+h=SSKJr SSKJr SSKJr SSKJ r J r J r "SS\R5r "SS\R5r"S S \R5r"S S \R5r"S S\R5rg)N)FlaxTransformer2DModel)FlaxDownsample2DFlaxResnetBlock2DFlaxUpsample2Dc\rSrSr%Sr\\S'\\S'Sr\\S'Sr \\S'Sr \\S 'S r \ \S 'S r \ \S 'S r\ \S'S r\ \S'S r\ \S'\R$r\R&\S'Sr\\S'SrSSjrSrg)FlaxCrossAttnDownBlock2Da Cross Attention 2D Downsizing block - original architecture from Unet transformers: https://huggingface.co/papers/2103.06104 Parameters: in_channels (:obj:`int`): Input channels out_channels (:obj:`int`): Output channels dropout (:obj:`float`, *optional*, defaults to 0.0): Dropout rate num_layers (:obj:`int`, *optional*, defaults to 1): Number of attention blocks layers num_attention_heads (:obj:`int`, *optional*, defaults to 1): Number of attention heads of each spatial transformer block add_downsample (:obj:`bool`, *optional*, defaults to `True`): Whether to add downsampling layer before each final output use_memory_efficient_attention (`bool`, *optional*, defaults to `False`): enable memory efficient attention https://huggingface.co/papers/2112.05682 split_head_dim (`bool`, *optional*, defaults to `False`): Whether to split the head dimension into a new axis for the self-attention computation. In most cases, enabling this flag should speed up the computation for Stable Diffusion 2.x and Stable Diffusion XL. dtype (:obj:`jnp.dtype`, *optional*, defaults to jnp.float32): Parameters `dtype` in_channels out_channelsdropout num_layersnum_attention_headsTadd_downsampleFuse_linear_projectiononly_cross_attentionuse_memory_efficient_attentionsplit_head_dimdtypetransformer_layers_per_blockc /n/n[UR5HnUS:Xa URO URn[ UURUR UR S9nURU5 [URURURUR-URURURURURUR S9 nURU5 M XlX lUR"(a$[%URUR S9Ulgg)Nrr r dropout_probr r n_headsd_headdepthrrrrrr)rangerr r rrrappendrrrrrrrresnets attentionsrrdownsamplers_0)selfr#r$ir res_block attn_blocks d/home/james-whalen/.local/lib/python3.13/site-packages/diffusers/models/unets/unet_2d_blocks_flax.pysetupFlaxCrossAttnDownBlock2D.setup>s t'A./1f$**$:K:KK)'!..!\\jj I NN9 %/ --00((D,D,DD77&*&@&@%)%>%>/3/R/R#22jj J   j )-(0 $   "243D3DDJJ"WD  cSn[URUR5HupgU"XUS9nU"XUS9nXQ4- nM UR(aUR U5nXQ4- nX4$N deterministic)zipr#r$rr%)r& hidden_statestembencoder_hidden_statesr2 output_statesresnetattns r*__call__!FlaxCrossAttnDownBlock2D.__call__`su  doo>LF"=mTM UbcM - -M?    // >M - -M++r-)r$r%r#NT)__name__ __module__ __qualname____firstlineno____doc__int__annotations__rfloatrrrboolrrrrjnpfloat32rrr+r:__static_attributes__r0r-r*r r s4GUJ  ND"'4'!