cCi]\ B S r SSKJrJrJr SSKrSSKJr SSKJr SSK J
r
J r SSK J
r
Jr SS KJrJr SS
KJrJr SS KJr SS KJr SS
KJrJr SSK J r SSK J r J r J!r!J"r"J#r#J$r$J%r%J&r&J'r'J(r( SSK)J*r*J+r+ \RX " \-5 r.Sr/Sr0 " S S\
5 r1 " S S\%5 r2 " S S\&5 r3 " S S\Rh 5 r5 " S S \+5 r6 " S S \*5 r7 " S S \$5 r8 " S! S"\8\#5 r9 " S# S$\ 5 r: " S% S&\!5 r; " S' S(\"5 r< " S) S*\ 5 r=/ S+Qr>g),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 )
LlamaForCausalLMLlamaForQuestionAnswering LlamaForSequenceClassificationLlamaForTokenClassification
LlamaModelLlamaPreTrainedModel LlamaRMSNormLlamaRotaryEmbeddingapply_rotary_pos_embeager_attention_forward)Olmo2DecoderLayerOlmo2MLPzLGAI-EXAONE/EXAONE-4.0-32B
Exaone4Configc ^ \ rS rSrSrSrS/rSSSSSSSS.rS/S /4S
S /S
/4S
/S
/4S .r SU 4S
jjr Sr
U =r $ )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-4.0-32B [LGAI-EXAONE/EXAONE-4.0-32B](https://huggingface.co/LGAI-EXAONE/EXAONE-4.0-32B)
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
```exaone4past_key_valuescolwiserowwise)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_tokenslayersnormc D > Xl X l X@l XPl X`l X0l Xpl Xl Xl Xl Xl
UU l Xl UU l
UU l UU l UU l U R c SnU R cI [# U R 5 Vs/ s H# nUS- U- S:w a UU R : a SOSPM% snU l SU R ; a SU l [' U R U R 5 [( TU ]T " SXUS.UD6 g s snf ) Nr sliding_attentionfull_attentionsliding_windowhybrid) bos_token_id eos_token_idtie_word_embeddings )
vocab_size hidden_sizenum_hidden_layersnum_attention_headsnum_key_value_headsintermediate_size
hidden_actmax_position_embeddingsinitializer_range rms_norm_eps use_cacheattention_dropout
rope_theta rope_scalingr5 sliding_window_pattern layer_typesrangecache_implementationr super__init__)selfr; r< r@ r= r> r? rA rB rC rD rE r7 r8 r9 rG rH rF r5 rI rJ kwargsi __class__s e/home/james-whalen/.local/lib/python3.13/site-packages/transformers/models/exaone4/modular_exaone4.pyrN Exaone4Config.__init__ s< 0 %&!2 #6 #6 !2 $'>$!2 ("!2 $ ( ,&<#& &%& " #
t 5 56 7A U 56!;DDZDZ@Z $%& 7 D t / / /(0D % d . .0F0FG
%Vi
ms
s *D )rF rL rA r< rC r@ rJ rB r> r= r? rD rH rG r5 rI rE r; )i i @ rV rV silui g{Gz?gh㈵>Tr r Fg @N rU N) __name__
__module__ __qualname____firstlineno____doc__
model_typekeys_to_ignore_at_inferencebase_model_tp_planbase_model_pp_planrN __static_attributes__
__classcell__rR s @rS r# r# < s t l J#4"5 &/%.%.%."+ )"+ & (9:#%56 8IJ!"_$56 $ ! +9
9
c \ rS rSrSrg)Exaone4RMSNormi r: NrZ r[ r\ r] rc r: rf rS rh rh rf rh c \ rS rSrSrg)Exaone4RotaryEmbeddingi r: Nri r: rf rS rl rl rj rf rl c Z ^ \ rS rSrS\S\4U 4S jjr\" SSSS9 SS \R S
