i0ddlZddlmZddlmZmZddlmZddlmZm Z ddl m Z ddl m Z dd lmZdd lmZmZdd lmZd d lmZd dlmZmZmZmZmZmZmZmZm Z m!Z!m"Z"GddeeZ#GddeZ$GddeZ%Gdde!Z&GddeZ'Gdde Z(Gdde"Z)GddeZ*Gdd eZ+Gd!d"eZ,Gd#d$eZ-Gd%d&eZ.gd'Z/y)(N)nn)Cache DynamicCache)PreTrainedConfig)create_causal_mask!create_sliding_window_causal_mask)BaseModelOutputWithPast)RopeParameters)Unpack)TransformersKwargsauto_docstring)check_model_inputs) MistralConfig) Qwen2AttentionQwen2DecoderLayerQwen2ForCausalLMQwen2ForQuestionAnsweringQwen2ForSequenceClassificationQwen2ForTokenClassificationQwen2MLP Qwen2ModelQwen2PreTrainedModel Qwen2RMSNormQwen2RotaryEmbeddingc*eZdZdZdZ ddedzdedzdedzdedzdedzd edzd edzd edzd edzd edzdedzdedzdedzdedzdedzdedzde dzdedzdedzde edzf(dZ y)MinistralConfiga This is the configuration class to store the configuration of a [`MinistralModel`]. It is used to instantiate an Ministral 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 Ministral-8B-Instruct-2410. [mistralai/Ministral-8B-Instruct-2410](https://huggingface.co/mistralai/Ministral-8B-Instruct-2410) [mistralai/Ministral-8B-Instruct-2410](https://huggingface.co/mistralai/Ministral-8B-Instruct-2410) 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 32000): Vocabulary size of the Ministral model. Defines the number of different tokens that can be represented by the `inputs_ids` passed when calling [`MinistralModel`] hidden_size (`int`, *optional*, defaults to 4096): Dimension of the hidden representations. intermediate_size (`int`, *optional*, defaults to 14336): Dimension of the MLP representations. num_hidden_layers (`int`, *optional*, defaults to 32): Number of hidden layers in the Transformer encoder. num_attention_heads (`int`, *optional*, defaults to 32): Number of attention heads for each attention layer in the Transformer encoder. num_key_value_heads (`int`, *optional*, defaults to 8): This is the number of key_value heads that should be used to implement Grouped Query Attention. If `num_key_value_heads=num_attention_heads`, the model will use Multi Head Attention (MHA), if `num_key_value_heads=1` the model will use Multi Query Attention (MQA) otherwise GQA is used. When converting a multi-head checkpoint to a GQA checkpoint, each group key and value head should be constructed by meanpooling all the original heads within that group. For more details, check out [this paper](https://huggingface.co/papers/2305.13245). If it is not specified, will default to `8`. head_dim (`int`, *optional*, defaults to `hidden_size // num_attention_heads`): The attention head dimension. hidden_act (`str` or `function`, *optional*, defaults to `"silu"`): The non-linear activation function (function or string) in the decoder. max_position_embeddings (`int`, *optional*, defaults to `4096*32`): The maximum sequence length that this model might ever be used with. Ministral's sliding window attention allows sequence of up to 4096*32 tokens. initializer_range (`float`, *optional*, defaults to 0.02): The standard deviation of the truncated_normal_initializer for initializing all weight matrices. rms_norm_eps (`float`, *optional*, defaults to 1e-06): The epsilon used by the rms normalization layers. use_cache (`bool`, *optional*, defaults to `True`): Whether or not the model should return the last key/values attentions (not used by all models). Only relevant if `config.is_decoder=True`. pad_token_id (`int`, *optional*): The id of the padding token. bos_token_id (`int`, *optional*, defaults to 1): The id of the "beginning-of-sequence" token. eos_token_id (`int`, *optional*, defaults to 2): The id of the "end-of-sequence" token. tie_word_embeddings (`bool`, *optional*, defaults to `False`): Whether the model's input and output word embeddings should be tied. rope_parameters (`RopeParameters`, *optional*): Dictionary containing the configuration parameters for the RoPE embeddings. The dictionary should contain a value for `rope_theta` and optionally parameters used for