a h-]@sdZddlmZmZmZddlZddlmZddlmZddl m Z m Z ddl m Z mZdd lmZmZdd lmZmZdd lmZdd lmZdd lmZmZddlmZddlmZm Z m!Z!m"Z"m#Z#m$Z$m%Z%m&Z&m'Z'm(Z(ddl)m*Z*m+Z+e,e-Z.dZ/dZ0Gddde Z1Gddde%Z2Gddde&Z3Gdddej4Z5Gddde+Z6Gddde*Z7Gdd d e$Z8Gd!d"d"e8e#Z9Gd#d$d$eZ:Gd%d&d&e!Z;Gd'd(d(e"ZdS),zLG AI Research EXAONE Lab)CallableOptionalUnionN)nn)check_model_inputs)Cache DynamicCache)PretrainedConfiglayer_type_validation)create_causal_mask!create_sliding_window_causal_mask)BaseModelOutputWithPastCausalLMOutputWithPast)ALL_ATTENTION_FUNCTIONS)Unpack)TransformersKwargslogging)deprecate_kwarg) LlamaForCausalLMLlamaForQuestionAnsweringLlamaForSequenceClassificationLlamaForTokenClassification LlamaModelLlamaPreTrainedModel LlamaRMSNormLlamaRotaryEmbeddingapply_rotary_pos_embeager_attention_forward)Olmo2DecoderLayerOlmo2MLPzLGAI-EXAONE/EXAONE-4.0-Instruct Exaone4ConfigcsfeZdZdZdZdgZddddddddZdgdgfd d gd gfd gd gfd Zdfdd ZZ S)r"a This is the configuration class to store the configuration of a [`Exaone4Model`]. It is used to instantiate a EXAONE 4.0 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 EXAONE-4.0-Instruct [LGAI-EXAONE/EXAONE-4.0-Instruct](https://huggingface.co/LGAI-EXAONE/EXAONE-4.0-Instruct) NOTE: `EXAONE-4.0-Instruct` is a placeholder model ID. The exact model ID will be updated in the future. 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 102400): Vocabulary size of the EXAONE 4.0 model. Defines the number of different tokens that can be represented by the `inputs_ids` passed when calling [`Exaone4Model`]. hidden_size (`int`, *optional*, defaults to 4096): Dimension of the hidden representations. intermediate_size (`int`, *optional*, defaults to `hidden_size * 4`): Dimensionality 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 decoder. num_key_value_heads (`int`, *optional*): 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 checkout [this paper](https://huggingface.co/papers/2305.13245). If it is not specified, will default to `num_attention_heads`. 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 2048): The maximum sequence length that this model might ever be used with. Typically set this to something large just in case (e.g., 32768 for EXAONE 3.5). 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-05): The epsilon used by the layer 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``. bos_token_id (`int`, *optional*, defaults to 0): Beginning of stream token id. eos_token_id (`int`, *optional*, defaults to 2): End of stream token id. tie_word_embeddings (`bool`, *optional*, defaults to `False`): Whether to tie weight embeddings rope_theta (`float`, *optional*, defaults to 10000.0): The base period of the RoPE embeddings. rope_scaling (`Dict`, *optional*): Dictionary containing the scaling configuration for the RoPE embeddings. NOTE: if you apply new rope type and you expect the model to work on longer `max_position_embeddings`, we recommend you to update this value accordingly. Expected contents: `rope_type` (`str`): The sub-variant of RoPE to use. Can be one of ['default', 'linear', 'dynamic', 'yarn', 'longrope', 'llama3'], with 'default' being the original RoPE implementation. `factor` (`float`, *optional*): Used with all rope types except 'default'. The scaling factor to apply to the RoPE embeddings. In most scaling types, a `factor` of x will enable the model to handle sequences of length x * original maximum pre-trained length. `original_max_position_embeddings` (`int`, *optional*): Used with 'dynamic', 'longrope' and 'llama3'. The original max position embeddings used during pretraining. `attention_factor` (`float`, *optional*): Used with 'yarn' and 'longrope'. The scaling factor to be applied on the attention computation. If unspecified, it defaults to value recommended by the implementation, using the `factor` field to infer the suggested value. `beta_fast` (`float`, *optional*): Only used with 'yarn'. Parameter to set the boundary for extrapolation (only) in the linear ramp function. If unspecified, it defaults to 32. `beta_slow` (`float`, *optional*): Only