\
\R \R 4 S \ \R S\ \ S \ \R S
\\ S\
\R \ \R \ \
\R 4 4S jj5 rSrU =r$ )Exaone4Attentioni config layer_idxc d > [ TU ] 5 Xl X l UR U l UR
U l UR U l [ USUR UR - 5 U l UR UR
- U l UR U l
SU l U R S- U l UR U l
UR U l UR U S:H U l [" R$ " U R U R U R - SS9U l [" R$ " U R U R
U R - SS9U l [" R$ " U R U R
U R - SS9U l [" R$ " U R U R - U R SS9U l [/ U R UR0 S9U l [/ U R UR0 S9U l g )Nhead_dimTg r3 F)biaseps)rM rN ro rp r> r? r< getattrrr num_key_value_groupsrF is_causalscalingr5 rI rJ
is_slidingr Linearq_projk_projv_projo_projrh rD q_normk_normrO ro rp rR s rS rN Exaone4Attention.__init__
s
"#)#=#= #)#=#= !--
F4F4F&JdJd4de
$*$>$>&B\B\$\!!'!9!9 }}d* $ 3 3&,&C&C# ,,Y7;NNii 0 0$2J2JT]]2Zafg ii 0 0$2J2JT]]2Zafg ii 0 0$2J2JT]]2Zafg ii 8 84== H$JZJZafg $T]]8K8KL $T]]8K8KL rf past_key_valuer' z4.58)new_nameversionr, position_embeddingsr- cache_positionrP returnc UR S S n/ UQSPU R P7nU R U5 R U5 R SS5 n U R U5 R U5 R SS5 n
U R
U5 R U5 R SS5 n U R U 5 n U R U
5 n
Uu pU R b U R ( a [ XX5 u pUb# SU0nUR XU R U5 u p[ nU R R S:w a [" U R R nU" U U U
U U4U R$ ( d SO U R& U R( U R ( a U R OS S.UD6u nnUR* " / UQSP76 R- 5 nU R/ U5 nUU4$ )Nr2 r r eagerrX )dropoutry r5 )shaperr r| view transposer} r~ r r r5 rz r updaterp r ro _attn_implementationr trainingrF ry reshape
contiguousr )rO r, r r- r' r rP input_shape hidden_shape query_states
key_states value_statescossin cache_kwargsattention_interface attn_output attn_weightss rS forwardExaone4Attention.forward" s $))#2. 88b8$--8 {{=166|DNNqRST [[ /44\BLLQPQR
{{=166|DNNqRST {{<0 [[,
& &$//';LVY'_ $L & . L (7'='=jX\XfXfht'u $J(? ;; + +w 6"9$++:Z:Z"[ $7
%
$}}C$2H2HLL26//4..t
%
%
! \ "));;;;FFH kk+. L((rf )rF ro rr r< rx rz r r} rp r> rw r? r r r| ry r5 rI r~ )NNN)rZ r[ r\ r] r# intrN r torchTensortupler r
LongTensorr r r rc rd re s @rS rn rn s M} M M0 %0A6R
26+/59
1) ||1) #5<< #= >1) ! . 1)
"% 1) !!1!12
1) +,1)
u||Xell 3XeELL>Q5RR S1) S1)rf rn c \ rS rSrSrg)
Exaone4MLPiW r: Nri r: rf rS r r W rj rf r c \ rS rSrSrg)Exaone4DecoderLayeri[ r: Nri r: rf rS r r [ rj rf r c \ rS rSr\rS/rSrg)Exaone4PreTrainedModeli_ r r: N)rZ r[ r\ r] r# config_class_no_split_modulesrc r: rf rS r r _ s L./rf r c " ^ \ rS rSrS\4U 4S jjr\" 5 SS\\R S\\R S\\R S\\ S\\R S \\
S
\\R S \\ S \\\4 4S
jj5 rSrU =r$ ) Exaone4Modelid ro c $ > [ TU ] U5 [ R " [ UR
5 Vs/ s H n[
X5 PM sn5 U l [ UR UR S9U l U R 5 g s snf )Nrt )
rM rN r
ModuleListrK r= r r/ rh r< rD r0 post_initr s rS rN Exaone4Model.__init__e ss
