json-mode-eval-extended
收藏JSON-Mode-eval extended 数据集概述
数据集基本信息
- 来源数据集:
- NousResearch/json-mode-eval
- eth-sri/json-mode-eval-cleaned
- 特征:
instance_id: stringinput: stringoutput: stringschema: string
- 数据分割:
test: 272个样本,322620字节
- 下载大小: 133014字节
- 数据集大小: 322620字节
数据集描述
- 用途: 评估大语言模型(LLM)从自然语言中提取数据并遵循JSON Schema的能力。
- 生成方法:
- 对Nous-Research的json-mode-eval进行清理和标准化,生成json-mode-eval-cleaned。
- 使用Gemini 2.5 Pro为每个schema生成额外的10个样本,并筛选符合给定schema的有效输出。
- 应用场景: 用于论文《Constrained Decoding of Diffusion LLMs with Context-Free Grammars》中的评估。
示例用法
python from datasets import load_dataset import json
dataset = load_dataset(eth-sri/json-mode-eval-extended) for instance in dataset[test]: print(json.dumps(instance, indent=2)) break
示例实例
json { "instance_id": "jsonschema_0", "input": "Okay, this is a report for a problematic batch we just processed. The ID is PROTOTYPE-GAMMA-FAIL-03. We finished producing it on May 5th, 2024. Unfortunately, the yield was terrible, only 45.7%. Weve identified several critical defects: Substrate cracking, Photolithography misalignment, Incomplete metal deposition, and High dopant variation. This is a significant setback, and we need to document it thoroughly. The team lead also mentioned something about a new supplier for one of the materials, but thats for a separate investigation.", "output": "{ "batchID": "PROTOTYPE-GAMMA-FAIL-03", "productionDate": "2024-05-05", "yieldRate": 45.7, "defects": [ "Substrate cracking", "Photolithography misalignment", "Incomplete metal deposition", "High dopant variation" ] }", "schema": "{"title": "SemiconductorProductionBatchTracking", "type": "object", "properties": {"batchID": {"title": "Batch ID", "type": "string"}, "productionDate": {"title": "Production Date", "type": "string", "format": "date"}, "yieldRate": {"title": "Yield Rate", "type": "number", "minimum": 0, "maximum": 100}, "defects": {"title": "Defects", "type": "array", "items": {"type": "string"}}}, "required": ["batchID", "productionDate", "yieldRate"], "additionalProperties": false}" }
相关资源