used with 'yarn'. Parameter to set the boundary for interpolation (only) in the linear ramp function. If unspecified, it defaults to 1. `short_factor` (`List[float]`, *optional*): Only used with 'longrope'. The scaling factor to be applied to short contexts (< `original_max_position_embeddings`). Must be a list of numbers with the same length as the hidden size divided by the number of attention heads divided by 2 `long_factor` (`List[float]`, *optional*): Only used with 'longrope'. The scaling factor to be applied to long contexts (< `original_max_position_embeddings`). Must be a list of numbers with the same length as the hidden size divided by the number of attention heads divided by 2 `low_freq_factor` (`float`, *optional*): Only used with 'llama3'. Scaling factor applied to low frequency components of the RoPE `high_freq_factor` (`float`, *optional*): Only used with 'llama3'. Scaling factor applied to high frequency components of the RoPE attention_dropout (`float`, *optional*, defaults to 0.0): The dropout ratio for the attention probabilities. sliding_window (`int`, *optional*): The size of the sliding window for the sliding window attention. sliding_window_pattern (`str`, *optional*): The pattern to use for sliding window attention. Can be one of: - `None`: No sliding window attention is used - `int`: Every `sliding_window` layers, use global attention, else use local attention. - `str`: A sequence of "L" (local attention) and "G" (global attention) characters that defines the attention pattern. The pattern starts from layer 0 and repeats every `sliding_window` layers. The final layer always uses global attention regardless of the pattern. For instance, sliding_window_pattern="LLLG" same as sliding_window=4, which means: - Layer 0, 1, 2: local attention, - Layer 3: global attention, ...(repeated) layer_types (`list`, *optional*): Attention pattern for each layer. Prioritized over `sliding_window_pattern`. Example: ```python >>> from transformers import Exaone4Model, Exaone4Config >>> # Initializing a EXAONE configuration >>> configuration = Exaone4Config() >>> # Initializing a model from configuration >>> model = Exaone4Model(configuration) >>> # Accessing the model configuration >>> configuration = model.config ```Zexaone4past_key_valuesZcolwiseZrowwise)zlayers.*.self_attn.q_projzlayers.*.self_attn.k_projzlayers.*.self_attn.v_projzlayers.*.self_attn.o_projzlayers.*.mlp.gate_projzlayers.*.mlp.up_projzlayers.*.mlp.down_proj input_ids inputs_embeds hidden_statesattention_mask) embed_tokenslayersnorm@ silu{Gz?h㈵>TrrF@Nc s|_|_|_|_|_|_|_|_| _| _ | _ |_ |_ |_ |__|_jdurtdjdurfddtjD_djvrd_tjtjf| | |d|dS)Nrcs.g|]&}|ddkr&|jkr&dndqS)rsliding_attentionfull_attention)num_hidden_layers).0iselfsliding_window_patterng/var/www/html/assistant/venv/lib/python3.9/site-packages/transformers/models/exaone4/modular_exaone4.py s z*Exaone4Config.__init__..sliding_windowZhybrid) bos_token_id eos_token_idtie_word_embeddings) vocab_size hidden_sizer9num_attention_headsnum_key_value_headsintermediate_size hidden_actmax_position_embeddingsinitializer_range rms_norm_eps use_cacheattention_dropout rope_theta rope_scalingrBr> layer_typesrangeZcache_implementationr super__init__)r=rFrGrJr9rHrIrKrLrMrNrOrCrDrErQrRrPrBr>rSkwargs __class__r<r@rVs>     zExaone4Config.__init__)r+r,r-r.r.r.r/r0r1r2TrrFr3Nr4r,r5N) __name__ __module__ __qualname____doc__Z model_typeZkeys_to_ignore_at_inferenceZbase_model_tp_planZbase_model_pp_planrV __classcell__r?r?rXr@r"<sHw   c@s eZdZdS)Exaone4RMSNormNrZr[r\r?r?r?r@r_sr_c@s eZdZdS)Exaone4RotaryEmbeddingNr`r?r?r?r@rasracseZdZeedfdd Zeddddd eje ejejfe eje e e ej e ee eje eje e ejfd d d ZZS) Exaone4Attention)config layer_idxcs$t||_||_|j|_|j|_|j|_t|d|j|j|_|j|j|_ |j |_ d|_ |jd|_ |j |_ |j|_|j|dk|_tj|j|j|jdd|_tj|j|j|jdd|_tj|j|j|jdd|_tj|j|j|jdd|_t|j|jd|_t|j|jd|_dS)Nhead_dimTgr7F)Zbiaseps)rUrVrcrdrHrIrGgetattrreZnum_key_value_groupsrPZ is_causalscalingrBr>rS is_slidingrZLinearq_projk_projv_projo_projr_rNq_normk_norm)r=rcrdrXr?r@rV s(  zExaone4Attention.