scaling in case you want to use RoPE with longer `max_position_embeddings`. sliding_window (`int`, *optional*, defaults to 4096): Sliding window attention window size. If not specified, will default to `4096`. attention_dropout (`float`, *optional*, defaults to 0.0): The dropout ratio for the attention probabilities. layer_types (`list`, *optional*): Attention pattern for each layer. ```python >>> from transformers import MinistralModel, MinistralConfig >>> # Initializing a Ministral 8B style configuration >>> configuration = MinistralConfig() >>> # Initializing a model from the Ministral 8B style configuration >>> model = MinistralModel(configuration) >>> # Accessing the model configuration >>> configuration = model.config ``` ministralN vocab_size hidden_sizeintermediate_sizenum_hidden_layersnum_attention_headsnum_key_value_headshead_dim hidden_actmax_position_embeddingsinitializer_range rms_norm_eps use_cache pad_token_id bos_token_id eos_token_idtie_word_embeddingsrope_parameterssliding_windowattention_dropout layer_typesc | |_||_||_||_||_| |_||_||_||_||_ ||_ ||_ ||}||_ ||_ | |_| |_| |_||_||_|j$|jdndg|z|_||_t)j*|fi|y)Nsliding_attentionfull_attention)r,r-r.r/r r(r!r"r#r$r1r&r%r'r)r*r+r2r3r0r__init__)selfr r!r"r#r$r%r&r'r(r)r*r+r,r-r.r/r0r1r2r3kwargss y/home/obispo/Crisostomo_bridge/mision_env/lib/python3.12/site-packages/transformers/models/ministral/modular_ministral.pyr7zMinistralConfig.__init__ls0)((#6 $'>$&!2!2#6 ,    &"5 #6 $!2("!2&    #'+':':'F#L\ ! "D  /!!$1&1)i}i8 r=Nsiluig{Gz?gư>TNrFNr<gN) __name__ __module__ __qualname____doc__ model_typeintstrfloatboolr listr7r;r:rrsoK ZJ"'"&(-(**,*+#!'.7*.%)!%#'#$#$+015%)*-(,+82$J824Z82: 82 : 82 !4Z 82!4Z82*82$J82"%t82!4<82dl82$;82Dj82Dj82 Dj!82""D[#82$($.%82&d '82(!4<)82*#Y%+82r;rc eZdZy) MinistralMLPNrArBrCrKr;r:rMrMr;rMc$eZdZdeffd ZxZS)MinistralAttention layer_idxct|||tj|j|j |j zd|_tj|j|j|j zd|_ tj|j|j|j zd|_ y)NF)bias) superr7rLinearr!r$r&q_projr%k_projv_proj)r8configrR __class__s r:r7zMinistralAttention.__init__s +ii 2 2F4N4NQUQ^Q^4^ejk ii 2 2F4N4NQUQ^Q^4^ejk ii 2 2F4N4NQUQ^Q^4^ejk r;)rArBrCrFr7 __classcell__r[s@r:rQrQsl#llr;rQc eZdZy)MinistralRMSNormNrNrKr;r:r_r_rOr;r_c eZdZy)MinistralDecoderLayerNrNrKr;r:rararOr;rac eZdZy)MinistralPreTrainedModelNrNrKr;r:rcrcrOr;rcc eZdZy)MinistralRotaryEmbeddingNrNrKr;r:rererOr;receZdZdeffd Zee ddejdzdejdzdejdzde dzdejdzd e dzd ejdzd e ed efd ZxZS)MinistralModelrZc(t|||`y)N)rUr7has_sliding_layers)r8rZr[s r:r7zMinistralModel.__init__s   #r;N input_idsattention_mask position_idspast_key_values inputs_embedsr+cache_positionr9returnc |du|duz r td||j|}|r|t|j}|F||j nd} t j | | |jdz|j}||jd}t|x} ts*|j|||||d} td i| td i| d} |} |j| |} |jd|jj D]}|| f| |j"||||| d|} !|j%| } t'| |r| Sd S) Nz:You must specify exactly one of input_ids or inputs_embeds)rZrr@)device)rZ input_embedsrkrormrl)r6r5)rkrlrmr+roposition_embeddings)last_hidden_statermrK) ValueError embed_tokensrrZget_seq_lengthtorcharangeshaperr unsqueeze isinstancedictrr rotary_emblayersr#attention_typenormr )r8rjrkrlrmrnr+ror9past_seen_tokenscausal_mask_mapping mask_kwargs hidden_statesrt decoder_layers r:forwardzMinistralModel.forwards  -t"; <YZ Z   --i8M 0*$++>O  !CRC^==?de "\\ "2]5H5H5K"KTaThThN  )33A6L?-F++ -"0"0#2 , K#5"C{"C%F%U%U#  & "oom\J![[)H4;;+H+HI M) 2=3O3OP) /#-$7  M  -0 &+/8O  >B  r;)NNNNNNN)rArBrCrr7rrry LongTensorTensorr FloatTensorrIr r r rr\r]s@r:rgrgs$$.2.204(,26!%26A ##d*A  t+A &&- A  A ((4/ A $;A ((4/A +,A  !A A r;rgc eZdZy)MinistralForCausalLMNrNrKr;r:rrrOr;rc eZdZy)"MinistralForSequenceClassificationNrNrKr;r:rrrOr;rc eZdZy)MinistralForTokenClassificationNrNrKr;r:rrrOr;rc eZdZy)MinistralForQuestionAnsweringNrNrKr;r:rrrOr;r)rrcrgrrrr)0ryr cache_utilsrrconfiguration_utilsr masking_utilsrr modeling_outputsr modeling_rope_utilsr processing_utilsr utilsr r utils.genericrmistral.configuration_mistralrqwen2.modeling_qwen2rrrrrrrrrrrrrMrQr_rarcrergrrrr__all__rKr;r:rs .3R71&7/9    H2m%5H2V 8 ll |  -  3  3 H ZH V +  )G  &A  $=  r;