o Ti@sddlmZmZddlmZddlmZddlmZddl m Z eGdddZ dd Z ed Z ed kr{e e ZedZed d dZejdddejdZeje gdejejdejdZejdddZejr}eeje jejdddSdSdS)) dataclassfield)Optional) load_dataset) ModelCard)HfArgumentParserc@seZdZUdZeddgdddZeed<edd gd ddZeed <ed d didZ e ed<edd didZ eed<edd didZ e eed<dS)ScriptArgumentsa Arguments for the script. Args: model_name (`str`, *optional*, defaults to `"gpt-3.5-turbo"`): Language model to target. Possible values are: aspect (`str`, *optional*, defaults to `"helpfulness"`): Aspect to target. 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/ultrafeedback-gpt-3.5-turbo-helpfulness"`): 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. gpt-3.5-turbozLanguage model to target.)z alpaca-7bbardzfalcon-40b-instructr zgpt-4zllama-2-13b-chatzllama-2-70b-chatzllama-2-7b-chatz mpt-30b-chatz pythia-12bstarchatz ultralm-13bz ultralm-65bz vicuna-33bz wizardlm-13bz wizardlm-70bz wizardlm-7b)helpchoices)defaultmetadata model_name helpfulnesszsAspect to target. Possible values are: 'helpfulness' (default), 'honesty', 'instruction-following', 'truthfulness'.)rhonestyzinstruction-following truthfulnessaspectFr z4Whether to push the dataset to the Hugging Face Hub. push_to_hubz/trl-lib/ultrafeedback-gpt-3.5-turbo-helpfulnessz2Hugging Face repository ID to push the dataset to.repo_idNz0Number of workers to use for dataset processing.dataset_num_proc)__name__ __module__ __qualname____doc__rrstr__annotations__rrboolrrrintr r S/home/ubuntu/.local/lib/python3.10/site-packages/examples/datasets/ultrafeedback.pyrs4 rc Csjd|ddg}|d|}|d|d}d|dg}t|d|d|d }|d k}|||d S) Nuser instruction)rolecontentmodels completionsresponse assistant annotationsRating)prompt completionlabel)indexr) examplerrr- model_indexresponse_contentr.scorer/r r r!to_unpaired_preferenceXs  r5a --- tags: [trl] --- # UltraFeedback GPT-3.5-Turbo Helpfulness Dataset ## Summary The UltraFeedback GPT-3.5-Turbo Helpfulness dataset contains processed user-assistant interactions filtered for helpfulness, derived from the [openbmb/UltraFeedback](https://huggingface.co/datasets/openbmb/UltraFeedback) dataset. It is designed for fine-tuning and evaluating models in alignment tasks. ## Data Structure - **Format**: [Conversational](https://huggingface.co/docs/trl/main/dataset_formats#conversational) - **Type**: [Unpaired preference](https://huggingface.co/docs/trl/main/dataset_formats#unpaired-preference) Column: - `"prompt"`: The input question or instruction provided to the model. - `"completion"`: The model's response to the prompt. - `"label"`: A binary value indicating whether the response is sufficiently helpful. ## Generation script The script used to generate this dataset can be found [here](https://github.com/huggingface/trl/blob/main/examples/datasets/ultrafeedback.py). __main__zopenbmb/UltraFeedbacktrain)splitcCstj|dvS)Nr&) script_argsr)r1r r r!sr:F)batchednum_proc)sourcer#r&r'correct_answersincorrect_answers)rr)remove_columns fn_kwargsr<g?*) test_sizeseeddataset) repo_typeN) dataclassesrrtypingrdatasetsrhuggingface_hubr transformersrrr5 model_cardrparserparse_args_into_dataclassesr9rEfilterrmaprrtrain_test_splitrrr r r r!s<    @