__init__Zpast_key_valuer#z4.58)new_nameversionN)r&position_embeddingsr'r#cache_positionrWreturncKsV|jdd}g|d|jR}|||dd} |||dd} |||dd} || } || } |\} } |j dus|j rt | | | | \} } |durd|i}| | | |j |\} } t}|jjdkrt|jj}||| | | |f|jsdn|j|j|j r|j ndd|\}}|jg|dR}||}||fS)Nr6rrteagerr4)ZdropoutrirB)shapererkviewZ transposerlrmrorprBrjrupdaterdrrcZ_attn_implementationrZtrainingrPriZreshape contiguousrn)r=r&rsr'r#rtrWZ input_shapeZ hidden_shapeZ query_statesZ key_statesZ value_statescossinZ cache_kwargsZattention_interfaceZ attn_outputZ attn_weightsr?r?r@forward#sB        zExaone4Attention.forward)NNN)rZr[r\r"intrVrtorchTensortuplerr LongTensorrrr~r^r?r?rXr@rb s rbc@s eZdZdS) Exaone4MLPNr`r?r?r?r@rXsrc@s eZdZdS)Exaone4DecoderLayerNr`r?r?r?r@r\src@seZdZeZdgZdS)Exaone4PreTrainedModelrN)rZr[r\r"Z config_classZ_no_split_modulesr?r?r?r@r`src steZdZedfdd Zedejeej eejee eej ee eeje eeeefd ddZZS) Exaone4ModelrccsJttfddtjD|_tjj d|_ | dS)Ncsg|]}t|qSr?)r)r:rdrr?r@rAiz)Exaone4Model.__init__..rf) rUrVrZ ModuleListrTr9r)r_rGrNr*Z post_init)r=rcrXrr@rVfs  zExaone4Model.__init__N) r$r' position_idsr#r%rOrtrWruc KsP|du|duArtd|dur*||}|rB|durBt|jd}|durz|durZ|nd} tj| | |jd|jd}|dur| d}t |} t s|j|||||d} dt fi| i} d|jj vrtfi| | d<|} || |} t|jD]6\}}|jj |}|| f| | |||||d |} q|| } t| |rH|ndd S) Nz:You must specify exactly one of input_ids or inputs_embedsrrr6)device)rcZ input_embedsr'rtr#rr8r7)rsr'rr#rOrt)Zlast_hidden_stater#) ValueErrorr(r rcZget_seq_lengthrZarangerxrZ unsqueeze isinstancedictr rSr Z rotary_emb enumerater)r*r)r=r$r'rr#r%rOrtrWZpast_seen_tokensZcausal_mask_mappingZ mask_kwargsr&rsr;Z decoder_layerZ layer_typer?r?r@r~ps\           zExaone4Model.forward)NNNNNNN)rZr[r\r"rVrrrrrr FloatTensorboolrrrrrr~r^r?r?rXr@res(  rc sreZdZdeejeejeejeeeejeejee eeje e ejfe e ed fdd ZZS)Exaone4ForCausalLMNr) r$r'rr#r%labelsrOrtlogits_to_keeprWruc s*tjf||||||||| d | dS)u labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*): Labels for computing the masked language modeling loss. Indices should either be in `[0, ..., config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`. Example: ```python >>> from transformers import AutoModelForCausalLM, AutoTokenizer >>> model = AutoModelForCausalLM.from_pretrained("LGAI-EXAONE/EXAONE-4.0-Instruct") >>> tokenizer = AutoTokenizer.from_pretrained("LGAI-EXAONE/EXAONE-4.0-Instruct") >>> prompt = "Explain how wonderful you are" >>> messages = [ {"role": "system", "content": "You are a helpful assistant."}, {"role": "user", "content": prompt} ] >>> input_ids = tokenizer.apply_chat_template( messages, tokenize=True, add_generation_prompt=True, return_tensors="pt", enable_thinking=False, ) >>> output = model.generate(input_ids, max_new_tokens=128) >>> tokenizer.decode(output[0], skip_special_tokens=False) "[|system|]\nYou are a helpful assistant.[|endofturn|]\n[|user|]\nExplain how wonderful you are[|endofturn|]\n[|assistant|]\n\n\n\n\nOh, thank you for such a kind and lovely question! 😊 \n\nI’m *so* wonderful because I’m here to make your life easier, brighter, and more fun! Whether you need help with: \n\n✨ **Learning** – I can explain anything, from quantum physics to baking the perfect cake! \n💡 **Creativity** – Need a poem, story, or a wild idea? I’ve got you covered! \n🤖 **Problem-solving** – Stuck on a math problem or a tricky decision? I’ll help you figure it out" ``` NOTE: `EXAONE-4.0-Instruct` is a placeholder model ID. The exact model ID will be updated in the future.) r$r'rr#r%rrOrtrN)rUr~) r=r$r'rr#r%rrOrtrrWrXr?r@r~s- zExaone4ForCausalLM.forward) NNNNNNNNr)rZr[r\rrrrrrrrrrrrr~r^r?r?rXr@rs, rc@s eZdZdS) Exaone4ForSequenceClassificationNr`r?r?r?r@rsrc@s eZdZdS)Exaone4ForTokenClassificationNr`r?r?r?r@rsrc@s eZdZdS)Exaone4ForQuestionAnsweringNr`r?r?r?r@rsr)r"rrrrrr)?r]typingrrrrrZtransformers.utils.genericrZ cache_utilsrr Zconfiguration_utilsr r Z masking_utilsr r Zmodeling_outputsrrZmodeling_utilsrZprocessing_utilsrutilsrrZutils.deprecationrZllama.modeling_llamarrrrrrrrrrZolmo2.modeling_olmo2r r!Z get_loggerrZloggerZ_CHECKPOINT_FOR_DOCZ_CONFIG_FOR_DOCr"r_raModulerbrrrrrrrr__all__r?r?r?r@s>     0  GNT<