mmEJ6KcKcEd eEd
3Ed e
#6#5#56;N;NO
fs B
r* r- position_idsr' r+ rE r rP r c US L US L- ( a [ S5 eUc U R U5 nU( a Uc [ U R S9nUcD Ub UR 5 OSn [
R " XUR S - UR S9nUc UR S5 n[ U=n
[ 5 ( dH U R UUUUUS.n S[ S 0 U D60n
SU R R ; a [ S 0 U D6U
S' Un U R X5 n
[! U R" 5 H1 u pU R R U nU" U 4U
U
U UUUUS .UD6n M3 U R% U 5 n [' U U( a US
9$ S S
9$ ) Nz:You must specify exactly one of input_ids or inputs_embeds)ro r r2 )device)ro input_embedsr- r r' r r4 r3 )r r- r r' rE r )last_hidden_stater' r: )
ValueErrorr. r
ro get_seq_lengthr aranger r unsqueeze
isinstancedictr
rJ r
rotary_emb enumerater/ r0 r )rO r* r- r r' r+ rE r rP past_seen_tokenscausal_mask_mapping mask_kwargsr, r rQ
decoder_layer
layer_types rS r Exaone4Model.forwardo s
-t"; < YZ Z - -i 8M 0 *$++ >O !CRC^ = = ?de "\\ ]5H5H5K"KTaThTh N )33A6L ?-FF ++ -"0"0#2 ,
K !"4"C{"C# #dkk&=&==;\;k_j;k#$78%
"oom J )$++ 6 A 003J ) $7 2: > ) /# - M !7 -0
& +/8O
>B
rf )r/ r0 )NNNNNNN)rZ r[ r\ r] r# rN r r r r r r FloatTensorboolr r r r r r rc rd re s @rS r r d s } 151537+/59$(59E
E , ,-E
! .E
u / /0 E
"% E
1 12
E
D>E
!!1!12E
+,E
u-- .E
E
rf r c : ^ \ rS rSr SS\\R S\\R S\\R S\\ S\\R S\\R S\\
S \\R S
\ \ \R 4 S \
\ S \4U 4S
jjjrSrU =r$ )Exaone4ForCausalLMi r* r- r r' r+ labelsrE r logits_to_keeprP r c
: > [ T U ] " SUUUUUUUUU S. U
D6 g)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-4.0-32B")
>>> tokenizer = AutoTokenizer.from_pretrained("LGAI-EXAONE/EXAONE-4.0-32B")
>>> 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"
```
) r* r- r r' r+ r rE r r Nr: )rM r ) rO r* r- r r' r+ r rE r r rP rR s rS r Exaone4ForCausalLM.forward s< X
)% +' ))
rf r: ) NNNNNNNNr )rZ r[ r\ r] r r r r r r r r r r r r r rc rd re s @rS r r s 151537+/59-1$(59347
E , ,-7
! .7
u / /0 7
"% 7
1 12
7
))*7
D>7
!!1!127
c5<< /07
+,7
7
7
rf r c \ rS rSrSrg) Exaone4ForSequenceClassificationi r: Nri r: rf rS r r rj rf r c \ rS rSrSrg) Exaone4ForTokenClassificationi r: Nri r: rf rS r r rj rf r c \ rS rSrSrg)Exaone4ForQuestionAnsweringi r: Nri r: rf rS r r rj rf r )r# r r r r r r )?r^ typingr r r r r transformers.utils.genericr cache_utilsr r
configuration_utilsr r
masking_utilsr
r modeling_outputsr r modeling_utilsr processing_utilsr utilsr r utils.deprecationr llama.modeling_llamar r r r r r r r r r olmo2.modeling_olmo2r! r"
get_loggerrZ logger_CHECKPOINT_FOR_DOC_CONFIG_FOR_DOCr# rh rl Modulern r r r r r r r r __all__r: rf rS r s , , 9 . J R 6 & 1 ?
H %2 !B
$ B
J \ 1 K)ryy K)\ + 0 1 0
Q
): Q
h8
) 8
v 'E $? "; rf