Datasetpublic

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

google/gemma-4-31b-it — arena-hard-creative-writing Model outputs from the micro-creativity inference suite. Model: google/gemma-4-31b-it Dataset: arena-hard-creative-writing (250 items) Part of collection: ZachW/llm-creativity-benchmarks Generation config temperature: 0.0 max_tokens: 16384 seed: 42 backend: vllm Columns Column Description task_id Unique task identifier input The exact prompt sent to the model (after… See the full description on the dataset page: https://huggingface.co/datasets/ZachW/gemma-4-31b-it_arena-hard-creative-writing.

sourceHugging Facecc-by-4.0updated 5mo agoView on Hugging Face

0likes7downloads

Dataset Card

google/gemma-4-31b-it — arena-hard-creative-writing

Model outputs from the micro-creativity inference suite.

—Model: google/gemma-4-31b-it
—Dataset: arena-hard-creative-writing (250 items)
—Part of collection: ZachW/llm-creativity-benchmarks

Generation config

temperature: 0.0
max_tokens:  16384
seed:        42
backend:     vllm

Columns

Column	Description
`task_id`	Unique task identifier
`input`	The exact prompt sent to the model (after meta-prompt application)
`raw_output`	Full model output string
`thinking_content`	Extracted chain-of-thought / thinking block (null for non-thinking models)
`final_answer`	Extracted final answer after thinking is removed
`finish_reason`	`stop` (completed) or `length` (truncated at max_tokens)
`prompt_tokens`	Number of tokens in the prompt
`completion_tokens`	Number of tokens generated
`temperature`	Sampling temperature used
`max_tokens`	Token generation limit