qyi&SSKJr SSKJr SSKJr SSKJrJr SSK J r \R"\ 5r \"SS 9\"S S \555r\"SS 9\"S S \555rS S /rg))Any)strict)PreTrainedConfig)auto_docstringlogging)SiglipVisionConfigzgoogle/gemma-3-4b-it) checkpointc >^\rSrSr%SrSrS/rSSSSSSSSSS. rS /S /4S S /S /4S /S /4S .rSr \ \ S'Sr \ \ S'Sr \ \ S'Sr\ \ S'Sr\ \ S'Sr\ \ S'Sr\ \ S'Sr\\ S'Sr\ \ S'S r\\ S!'S"r\\ S#'S$r\\ S%'S&r\ S'-\ S('S)r\ \\ -S'-\ S*'S+r\ S'-\ S,'S$r\\ S-'S'r\ S'-\ S.'S/r!\\ S0'S1r"\ \-S'-\ S2'Sr#\ \ S3'S4r$\ S'-\ S5'S'r%\\S'-\ S6'S'r&\S'-\ S7'S'r'\S'-\ S8'S/r(\S'-\ S9'S:S;S<.r)U4S=jr*S>r+S?r,S@r-U=r.$)AGemma3TextConfig!a query_pre_attn_scalar (`float`, *optional*, defaults to 256): scaling factor used on the attention scores final_logit_softcapping (`float`, *optional*): Scaling factor when applying tanh softcapping on the logits. attn_logit_softcapping (`float`, *optional*): Scaling factor when applying tanh softcapping on the attention scores. use_bidirectional_attention (`bool`, *optional*, defaults to `False`): If True, the model will attend to all text tokens instead of using a causal mask. This does not change behavior for vision tokens. ```python >>> from transformers import Gemma3TextModel, Gemma3TextConfig >>> # Initializing a Gemma3Text gemma3_text-7b style configuration >>> configuration = Gemma3TextConfig() >>> # Initializing a model from the gemma3_text-7b style configuration >>> model = Gemma3TextModel(configuration) >>> # Accessing the model configuration >>> configuration = model.config ``` gemma3_textpast_key_valuescolwisereplicated_with_grad_allreducerowwise) zlayers.*.self_attn.q_projzlayers.*.self_attn.k_projzlayers.*.self_attn.v_projzlayers.*.self_attn.q_normzlayers.*.self_attn.k_normzlayers.*.self_attn.o_projzlayers.*.mlp.gate_projzlayers.*.mlp.up_projzlayers.*.mlp.down_proj input_ids inputs_embeds hidden_statesattention_mask) embed_tokenslayersnormi@ vocab_sizei hidden_sizei$intermediate_sizenum_hidden_layersnum_attention_headsnum_key_value_headshead_dimgelu_pytorch_tanhhidden_activationimax_position_embeddings{Gz?initializer_rangegư> rms_norm_epsT use_cacherN pad_token_id eos_token_idr bos_token_idtie_word_embeddingsrope_parametersFattention_biasgattention_dropoutquery_pre_attn_scalarisliding_window layer_typesfinal_logit_softcappingattn_logit_softcappinguse_bidirectional_attentiong.Ag@)globallocalc b>UR(aURS-S-UlURSS5UlURcM[ UR 5Vs/sH'n[US-UR-5(aSOSPM) snUl[TU]$"S0UD6 gs snf)Nr r.sliding_window_patternsliding_attentionfull_attention) r:r6get_sliding_window_patternr7rangerboolsuper __post_init__)selfkwargsi __class__s l/home/mande/repo/quber/.venv/lib/python3.13/site-packages/transformers/models/gemma3/configuration_gemma3.pyrHGemma3TextConfig.__post_init__hs  + +#'#6#6!