&$&+0"D0 ND {{E399"() #) XD ,r-r c\rSrSr%Sr\\S'\\S'Sr\\S'Sr \\S'S r \ \S '\ Rr\ R\S 'S rSS jrSrg)FlaxDownBlock2Doa& Flax 2D downsizing block Parameters: in_channels (:obj:`int`): Input channels out_channels (:obj:`int`): Output channels dropout (:obj:`float`, *optional*, defaults to 0.0): Dropout rate num_layers (:obj:`int`, *optional*, defaults to 1): Number of attention blocks layers add_downsample (:obj:`bool`, *optional*, defaults to `True`): Whether to add downsampling layer before each final output dtype (:obj:`jnp.dtype`, *optional*, defaults to jnp.float32): Parameters `dtype` r r r rrrTrrch/n[UR5H\nUS:Xa URO URn[ UURUR UR S9nURU5 M^ XlUR(a$[URUR S9Ul gg)Nrrr ) r!rr r rrrr"r#rrr%)r&r#r'r r(s r*r+FlaxDownBlock2D.setupst'A./1f$**$:K:KK)'!..!\\jj I NN9 %(   "243D3DDJJ"WD  r-cSnURHnU"XUS9nXA4- nM UR(aURU5nXA4- nX4$r/)r#rr%)r&r4r5r2r7r8s r*r:FlaxDownBlock2D.__call__s[ llF"=mTM - -M#    // >M - -M++r-)r%r#Nr<)r=r>r?r@rArBrCrrDrrrErFrGrr+r:rHr0r-r*rJrJosP$GUJND{{E399"X$ ,r-rJc\rSrSr%Sr\\S'\\S'\\S'Sr\\S'Sr \\S 'Sr \\S 'S r \ \S 'S r \ \S'S r\ \S'S r\ \S'S r\ \S'\R$r\R&\S'Sr\\S'SrSSjrSrg)FlaxCrossAttnUpBlock2Da Cross Attention 2D Upsampling block - original architecture from Unet transformers: https://huggingface.co/papers/2103.06104 Parameters: in_channels (:obj:`int`): Input channels out_channels (:obj:`int`): Output channels dropout (:obj:`float`, *optional*, defaults to 0.0): Dropout rate num_layers (:obj:`int`, *optional*, defaults to 1): Number of attention blocks layers num_attention_heads (:obj:`int`, *optional*, defaults to 1): Number of attention heads of each spatial transformer block add_upsample (:obj:`bool`, *optional*, defaults to `True`): Whether to add upsampling layer before each final output use_memory_efficient_attention (`bool`, *optional*, defaults to `False`): enable memory efficient attention https://huggingface.co/papers/2112.05682 split_head_dim (`bool`, *optional*, defaults to `False`): Whether to split the head dimension into a new axis for the self-attention computation. In most cases, enabling this flag should speed up the computation for Stable Diffusion 2.x and Stable Diffusion XL. dtype (:obj:`jnp.dtype`, *optional*, defaults to jnp.float32): Parameters `dtype` r r prev_output_channelr rrrrT add_upsampleFrrrrrrc /n/n[UR5GHnX0RS- :Xa URO URnUS:Xa URO URn[ XT-URUR URS9nURU5 [URURURUR-URURURURURURS9 nURU5 GM XlX lUR$(a$['URURS9Ulgg)Nrrrrr )r!rr r rSrrrr"rrrrrrrr#r$rTr upsamplers_0)r&r#r$r'res_skip_channelsresnet_in_channelsr(r)s r*r+FlaxCrossAttnUpBlock2D.setups8 t'A56//A:M5M 0 0TXTeTe =>!V!9!9IZIZ ).B!..!\\jj I NN9 %/ --00((D,D,DD77&*&@&@%)%>%>/3/R/R#22jj J   j )/(2 $    .t/@/@ SD  r-c[URUR5H3upgUSnUSSn[R"X4SS9nU"XUS9nU"XUS9nM5 UR (aUR U5nU$N)axisr1)r3r#r$rF concatenaterTrV) r&r4res_hidden_states_tupler5r6r2r8r9res_hidden_statess r*r:FlaxCrossAttnUpBlock2D.__call__s doo>LF 7 ; &=cr&B #OO],NUWXM"=mTM UbcM?    --mr?r@rArBrCrrDrrrTrErrrrrFrGrrr+r:rHr0r-r*rQrQs4GUJ  L$"'4'!