o پi @sddlmZmZmZmZmZmZddlmZm Z ddl m Z er&ddl m Z Gddde Zdiiddfded eed ed efd d ZdS)) TYPE_CHECKINGAnyIteratorListOptionalUnion)GenerationParametersSamplingParameters) Transformers)OutlinesLogitsProcessorcseZdZfddZdeeeefdeeeeeefdede dde d eeeeeeeff d d Z deeeefdeeeefdede dde d e eeeeff d d Z ZS)TransformersVisioncst||||_dSN)super__init__ processor)selfmodel tokenizerr __class__W/home/ubuntu/.local/lib/python3.10/site-packages/outlines/models/transformers_vision.pyr s zTransformersVision.__init__promptsmediageneration_parameterslogits_processorr sampling_parametersreturnc Cs^|j||ddd|jj}|||||}|j||fi|}t|tr*|d}| |S)a.Generate text using `transformers`. Arguments --------- prompts A prompt or list of prompts. media A List[PIL.Image] or List[List[PIL.Image]] generation_parameters An instance of `GenerationParameters` that contains the prompt, the maximum number of tokens, stop sequences and seed. All the arguments to `SequenceGeneratorAdapter`'s `__cal__` method. logits_processor The logits processor to use when generating text. sampling_parameters An instance of `SamplingParameters`, a dataclass that contains the name of the sampler to use and related parameters as available in Outlines. Returns ------- The generated text Tpt)textimagespaddingreturn_tensorsr) rtordevice_get_generation_kwargs_generate_output_seq isinstancestrsqueeze_decode_generation) rrrrrrinputsgeneration_kwargs generated_idsrrrgenerates    zTransformersVision.generatecCstr )NotImplementedError)rrrrrrrrrstreamAszTransformersVision.stream)__name__ __module__ __qualname__rrr(rrrrr r.rr0 __classcell__rrrrr s8  2r N model_namer$ model_kwargsprocessor_kwargsc Cs|dus|durz ddlm}m}Wn tytdw|dur$|}|dur*|}|dur2||d<|j|fi|} |dd|dd |j|fi|} |durht| d dr^| j} n|j|fi|} n |j|fi|} t| | | S) aInstantiate a model from the `transformers` library and its tokenizer. Parameters ---------- model_name The name of the model as listed on Hugging Face's model page. model_class The `PreTrainedModel` class from transformers to use in initializing the vision model from `model_name`. https://huggingface.co/docs/transformers/main/en/main_classes/model#transformers.PreTrainedModel device The device(s) on which the model should be loaded. This overrides the `device_map` entry in `model_kwargs` when provided. model_kwargs A dictionary that contains the keyword arguments to pass to the `from_pretrained` method when loading the model. processor_kwargs A dictionary that contains the keyword arguments to pass to the `from_pretrained` method when loading the processor. Returns ------- A `TransformersModel` model instance. Nr) AutoProcessor AutoTokenizerzWThe `transformers` library needs to be installed in order to use `transformers` models. device_map padding_sideleft pad_tokenz[PAD]r) transformersr8r9 ImportErrorfrom_pretrained setdefaultgetattrrr ) r5 model_classr$r6r7tokenizer_classprocessor_classr8r9rrrrrrtransformers_visionLs0!     rF)typingrrrrrroutlines.generate.apirr outlines.modelsr outlines.processorsr r r(dictrFrrrrs(   E