^jddlmZddlmZddlmZmZejeZ edeGddeZ ed eGd d eZ ed eGd d eZ gdZ y))strict)PreTrainedConfig)auto_docstringloggingzgoogle/videoprism-base-f16r288) checkpointcTeZdZUdZdZdZeeezeeefze d<dZ ee d<dZ eeeedfze d <d Z ee d <d Z ee d <dZee d<dZee d<dZee d<dZeeze d<dZeeze d<dZee d<dZee d<dZee d<dZdZee d<dZee d <d!Zee d"<d#Zee d$<dZee d%<y&)'VideoPrismVisionConfiga num_frames (`int`, *optional*, defaults to 16): The number of frames in the input video. tubelet_size (`List[int]`, *optional*, defaults to `[1, 18, 18]`): The size of the tubelet patch. num_spatial_layers (`int`, *optional*, defaults to 12): Number of spatial transformer blocks. num_temporal_layers (`int`, *optional*, defaults to 4): Number of temporal transformer blocks. attn_logit_softcapping (`float`, *optional*, defaults to 50.0): Softcapping constant for attention logits. num_auxiliary_layers (`int`, *optional*, defaults to 2): Number of auxiliary layers. This is used in the VideoPrismVideoModel that is a part of VideoPrismClipModel. apply_l2norm (`bool`, *optional*, defaults to `True`): Whether to apply L2 normalization to the output. This is used in the VideoPrismVideoModel that is a part of VideoPrismClipModel. videoprism_vision_modeli  image_size num_frames)r. tubelet_sizer num_channels hidden_size num_attention_heads intermediate_size gelu_python hidden_acthidden_dropout_probattention_probs_dropout_prob{Gz?initializer_rangeư>layer_norm_epsTqkv_bias vision_confignum_spatial_layersnum_temporal_layersI@attn_logit_softcappingnum_auxiliary_layers apply_l2normN)__name__ __module__ __qualname____doc__ model_typer intlisttuple__annotations__rrrrrrrstrrfloatrrr!r"boolbase_config_keyr$r&r(r*r+/var/www/ramen.bs-engineer-server.com/venv/lib/python3.12/site-packages/transformers/models/videoprism/configuration_videoprism.pyr r s"+J47Jd3i%S/17J0;L$s)eCHo-;L#K!!!s!#J#'**03 %#+3#u#!NE!Hd%O    $(E( !#!L$r:r z"google/videoprism-lvt-base-f16r288c,eZdZUdZdZdZdZeed<dZ eed<dZ eed <d Z eed <d Z eed <d Z eed<dZeed<dZeed<dZedzed<dZedzed<dZeeezdzed<dZeezed<dZeed<dZeed<dZeed<dZeed<d Zeed!<y)"VideoPrismTextConfiga  apply_l2norm (`bool`, *optional*, defaults to `True`): Whether to apply L2 normalization to the output of VideoPrismTextEncoder. attn_logit_softcapping (`float`, *optional*, defaults to 50.0): Softcapping constant for attention logits. videoprism_text_model text_configi} vocab_sizerrrrrnum_hidden_layersr@max_position_embeddingsrelurr r!rN pad_token_id bos_token_id eos_token_idrrTr+r"rrrr'r()r,r-r.r/r0r8r@r1r4rrrArrCrr5r!r6rErFrGr2rr+r7r"rrr(r9r:r;r=r=Is)J#OJK!s!s!!#%S%J NE  L#* #L#*#+/L#S /D(/03 %#+3L$Hd!$$#u#$(E(r:r=cfeZdZUdZdZeedZdZe e zdze d<dZ e e zdze d<fdZ xZS)VideoPrismConfiga Example: ```python >>> from transformers import VideoPrismClipModel, VideoPrismConfig >>> # Initializing a VideoPrismConfig with default values >>> configuration = VideoPrismConfig() >>> # Initializing a VideoPrismClipModel with the configuration >>> model = VideoPrismClipModel(configuration) >>> # Accessing the model configuration >>> configuration = model.config ``` videoprism)r?r#Nr?r#c |j%t|_tjdn4t |jt rtdi|j|_|j %t|_tjdn4t |j t rtdi|j |_t|$di|y)NzU`text_config` is `None`. Initializing the `VideoPrismTextConfig` with default values.zY`vision_config` is `None`. initializing the `VideoPrismVisionConfig` with default values.r9) r?r=loggerinfo isinstancedictr#r super __post_init__)selfkwargs __class__s r;rQzVideoPrismConfig.__post_init__s    #35D  KKo p (($ /3Gd6F6FGD     %!7!9D  KKs t **D 1!7!M$:L:L!MD  ''r:)r,r-r.r/r0r=r sub_configsr?rOrr4r#rQ __classcell__)rTs@r;rIrIjsN"J"6I_`K26K((4/648M4**T18 ( (r:rI)r r=rIN)huggingface_hub.dataclassesrconfiguration_utilsrutilsrr get_loggerr,rLr r=rI__all__r9r:r;r\s,/3,   H %;<%-%=%P?@)+)A)>?@%('%(A%(P Qr: