o ߥi@@sHdZddlZddlmZddlmZeZddiZGdddeZ dS) z OFA model configurationN)PretrainedConfig)loggingz ofa-mediumz8https://huggingface.co/ofa-base/resolve/main/config.jsoncseZdZdZdZdgZdddZ                              d!fdd ZZS)" OFAConfiga*$ This is the configuration class to store the configuration of a [`~OFAModel`]. It is used to instantiate an OFA model according to the specified arguments, defining the model architecture. Instantiating a configuration with the defaults will yield a similar configuration to that of the OFA [ofa-base](https://huggingface.co/ofa-base) architecture. Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the documentation from [`PretrainedConfig`] for more information. Args: vocab_size (`int`, *optional*, defaults to 50265): Vocabulary size of the OFA model. Defines the number of different tokens that can be represented by the `inputs_ids` passed when calling [`~OFAModel`] or [`~TFOFAModel`]. d_model (`int`, *optional*, defaults to 1024): Dimension of the layers and the pooler layer. encoder_layers (`int`, *optional*, defaults to 12): Number of encoder layers. decoder_layers (`int`, *optional*, defaults to 12): Number of decoder layers. encoder_attention_heads (`int`, *optional*, defaults to 16): Number of attention heads for each attention layer in the Transformer encoder. decoder_attention_heads (`int`, *optional*, defaults to 16): Number of attention heads for each attention layer in the Transformer decoder. decoder_ffn_dim (`int`, *optional*, defaults to 4096): Dimension of the "intermediate" (often named feed-forward) layer in decoder. encoder_ffn_dim (`int`, *optional*, defaults to 4096): Dimension of the "intermediate" (often named feed-forward) layer in decoder. activation_function (`str` or `function`, *optional*, defaults to `"gelu"`): The non-linear activation function (function or string) in the encoder and pooler. If string, `"gelu"`, `"relu"`, `"silu"` and `"gelu_new"` are supported. dropout (`float`, *optional*, defaults to 0.1): The dropout probability for all fully connected layers in the embeddings, encoder, and pooler. attention_dropout (`float`, *optional*, defaults to 0.0): The dropout ratio for the attention probabilities. activation_dropout (`float`, *optional*, defaults to 0.0): The dropout ratio for activations inside the fully connected layer. classifier_dropout (`float`, *optional*, defaults to 0.0): The dropout ratio for classifier. max_position_embeddings (`int`, *optional*, defaults to 1024): The maximum sequence length that this model might ever be used with. Typically set this to something large just in case (e.g., 512 or 1024 or 2048). init_std (`float`, *optional*, defaults to 0.02): The standard deviation of the truncated_normal_initializer for initializing all weight matrices. encoder_layerdrop: (`float`, *optional*, defaults to 0.0): The LayerDrop probability for the encoder. See the [LayerDrop paper](see https://arxiv.org/abs/1909.11556) for more details. decoder_layerdrop: (`float`, *optional*, defaults to 0.0): The LayerDrop probability for the decoder. See the [LayerDrop paper](see https://arxiv.org/abs/1909.11556) for more details. use_cache (`bool`, *optional*, defaults to `True`): Whether or not the model should return the last key/values attentions (not used by all models). is_encoder_decoder (`bool`, *optional*, defaults to `True`): Whether or not the model is used as an encoder/decoder. scale_embedding (`bool`, *optional*, defaults to `False`): Whether or not to scale the embedding. If True, embedding = Math.qrt(d_model) * embedding pad_token_id (`int`, *optional*, defaults to `1`): The id of the _padding_ token. bos_token_id (`int`, *optional*, defaults to `0`): The id of the _beginning-of-stream_ token. decoder_start_token_id (`int`, *optional*, defaults to `0`): If an encoder-decoder model starts decoding with a different token than _bos_, the id of that token. eos_token_id (`int`, *optional*, defaults to `2`): The id of the _end-of-stream_ token. forced_eos_token_id (`int`, *optional*, defaults to `2`): The id of the token to force as the last generated token when `max_length` is reached. encoder_normalize_before (`bool`, *optional*, defaults to `True`): Whether or not to use layer normalization before the self attention and fc layer within encoder layer decoder_normalize_before (`bool`, *optional*, defaults to `True`): Whether or not to use layer normalization before the self attention and fc layer within decoder layer normformer (`bool`, *optional*, defaults to `True`): Whether or not to use layer normalization between the self attention layers and fc layer within encoder&decoder layer encoder_drop_path_rate (`float`, *optional*, defaults to `0.0`): The drop path rate using in the encoder. see more about drop path [drop path](https://arxiv.org/abs/1605.07648) decoder_drop_path_rate (`float`, *optional*, defaults to `0.0`): The drop path rate using in the decoder. see more about drop path [drop path](https://arxiv.org/abs/1605.07648) layernorm_embedding (`bool`, *optional*, defaults to `True`): Whether or not to use layer normalization for text input embedding in encoder and decoder. patch_layernorm_embedding (`bool`, *optional*, defaults to `True`): Whether or not to use layer normalization for image patch input embedding in encoder and decoder. entangle_position_embedding (`bool`, *optional*, defaults to `False`): Whether or not to entangle position embedding to input embedding. resnet_type (`str`, *optional*, defaults to `"resnet101"`): The image encoder's type in OFA, only works when use_ofasys=False. `"resnet18"`, `"resnet34"`, `"resnet50"`, `"resnet101"` and `"resnet152"` are supported. resnet_model_path (`str`, *optional*, defaults to `None`): The path where can load resnet model. If None, will use random initialized weights. resnet_drop_path_rate The drop path rate using in resnet for image encoding, see more about drop path [drop path](https://arxiv.org/abs/1605.07648) token_bucket_size (`int`, *optional*, defaults to `256`): The number of token buckets to use for each attention layer. image_bucket_size (`int`, *optional*, defaults to `42`): The number of image buckets to use for each attention layer. add_type_embedding (`bool`, *optional*, defaults to `True`): Whether or not to add type embedding to the input while encoding. So far, type means the type of modality, and only Text&Image modalities is supported, `0`=Text, `1`=Image share_decoder_input_output_embed (`bool`, *optional*, defaults to `True`): Whether or not to share the input embedding table as the weights the output projection in decoder. If False, using a new linear projection. attn_scale_factor (`float`, *optional*, defaults to `2.0`): The position embedding scaling factor. If it works, position_embedding = position_embedding * float(d_model / num_attention_heads * attn_scale_factor)**-0.5 code_layernorm_embedding (`bool`, *optional*, defaults to `True`): Whether or not to user layer normalization for code generation code_image_size (`int`, *optional*, defaults to `128`): Image size of generated images. Also used in calculating the image's position id for attention bias. interpolate_position (`bool`, *optional*, defaults to `False`): Deprecated now, will be deleted in next version. orig_patch_image_size (`int`, *optional*, defaults to `224`): Deprecated now, will be deleted in next version. share_attn_bias (`bool`, *optional*, defaults to `False`): Whether or not to share attn_bias cross transformer layers use_image_feature (`bool`, *optional*, defaults to `True`): Whether or not the model have image modality. disable_entangle (`bool`, *optional*, defaults to `False`): Whether or not to disable the entangle relative configs. use_ofasys (`bool`, *optional*, defaults to `False`): Whether or not the model is come from OFA-Sys. If True, the model structure will be some differences from OFA vit_type (`str`, *optional*, defaults to `"vit_base"`): The image encoder's type in OFA-Sys, only works when use_ofasys=True. `"vit_base"`, `"vit_large"`, `"vit_large_336"` and `"vit_huge"` are supported. vit_drop_path_rate The drop path rate using the image encoder vit. see more about drop path [drop path](https://arxiv.org/abs/1605.07648) ofapast_key_valuesencoder_attention_headsd_model)num_attention_heads hidden_sizeATgelu皙?{Gz?Fr resnet101N*@vit_base?c:; s||_||_||_||_||_||_||_||_||_||_ ||_ ||_ | |_ ||_ | |_| |_||_| |_||_||_||_||_||_||_||_||_| |_|!|_|"|_|#|_|$|_|%|_|&|_ |'|_!|(|_"|)|_#|*|_$|+|_%|,|_&|-|_'|.|_(|/|_)|0|_*|1|_+|2|_,|3|_-|4|_.|5|_/|6|_0|7|_1|8|_2|9|_3t4j5d|||| ||d|:|j6dur|:7ddr|j8|_6t9:d|j8ddSdSdS)N) pad_token_id bos_token_id eos_token_idis_encoder_decoderdecoder_start_token_idforced_eos_token_idforce_bos_token_to_be_generatedFz:Please make sure the config includes `forced_bos_token_id=zT` in future versions. The config can simply be saved and uploaded again to be fixed.); vocab_sizemax_position_embeddingsrencoder_ffn_dimencoder_layersrdecoder_ffn_dimdecoder_layersdecoder_attention_headsdropoutattention_dropoutactivation_dropoutactivation_functioninit_stdencoder_layerdropdecoder_layerdropclassifier_dropout use_cachenum_hidden_layersscale_embeddingencoder_normalize_beforedecoder_normalize_before normformerencoder_drop_path_ratedecoder_drop_path_ratelayernorm_embeddingpatch_layernorm_embedding resnet_typeresnet_model_pathresnet_drop_path_ratetoken_bucket_sizeimage_bucket_sizeadd_type_embedding share_decoder_input_output_embedattn_scale_factorcode_layernorm_embeddingcode_image_sizeentangle_position_embeddinginterpolate_positionorig_patch_image_sizeshare_attn_biasuse_image_featuredisable_entangle use_ofasysvit_typevit_drop_path_rateuse_gamma_featuregamma exclude_mlptemperature_init_valueremove_decoder_type_embeddingmlp_dimsuper__init__forced_bos_token_idgetr warningswarn);selfr'r(r*r)rr,r+r-r3r4r6r"r1rr.r/r0r2r5r8rr r#r!r$r9r:r;r<r=r>r?r@rArBrCrDrErFrGrHrIrJrKrLrMrNrOrPrQrRrSrTrUrVrWrXkwargs __class__r&g/home/ubuntu/.local/lib/python3.10/site-packages/modelscope/models/multi_modal/ofa/configuration_ofa.pyrZs;  zOFAConfig.__init__)9r r r rrr rrrrTTrrrrrrrFrrrrrTTTrrTTrNrrrTTrTrFFrFTFFrrFrTNFr) __name__ __module__ __qualname____doc__ model_typekeys_to_ignore_at_inference attribute_maprZ __classcell__r&r&rarcrsr) rgr] transformersrtransformers.utilsr get_loggerlogger!OFA_PRETRAINED_CONFIG_ARCHIVE_MAPrr&r&r&rcs