&$&+0"D0 ND {{E399"() #)!TF r-rQc\rSrSr%Sr\\S'\\S'\\S'Sr\\S'Sr \\S 'S r \ \S '\ Rr\ R\S 'S rSSjrSrg) FlaxUpBlock2Dia| Flax 2D upsampling block Parameters: in_channels (:obj:`int`): Input channels out_channels (:obj:`int`): Output channels prev_output_channel (:obj:`int`): Output channels from the previous block dropout (:obj:`float`, *optional*, defaults to 0.0): Dropout rate num_layers (:obj:`int`, *optional*, defaults to 1): Number of attention blocks layers add_downsample (:obj:`bool`, *optional*, defaults to `True`): Whether to add downsampling layer before each final output dtype (:obj:`jnp.dtype`, *optional*, defaults to jnp.float32): Parameters `dtype` r r rSr rrrTrTrc/n[UR5HnX RS- :Xa URO URnUS:Xa URO URn[ XC-URUR URS9nURU5 M Xl UR(a$[URURS9Ul gg)Nrrrr ) r!rr r rSrrrr"r#rTrrV)r&r#r'rWrXr(s r*r+FlaxUpBlock2D.setup"st'A56//A:M5M 0 0TXTeTe =>!V!9!9IZIZ ).B!..!\\jj I NN9 %(    .t/@/@ SD  r-cURH*nUSnUSSn[R"X4SS9nU"XUS9nM, UR(aUR U5nU$r[)r#rFr^rTrV)r&r4r_r5r2r8r`s r*r:FlaxUpBlock2D.__call__6sillF 7 ; &=cr&B #OO],NUWXM"=mTM #    --mr?r@rArBrCrrDrrTrErFrGrr+r:rHr0r-r*rcrcsV(GUJL${{E399"T( r-rcc\rSrSr%Sr\\S'Sr\\S'Sr \\S'Sr \\S'S r \ \S 'S r \ \S 'S r\ \S '\R r\R"\S 'Sr\\S'SrSSjrSrg)FlaxUNetMidBlock2DCrossAttniEa  Cross Attention 2D Mid-level block - original architecture from Unet transformers: https://huggingface.co/papers/2103.06104 Parameters: in_channels (:obj:`int`): Input channels dropout (:obj:`float`, *optional*, defaults to 0.0): Dropout rate num_layers (:obj:`int`, *optional*, defaults to 1): Number of attention blocks layers num_attention_heads (:obj:`int`, *optional*, defaults to 1): Number of attention heads of each spatial transformer block use_memory_efficient_attention (`bool`, *optional*, defaults to `False`): enable memory efficient attention https://huggingface.co/papers/2112.05682 split_head_dim (`bool`, *optional*, defaults to `False`): Whether to split the head dimension into a new axis for the self-attention computation. In most cases, enabling this flag should speed up the computation for Stable Diffusion 2.x and Stable Diffusion XL. dtype (:obj:`jnp.dtype`, *optional*, defaults to jnp.float32): Parameters `dtype` r r rrrrFrrrrrc H[URURURURS9/n/n[ UR 5Hn[ URURURUR-URURURURURS9nURU5 [URURURURS9nURU5 M Xl X lg)Nr)r rrrrrrr)rr rrr!rrrrrrrr"r#r$)r&r#r$_r)r(s r*r+!FlaxUNetMidBlock2DCrossAttn.setupfs  ,,!--!\\jj    t'A/ ,,00''4+C+CC77&*&@&@/3/R/R#22jj J   j )) ,,!--!\\jj I NN9 %'(* $r-cURS"X5n[URURSS5HupVU"XUS9nU"XUS9nM U$)Nrrr1)r#r3r$)r&r4r5r6r2r9r8s r*r:$FlaxUNetMidBlock2DCrossAttn.__call__sX Q < ab1ABLD UbcM"=mTMCr-)r$r#Nr<)r=r>r?r@rArBrCrrDrrrrErrrFrGrrr+r:rHr0r-r*ririEsv,GUJ  "'4'+0"D0 ND {{E399"() #)#%Jr-ri) flax.linenlinennn jax.numpynumpyrFattention_flaxr resnet_flaxrrrModuler rJrQrcrir0r-r*rwsp3MMV,ryyV,r7,bii7,tYRYYYx=BII=@L"))Lr-