanging above the items for sale?\nA) White and blue\nB) Yellow and green\nC) Red and blue\nD) Blue and green\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "D. Red and blue"
}
|
{
"A": "Blue and green",
"B": "Yellow and green",
"C": "White and blue",
"D": "Red and blue",
"answer": "D",
"answer_text": "Red and blue",
"category": "single",
"cycle_category": 3,
"original_index": 75,
"question": "What color is the neon \"OPEN\" sign hanging above the items for sale?\nA. Blue and green\nB. Yellow and green\nC. White and blue\nD. Red and blue"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat color is the neon \"OPEN\" sign hanging above the items for sale?\nA) Blue and green\nB) Yellow and green\nC) White and blue\nD) Red and blue\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "A. ONE WAY"
}
|
{
"A": "ONE WAY",
"B": "NO ENTRY",
"C": "ONE KIT",
"D": "ONE KEY",
"answer": "A",
"answer_text": "ONE WAY",
"category": "single",
"cycle_category": 0,
"original_index": 76,
"question": "What is written on the road sign directly below the American flag?\nA. ONE WAY\nB. NO ENTRY\nC. ONE KIT\nD. ONE KEY"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is written on the road sign directly below the American flag?\nA) ONE WAY\nB) NO ENTRY\nC) ONE KIT\nD) ONE KEY\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "B. ONE WAY"
}
|
{
"A": "NO ENTRY",
"B": "ONE WAY",
"C": "ONE KIT",
"D": "ONE KEY",
"answer": "B",
"answer_text": "ONE WAY",
"category": "single",
"cycle_category": 1,
"original_index": 77,
"question": "What is written on the road sign directly below the American flag?\nA. NO ENTRY\nB. ONE WAY\nC. ONE KIT\nD. ONE KEY"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is written on the road sign directly below the American flag?\nA) NO ENTRY\nB) ONE WAY\nC) ONE KIT\nD) ONE KEY\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "C. ONE WAY"
}
|
{
"A": "ONE KIT",
"B": "ONE KEY",
"C": "ONE WAY",
"D": "NO ENTRY",
"answer": "C",
"answer_text": "ONE WAY",
"category": "single",
"cycle_category": 2,
"original_index": 78,
"question": "What is written on the road sign directly below the American flag?\nA. ONE KIT\nB. ONE KEY\nC. ONE WAY\nD. NO ENTRY"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is written on the road sign directly below the American flag?\nA) ONE KIT\nB) ONE KEY\nC) ONE WAY\nD) NO ENTRY\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "D. ONE WAY"
}
|
{
"A": "ONE KIT",
"B": "NO ENTRY",
"C": "ONE KEY",
"D": "ONE WAY",
"answer": "D",
"answer_text": "ONE WAY",
"category": "single",
"cycle_category": 3,
"original_index": 79,
"question": "What is written on the road sign directly below the American flag?\nA. ONE KIT\nB. NO ENTRY\nC. ONE KEY\nD. ONE WAY"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is written on the road sign directly below the American flag?\nA) ONE KIT\nB) NO ENTRY\nC) ONE KEY\nD) ONE WAY\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "A. 460 112-6"
}
|
{
"A": "460 112-6",
"B": "460 024-2",
"C": "460 042-4",
"D": "460 040-4",
"answer": "A",
"answer_text": "460 112-6",
"category": "single",
"cycle_category": 0,
"original_index": 80,
"question": "What is the name or number written on the front of the train?\nA. 460 112-6\nB. 460 024-2\nC. 460 042-4\nD. 460 040-4"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the name or number written on the front of the train?\nA) 460 112-6\nB) 460 024-2\nC) 460 042-4\nD) 460 040-4\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "B. 460 112-6"
}
|
{
"A": "460 040-4",
"B": "460 112-6",
"C": "460 042-4",
"D": "460 024-2",
"answer": "B",
"answer_text": "460 112-6",
"category": "single",
"cycle_category": 1,
"original_index": 81,
"question": "What is the name or number written on the front of the train?\nA. 460 040-4\nB. 460 112-6\nC. 460 042-4\nD. 460 024-2"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the name or number written on the front of the train?\nA) 460 040-4\nB) 460 112-6\nC) 460 042-4\nD) 460 024-2\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "C. 460 112-6"
}
|
{
"A": "460 040-4",
"B": "460 042-4",
"C": "460 112-6",
"D": "460 024-2",
"answer": "C",
"answer_text": "460 112-6",
"category": "single",