#;q"@D (.zz2JA'N$    #t556 6A(,QUd6R6R,R'S'S#Yii6 D  ''  s%.B,cURUR-S:wa&[SURSURS35eg)zOPart of `@strict`-powered validation. Validates the architecture of the config.rzThe hidden size (z6) is not a multiple of the number of attention heads (z).N)rr! ValueError)rIs rMvalidate_architecture&Gemma3TextConfig.validate_architecturewsS   d66 6! ;#D$4$4#5622327  <c URSS5nSS0SS0S.nURb UROUUlUbURSRU5 URRS5cSS0URS'URSR SURSUR S55 URRS5cSS0URS'URSR SURS UR S 55 UR 5 U$) N rope_scaling rope_typedefault)r@rArA rope_thetar;r@rope_local_base_freqr<)popr2updaterC setdefault default_thetastandardize_rope_params)rIrJrUdefault_rope_paramss rMconvert_rope_params_to_dict,Gemma3TextConfig.convert_rope_params_to_dictsIzz.$7 #.y!9*I6 8<7K7K7Wt33]p  #  !1 2 9 9, G    # #$4 5 =6A95MD !1 2 -.99 &**\43E3Eh3OP     # #$7 8 @9Di8PD !4 5 01<< &**%;T=O=OPW=XY $$& rS)rDr7r2r6)/__name__ __module__ __qualname____firstlineno____doc__ model_typekeys_to_ignore_at_inferencebase_model_tp_planbase_model_pp_planrint__annotations__rrrr!r#r%r'strr(r*floatr+r,rFr-r/listr0r1r2dictr3r4r5r6r7r8r9r:r]rHrQr`__static_attributes__ __classcell__rLs@rMr r !s,J#4"5%.%.%.%E%E%."+ )"+ &(9:#%568IJ!"_$56 JK!s!s    Hc0s0#*S*#u#L%It L#* +,L#S /D(, L#* $$#'OTD[' ND ,/sU{T)/!$3$!%NC$J%$(KcT!(,0UT\0+/EDL//44*X>M (rSr c^\rSrSr%SrSrSSSS.r\\S.r S r \\ \ \ 4-S -\S 'S r\\ \ \ 4-S -\S 'S r\S -\S 'Sr\S -\S'Sr\S -\S'Sr\S -\S'Sr\S -\S'Sr\S -\S'U4SjrSrU=r$) Gemma3Configa mm_tokens_per_image (`int`, *optional*, defaults to 256): The number of tokens per image embedding. boi_token_index (`int`, *optional*, defaults to 255999): The begin-of-image token index to wrap the image prompt. eoi_token_index (`int`, *optional*, defaults to 256000): The end-of-image token index to wrap the image prompt. Example: ```python >>> from transformers import Gemma3ForConditionalGeneration, Gemma3Config, SiglipVisionConfig, Gemma3TextConfig >>> # Initializing a Siglip-like vision config >>> vision_config = SiglipVisionConfig() >>> # Initializing a Gemma3 Text config >>> text_config = Gemma3TextConfig() >>> # Initializing a Gemma3 gemma-3-4b style configuration >>> configuration = Gemma3Config(vision_config, text_config) >>> # Initializing a model from the gemma-3-4b style configuration >>> model = Gemma3TextConfig(configuration) >>> # Accessing the model configuration >>> configuration = model.config ```gemma3image_token_indexboi_token_indexeoi_token_index)image_token_id boi_token_id eoi_token_id) text_config vision_configNr~rr$mm_tokens_per_imageiiir)r*Tr1c >URc%[5Ul[RS5 O9[ UR[ 5(a[S0URD6Ul[ UR [ 5(a[S0UR D6UlO1UR c$[5Ul[RS5 [TU]$"S0UD6 g)Nz@text_config is None, using default Gemma3TextConfig text config.zFvision_config is None, using default SiglipVisionConfig vision config.rB) r~r loggerinfo isinstancerprr rGrH)rIrJrLs rMrHGemma3Config.__post_init__s    #/1D  KKZ [ (($ / //C$2B2BCD  d(($ / /!3!Id6H6H!ID     '!3!5D  KK` a ''rS)rbrcrdrerfrg attribute_mapr r sub_configsr~rprmrrlrrrkryrzrxr*rnr1rFrHrqrrrss@rMrurus :J-))M (+K =AK!DcN2T9@@DM%S#X6=D&)t)")OS4Z)")OS4Z)$+sTz+&*ut|*'++ ( (rSruN)typingrhuggingface_hub.dataclassesrconfiguration_utilsrutilsrrsiglipr get_loggerrbrr ru__all__rBrSrMrs*.3,'   H %12w'w3wt12?(#?(3?(D - .rS