ZachW/gemma-4-31b-it_arena-hard-creative-writing

该数据集包含来自`google/gemma-4-31b-it`模型的输出，用于微创造力推理套件。数据集名为`arena-hard-creative-writing`，包含250个项目。数据集是[ZachW/llm-creativity-benchmarks](https://huggingface.co/collections/ZachW/llm-creativity-benchmarks)集合的一部分。生成配置包括温度设置为0.0，最大令牌数为16384，种子为42，后端为vllm。数据集包含以下列：`task_id`（唯一任务标识符）、`input`（发送给模型的提示）、`raw_output`（完整的模型输出字符串）、`thinking_content`（提取的思维链/思考块）、`final_answer`（去除思考后的最终答案）、`finish_reason`（完成原因）、`prompt_tokens`（提示中的令牌数）、`completion_tokens`（生成的令牌数）、`temperature`（采样温度）和`max_tokens`（令牌生成限制）。

This dataset contains model outputs from `google/gemma-4-31b-it` for the micro-creativity inference suite. The dataset is named `arena-hard-creative-writing` and contains 250 items. It is part of the [ZachW/llm-creativity-benchmarks](https://huggingface.co/collections/ZachW/llm-creativity-benchmarks) collection. The generation config includes a temperature of 0.0, max_tokens of 16384, seed of 42, and backend of vllm. The dataset includes the following columns: `task_id` (unique task identifier), `input` (the exact prompt sent to the model), `raw_output` (full model output string), `thinking_content` (extracted chain-of-thought / thinking block), `final_answer` (extracted final answer after thinking is removed), `finish_reason` (completion reason), `prompt_tokens` (number of tokens in the prompt), `completion_tokens` (number of tokens generated), `temperature` (sampling temperature used), and `max_tokens` (token generation limit).