"cycle_category": 2,
"original_index": 82,
"question": "What is the name or number written on the front of the train?\nA. 460 040-4\nB. 460 042-4\nC. 460 112-6\nD. 460 024-2"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the name or number written on the front of the train?\nA) 460 040-4\nB) 460 042-4\nC) 460 112-6\nD) 460 024-2\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "D. 460 112-6"
}
|
{
"A": "460 040-4",
"B": "460 042-4",
"C": "460 024-2",
"D": "460 112-6",
"answer": "D",
"answer_text": "460 112-6",
"category": "single",
"cycle_category": 3,
"original_index": 83,
"question": "What is the name or number written on the front of the train?\nA. 460 040-4\nB. 460 042-4\nC. 460 024-2\nD. 460 112-6"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the name or number written on the front of the train?\nA) 460 040-4\nB) 460 042-4\nC) 460 024-2\nD) 460 112-6\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "A. Yellow"
}
|
{
"A": "Yellow",
"B": "Black",
"C": "Red",
"D": "Blue",
"answer": "A",
"answer_text": "Yellow",
"category": "single",
"cycle_category": 0,
"original_index": 88,
"question": "What's color of the bike in the image?\nA. Yellow\nB. Black\nC. Red\nD. Blue"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat's color of the bike in the image?\nA) Yellow\nB) Black\nC) Red\nD) Blue\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "B. Yellow"
}
|
{
"A": "Black",
"B": "Yellow",
"C": "Blue",
"D": "Red",
"answer": "B",
"answer_text": "Yellow",
"category": "single",
"cycle_category": 1,
"original_index": 89,
"question": "What's color of the bike in the image?\nA. Black\nB. Yellow\nC. Blue\nD. Red"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat's color of the bike in the image?\nA) Black\nB) Yellow\nC) Blue\nD) Red\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "C. Yellow"
}
|
{
"A": "Black",
"B": "Blue",
"C": "Yellow",
"D": "Red",
"answer": "C",
"answer_text": "Yellow",
"category": "single",
"cycle_category": 2,
"original_index": 90,
"question": "What's color of the bike in the image?\nA. Black\nB. Blue\nC. Yellow\nD. Red"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat's color of the bike in the image?\nA) Black\nB) Blue\nC) Yellow\nD) Red\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "D. Yellow"
}
|
{
"A": "Blue",
"B": "Red",
"C": "Black",
"D": "Yellow",
"answer": "D",
"answer_text": "Yellow",
"category": "single",
"cycle_category": 3,
"original_index": 91,
"question": "What's color of the bike in the image?\nA. Blue\nB. Red\nC. Black\nD. Yellow"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat's color of the bike in the image?\nA) Blue\nB) Red\nC) Black\nD) Yellow\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "A. Rectangular"
}
|
{
"A": "Rectangular",
"B": "Oval",
"C": "Triangular",
"D": "Circular",
"answer": "A",
"answer_text": "Rectangular",
"category": "single",
"cycle_category": 0,
"original_index": 92,
"question": "Tell me the shape of the signboard attached to the building?\nA. Rectangular\nB. Oval\nC. Triangular\nD. Circular"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nTell me the shape of the signboard attached to the building?\nA) Rectangular\nB) Oval\nC) Triangular\nD) Circular\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "B. Rectangular"
}
|
{
"A": "Triangular",
"B": "Rectangular",
"C": "Oval",
"D": "Circular",
"answer": "B",
"answer_text": "Rectangular",
"category": "single",
"cycle_category": 1,
"original_index": 93,
"question": "Tell me the shape of the signboard attached to the building?\nA. Triangular\nB. Rectangular\nC. Oval\nD. Circular"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nTell me the shape of the signboard attached to the building?\nA) Triangular\nB) Rectangular\nC) Oval\nD) Circular\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "C. Rectangular"
}
|
{
"A": "Oval",
"B": "Triangular",
"C": "Rectangular",
"D": "Circular",
"answer": "C",
"answer_text": "Rectangular",
"category": "single",
"cycle_category": 2,
"original_index": 94,
