o ۷iX.@sddlmZddlZddlmZddlmZmZddlmZddl m Z m Z ddl m Z dd lmZdd lmZmZdd lmZd d lmZd dlmZmZmZmZmZmZmZmZm Z m!Z!m"Z"GdddeeZ#GdddeZ$GdddeZ%Gddde!Z&GdddeZ'Gddde Z(Gddde"Z)GdddeZ*Gdd d eZ+Gd!d"d"eZ,Gd#d$d$eZ-Gd%d&d&eZ.gd'Z/dS)()OptionalN)nn)Cache DynamicCache)PretrainedConfig)create_causal_mask!create_sliding_window_causal_mask)BaseModelOutputWithPast)Unpack)TransformersKwargsauto_docstring)check_model_inputs) MistralConfig) Qwen2AttentionQwen2DecoderLayerQwen2ForCausalLMQwen2ForQuestionAnsweringQwen2ForSequenceClassificationQwen2ForTokenClassificationQwen2MLP Qwen2ModelQwen2PreTrainedModel Qwen2RMSNormQwen2RotaryEmbeddingc@sFeZdZdZdZ               dddZdS)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_theta (`float`, *optional*, defaults to 10000.0): The base period of the RoPE 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 ``` ministral}8 Nsilu{Gz?ư>TrF@cKstj|f| |||d|||_| |_||_||_||_||_||_||_ |dur,|}||_ ||_ | |_ | |_ | |_||_||_||_|jdurX|jdurPdndg||_dSdS)N) pad_token_id bos_token_id eos_token_idtie_word_embeddingssliding_attentionfull_attention)r__init__ vocab_sizemax_position_embeddings hidden_sizeintermediate_sizenum_hidden_layersnum_attention_headssliding_windowhead_dimnum_key_value_heads hidden_actinitializer_range rms_norm_eps use_cache rope_thetaattention_dropout layer_types)selfr1r3r4r5r6r9r8r:r2r;r<r=r*r+r,r-r>r7r?r@kwargsrCe/home/ubuntu/vllm_env/lib/python3.10/site-packages/transformers/models/ministral/modular_ministral.pyr0ksD  zMinistralConfig.__init__)rrr r!r!r"Nr#r$r%r&TNr'rFr(rr)N)__name__ __module__ __qualname____doc__ model_typer0rCrCrCrDrs0Krc@ eZdZdS) MinistralMLPNrErFrGrCrCrCrDrKrKcs"eZdZdeffdd ZZS)MinistralAttention layer_idxcsft||tj|j|j|jdd|_tj|j|j|jdd|_ tj|j|j|jdd|_ dS)NF)bias) superr0rLinearr3r6r8q_projr9k_projv_proj)rAconfigrO __class__rCrDr0s zMinistralAttention.__init__)rErFrGintr0 __classcell__rCrCrWrDrNsrNc@rJ)MinistralRMSNormNrLrCrCrCrDr[rMr[c@rJ)MinistralDecoderLayerNrLrCrCrCrDr\rMr\c@rJ)MinistralPreTrainedModelNrLrCrCrCrDr]rMr]c@rJ)MinistralRotaryEmbeddingNrLrCrCrCrDr^rMr^cseZdZdeffdd Zee       ddeej deej deej dee d eej d ee d eej d eed efddZZS)MinistralModelrVcst||`dS)N)rQr0has_sliding_layers)rArVrWrCrDr0s zMinistralModel.__init__N input_idsattention_mask position_idspast_key_values inputs_embedsr=cache_positionrBreturnc Ks:|du|duAr td|dur||}|r!|dur!t|jd}|dur=|dur-|nd} tj| | |jd|jd}|durF| d}t |} t sf|j|||||d} t d i| t d i| d} |} || |} |jd|jjD]}|| f| |j||||| d|} qw|| } t| |r|d Sdd S) Nz:You must specify exactly one of input_ids or inputs_embeds)rVrr')device)rV input_embedsrbrfrdrc)r/r.)rbrcrdr=rfposition_embeddings)last_hidden_staterdrC) ValueError embed_tokensrrVget_seq_lengthtorcharangeshaperh unsqueeze isinstancedictrr rotary_emblayersr5attention_typenormr )rArarbrcrdrer=rfrBpast_seen_tokenscausal_mask_mapping mask_kwargs hidden_statesrj decoder_layerrCrCrDforwards\        zMinistralModel.forward)NNNNNNN)rErFrGrr0rr rro LongTensorTensorr FloatTensorboolr r r r~rZrCrCrWrDr_s<  r_c@rJ)MinistralForCausalLMNrLrCrCrCrDrrMrc@rJ)"MinistralForSequenceClassificationNrLrCrCrCrDrrMrc@rJ)MinistralForTokenClassificationNrLrCrCrCrDrrMrc@rJ)MinistralForQuestionAnsweringNrLrCrCrCrDrrMr)rr]r_rrrr)0typingrror cache_utilsrrconfiguration_utilsr masking_utilsrr modeling_outputsr processing_utilsr utilsr r utils.genericrmistral.configuration_mistralrqwen2.modeling_qwen2rrrrrrrrrrrrrKrNr[r\r]r^r_rrrr__all__rCrCrCrDs4       4  M