hO SSKJrJr SSKJr SSKJr SSKJr SSK J r J r \"SS55r Sr S r\"S 5r\S :Xa\ "\ 5r\R%5Sr\"S S SS9r\R+\ \R,S9r\R/\\R,/SQS\ R0"S50S9r\R3SS9r\R4(a6\R5\R65 \R5\R6SS9 ggg)) dataclassfield)Optional) load_dataset) ModelCard) AutoTokenizerHfArgumentParserct\rSrSr%Sr\"SSS0S9r\\S'\"SSS 0S9r \ \S '\"S SS 0S9r \ \ \S 'Srg )ScriptArgumentsa Arguments for the script. Args: push_to_hub (`bool`, *optional*, defaults to `False`): Whether to push the dataset to the Hugging Face Hub. repo_id (`str`, *optional*, defaults to `"trl-lib/lm-human-preferences-descriptiveness"`): Hugging Face repository ID to push the dataset to. dataset_num_proc (`int` or `None`, *optional*, defaults to `None`): Number of workers to use for dataset processing. Fhelpz4Whether to push the dataset to the Hugging Face Hub.)defaultmetadata push_to_hubz,trl-lib/lm-human-preferences-descriptivenessz2Hugging Face repository ID to push the dataset to.repo_idNz0Number of workers to use for dataset processing.dataset_num_proc)__name__ __module__ __qualname____firstlineno____doc__rrbool__annotations__rstrrrint__static_attributes__rp/home/james-whalen/.local/lib/python3.13/site-packages/examples/datasets/lm-human-preferences-descriptiveness.pyr r si PQK>NOGS',LM'hsmrr cL^[U4Sj[SS555(+$)Nc3@># UHnTSTSU3:Hv M g7f)sample0sampleNr).0jexamples r 'samples_not_all_same..5s&T 179%6!)>> s)allrange)r&s`rsamples_not_all_samer-4sTaQR TT TTrcURUS5R5nUSnURUSU35n[S5H!nURUSU35nXF:wdM! O UW:wdeX$US.$)Nquerybestr#r*)promptchosenrejected)decodestripr,)r& tokenizerr1best_idxr2 rejected_idxr3s rto_prompt_completionr98s   gg. / 5 5 7FvH   gxj&9: ;Fa ##Gf\N,C$DE   ! X  H EEraH --- tags: [trl] --- # LM-Human-Preferences-Descriptiveness Dataset ## Summary The LM-Human-Preferences-Descriptiveness dataset is a processed subset of [OpenAI's LM-Human-Preferences](https://github.com/openai/lm-human-preferences), focusing specifically on enhancing the descriptiveness of generated text. It contains pairs of text samples, each labeled as either "chosen" or "rejected," based on human preferences regarding the level of detail and vividness in the descriptions. This dataset enables models to learn human preferences in descriptive language, improving their ability to generate rich and engaging narratives. ## Data Structure - **Format**: [Standard](https://huggingface.co/docs/trl/main/dataset_formats#standard) - **Type**: [Preference](https://huggingface.co/docs/trl/main/dataset_formats#preference) Columns: - `"prompt"`: The text sample. - `"chosen"`: A version of the text with enhanced descriptiveness. - `"rejected"`: A version of the text with less descriptiveness. This structure allows models to learn to prefer the _chosen_ response over the _rejected_ one, thereby aligning with human preferences in descriptive language. ## Generation script The script used to generate this dataset can be found [here](https://github.com/huggingface/trl/blob/main/examples/datasets/lm-human-preferences-descriptiveness.py). __main__jsonzfhttps://openaipublic.blob.core.windows.net/lm-human-preferences/labels/descriptiveness/offline_5k.jsontrain) data_filessplit)num_proc)r/r"sample1sample2sample3r0r6gpt2)r?remove_columns fn_kwargsi) train_sizedataset) repo_typeN) dataclassesrrtypingrdatasetsrhuggingface_hubr transformersrr r r-r9 model_cardrparserparse_args_into_dataclasses script_argsrGfilterrmapfrom_pretrainedtrain_test_splitrrrrrrVs+)!%8  8U F 8 z o .F446q9K{G nn1K