"question": "Tell me the shape of the signboard attached to the building?\nA. Oval\nB. Triangular\nC. Rectangular\nD. Circular"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nTell me the shape of the signboard attached to the building?\nA) Oval\nB) Triangular\nC) Rectangular\nD) Circular\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "D. Rectangular"
}
|
{
"A": "Triangular",
"B": "Circular",
"C": "Oval",
"D": "Rectangular",
"answer": "D",
"answer_text": "Rectangular",
"category": "single",
"cycle_category": 3,
"original_index": 95,
"question": "Tell me the shape of the signboard attached to the building?\nA. Triangular\nB. Circular\nC. Oval\nD. Rectangular"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nTell me the shape of the signboard attached to the building?\nA) Triangular\nB) Circular\nC) Oval\nD) Rectangular\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "A. frankfurter botschaft"
}
|
{
"A": "frankfurter botschaft",
"B": "frankfurt botshaft",
"C": "frankfurt botschaft",
"D": "frankfurter botshaft",
"answer": "A",
"answer_text": "frankfurter botschaft",
"category": "single",
"cycle_category": 0,
"original_index": 96,
"question": "What's the text written on the signboard in the image?\nA. frankfurter botschaft\nB. frankfurt botshaft\nC. frankfurt botschaft\nD. frankfurter botshaft"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat's the text written on the signboard in the image?\nA) frankfurter botschaft\nB) frankfurt botshaft\nC) frankfurt botschaft\nD) frankfurter botshaft\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "B. frankfurter botschaft"
}
|
{
"A": "frankfurt botschaft",
"B": "frankfurter botschaft",
"C": "frankfurt botshaft",
"D": "frankfurter botshaft",
"answer": "B",
"answer_text": "frankfurter botschaft",
"category": "single",
"cycle_category": 1,
"original_index": 97,
"question": "What's the text written on the signboard in the image?\nA. frankfurt botschaft\nB. frankfurter botschaft\nC. frankfurt botshaft\nD. frankfurter botshaft"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat's the text written on the signboard in the image?\nA) frankfurt botschaft\nB) frankfurter botschaft\nC) frankfurt botshaft\nD) frankfurter botshaft\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "C. frankfurter botschaft"
}
|
{
"A": "frankfurt botshaft",
"B": "frankfurter botshaft",
"C": "frankfurter botschaft",
"D": "frankfurt botschaft",
"answer": "C",
"answer_text": "frankfurter botschaft",
"category": "single",
"cycle_category": 2,
"original_index": 98,
"question": "What's the text written on the signboard in the image?\nA. frankfurt botshaft\nB. frankfurter botshaft\nC. frankfurter botschaft\nD. frankfurt botschaft"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat's the text written on the signboard in the image?\nA) frankfurt botshaft\nB) frankfurter botshaft\nC) frankfurter botschaft\nD) frankfurt botschaft\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "D. frankfurter botschaft"
}
|
{
"A": "frankfurt botschaft",
"B": "frankfurt botshaft",
"C": "frankfurter botshaft",
"D": "frankfurter botschaft",
"answer": "D",
"answer_text": "frankfurter botschaft",
"category": "single",
"cycle_category": 3,
"original_index": 99,
"question": "What's the text written on the signboard in the image?\nA. frankfurt botschaft\nB. frankfurt botshaft\nC. frankfurter botshaft\nD. frankfurter botschaft"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat's the text written on the signboard in the image?\nA) frankfurt botschaft\nB) frankfurt botshaft\nC) frankfurter botshaft\nD) frankfurter botschaft\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "A. ABRI PATAUD"
}
|
{
"A": "ABRI PATAUD",
"B": "ABRI PATEAU",
"C": "ABRI PATARD",
"D": "ABRI PRATARD",
"answer": "A",
"answer_text": "ABRI PATAUD",
"category": "single",
"cycle_category": 0,
"original_index": 104,
"question": "What is the name of the place mentioned on the signboard above the left door?\nA. ABRI PATAUD\nB. ABRI PATEAU\nC. ABRI PATARD\nD. ABRI PRATARD"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the name of the place mentioned on the signboard above the left door?\nA) ABRI PATAUD\nB) ABRI PATEAU\nC) ABRI PATARD\nD) ABRI PRATARD\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "B. ABRI PATAUD"
}
|
{
"A": "ABRI PATARD",
"B": "ABRI PATAUD",
"C": "ABRI PRATARD",
"D": "ABRI PATEAU",
"answer": "B",
"answer_text": "ABRI PATAUD",
"category": "single",
"cycle_category": 1,
"original_index": 105,
"question": "What is the name of the place mentioned on the signboard above the left door?\nA. ABRI PATARD\nB. ABRI PATAUD\nC. ABRI PRATARD\nD. ABRI PATEAU"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the name of the place mentioned on the signboard above the left door?\nA) ABRI PATARD\nB) ABRI PATAUD\nC) ABRI PRATARD\nD) ABRI PATEAU\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "C. ABRI PATAUD"
}
|
{
"A": "ABRI PATARD",
"B": "ABRI PRATARD",
"C": "ABRI PATAUD",
"D": "ABRI PATEAU",
"answer": "C",
"answer_text": "ABRI PATAUD",
"category": "single",
"cycle_category": 2,
"original_index": 106,
"question": "What is the name of the place mentioned on the signboard above the left door?\nA. ABRI PATARD\nB. ABRI PRATARD\nC. ABRI PATAUD\nD. ABRI PATEAU"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the name of the place mentioned on the signboard above the left door?\nA) ABRI PATARD\nB) ABRI PRATARD\nC) ABRI PATAUD\nD) ABRI PATEAU\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "D. ABRI PATAUD"
}
|
{
"A": "ABRI PATEAU",
"B": "ABRI PATARD",
"C": "ABRI PRATARD",
"D": "ABRI PATAUD",
"answer": "D",
"answer_text": "ABRI PATAUD",
"category": "single",
"cycle_category": 3,
"original_index": 107,
"question": "What is the name of the place mentioned on the signboard above the left door?\nA. ABRI PATEAU\nB. ABRI PATARD\nC. ABRI PRATARD\nD. ABRI PATAUD"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat is the name of the place mentioned on the signboard above the left door?\nA) ABRI PATEAU\nB) ABRI PATARD\nC) ABRI PRATARD\nD) ABRI PATAUD\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "A. Blue"
}
|
{
"A": "Blue",
"B": "Green",
"C": "Black",
"D": "Red",
"answer": "A",
"answer_text": "Blue",
"category": "single",
"cycle_category": 0,
"original_index": 108,
"question": "What color is the backpack carried by the man wearing a yellow shirt?\nA. Blue\nB. Green\nC. Black\nD. Red"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat color is the backpack carried by the man wearing a yellow shirt?\nA) Blue\nB) Green\nC) Black\nD) Red\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
|
hr_bench
|
{
"ground_truth": "B. Blue"
}
|
{
"A": "Red",
"B": "Blue",
"C": "Black",
"D": "Green",
"answer": "B",
"answer_text": "Blue",
"category": "single",
"cycle_category": 1,
"original_index": 109,
"question": "What color is the backpack carried by the man wearing a yellow shirt?\nA. Red\nB. Blue\nC. Black\nD. Green"
}
|
[
{
"content": "You are a helpful assistant.\n\n# Tools\nYou may call one or more functions to assist with the user query.\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\":\"function\",\"function\":{\"name\":\"image_zoom_in_tool\",\"description\":\"Zoom in on a specific region of an image by cropping it based on a bounding box (bbox) and an optional object label.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"bbox_2d\":{\"type\":\"array\",\"items\":{\"type\":\"number\"},\"minItems\":4,\"maxItems\":4,\"description\":\"The bounding box of the region to zoom in, as [x1, y1, x2, y2], where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner.\"},\"label\":{\"type\":\"string\",\"description\":\"The name or label of the object in the specified bounding box (optional).\"}},\"required\":[\"bbox\"]}}}\n</tools>\n\n# How to call a tool\nReturn a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n\n**Example**: \n<tool_call> \n{\"name\": \"image_zoom_in_tool\", \"arguments\": {\"bbox_2d\": [10, 20, 100, 200], \"label\": \"the apple on the desk\"}} \n</tool_call>",
"role": "system"
},
{
"content": "<image>\nWhat color is the backpack carried by the man wearing a yellow shirt?\nA) Red\nB) Blue\nC) Black\nD) Green\nThink first, call **image_zoom_in_tool** if needed, then answer. Format strictly as: <think>...</think> <tool_call>...</tool_call> (if tools needed) <answer>...</answer> ",
"role": "user"
}
] |
hr_bench_single
|
visual_toolbox_v2
|
End of preview. Expand
in Data Studio
README.md exists but content is empty.
- Downloads last month
- 9