From e522e8f97e670471a00779ac8506253c5c8b3ac2 Mon Sep 17 00:00:00 2001
From: Kevin Musgrave <tkm45@cornell.edu>
Date: Wed, 24 Jul 2024 21:15:55 +0200
Subject: [PATCH 01/16] Update setup.py

---
 setup.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/setup.py b/setup.py
index 2e8a8dee..0818062c 100644
--- a/setup.py
+++ b/setup.py
@@ -40,7 +40,7 @@
     python_requires=">=3.0",
     install_requires=[
         "numpy < 2.0",
-        "scikit-learn",
+        "scikit-learn <= 1.4.1.post1",
         "tqdm",
         "torch >= 1.6.0",
     ],

From 2ab5840b5f677c20c461fa8c6c0b58f27f2cf36e Mon Sep 17 00:00:00 2001
From: Kevin Musgrave <tkm45@cornell.edu>
Date: Wed, 24 Jul 2024 21:16:57 +0200
Subject: [PATCH 02/16] Update __init__.py

---
 src/pytorch_metric_learning/__init__.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/pytorch_metric_learning/__init__.py b/src/pytorch_metric_learning/__init__.py
index e5e59e38..ff4a5f40 100644
--- a/src/pytorch_metric_learning/__init__.py
+++ b/src/pytorch_metric_learning/__init__.py
@@ -1 +1 @@
-__version__ = "2.6.0"
+__version__ = "2.6.1.dev0"

From 64d8d2eaaed54a5482f169817737d1be4c344f50 Mon Sep 17 00:00:00 2001
From: Kevin Musgrave <tkm45@cornell.edu>
Date: Thu, 25 Jul 2024 03:26:32 +0200
Subject: [PATCH 03/16] Update __init__.py

---
 src/pytorch_metric_learning/__init__.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/pytorch_metric_learning/__init__.py b/src/pytorch_metric_learning/__init__.py
index ff4a5f40..fab833f3 100644
--- a/src/pytorch_metric_learning/__init__.py
+++ b/src/pytorch_metric_learning/__init__.py
@@ -1 +1 @@
-__version__ = "2.6.1.dev0"
+__version__ = "2.6.1"

From f21a8315d8bcfdff034e360c9633aa7535040a9a Mon Sep 17 00:00:00 2001
From: ir2718 <irep2718@gmail.com>
Date: Wed, 16 Oct 2024 23:29:36 +0200
Subject: [PATCH 04/16] implement tcm loss

---
 docs/imgs/tcm_loss_equation.png               | Bin 0 -> 39338 bytes
 docs/losses.md                                |  29 +++++++++
 .../losses/__init__.py                        |   1 +
 .../losses/tcm_loss.py                        |  60 ++++++++++++++++++
 tests/losses/test_tcm_loss.py                 |  46 ++++++++++++++
 5 files changed, 136 insertions(+)
 create mode 100644 docs/imgs/tcm_loss_equation.png
 create mode 100644 src/pytorch_metric_learning/losses/tcm_loss.py
 create mode 100644 tests/losses/test_tcm_loss.py

diff --git a/docs/imgs/tcm_loss_equation.png b/docs/imgs/tcm_loss_equation.png
new file mode 100644
index 0000000000000000000000000000000000000000..b8e62954021bd84dda90631bd981db333ffeb674
GIT binary patch
literal 39338
zcmeFZ^;?x+^ewtED3J!~1`!15ZZ;_`C|y#5bhk>kh#(-SfGCa9DFV_hB@NPD(sk$l
ze9w9AIltXM;O@sq1$MmeT5HZZ#+YLgp`oU58=C?ffk50=QbcMY5E$g}{~ws>@b_xh
z_X_yuhO3N{HYPm$FfGF2YjQU^UAIS$R&JhVE|v&u2S<BL4%eqHmX;2#HjZxF*Bd0@
zNkXV6$+}pYxjl1qpwoV4Z;5bqd1h;AVRoO6TY!`9`7>8HM<+82GddnF9zISw9v%^X
zZV_$)I$60#+B+e1xCjItLJ29Y?UlAM<!z{aa4ESnn!{(qHtS-&_vMpbP%fH2PjYMW
zJ>98W9{f&9T<f>o>tA{VO6P0)WaJz5lYW!aeE5y;n;h38r+ZE}d2g~?Y9u&YawST?
z?Ayk3v#@v{|C0B5@$!6QvvAw~N!ro3ZsG0yF$)<?G=~4{!R!871o3};r{uwN3#R|;
zJ5k@_<HP^Y$NeUG`Ty$~C76`t5&!2i+8O@;H~Ig(COI*Kmhy~3LTJ_Kxs3GxYhg1h
zUd2Qs;bEVOHl~E7TgrmgaiY~GgZdWgVSn>~mVBJ=x@peE1A_nc(A@tHo0Sv8djEcK
zRMagCo|LJ8Dc|jStw72SwN*C}v6F>__TJtk5n=|V&)L~#BL#Y_?Cc5fB}1mj<03;W
z1qFqU{g@@ci~WyRZ%-!_JvKTM*~4RFTQlXNKI4Wc@^tHkkFyBy5E7IVD8$i(7#F`2
ziMG1O%KCG<-ZvXJXhS6+K!V0=7L7{WhuGNISSw7H3Cq#Z@%dB@iETLryd+VC3ht4p
z;e78F8ynka+$=q<M~|ErJ4tez=|tTw&%L851#xKu&$-Xf&p+da{jX0<!Ix2_zCl31
zKa<G*WOjZ&$p7pxWn=r#cg8nw-Z-^Ud+J`K$zW>h=wy_X5TXa6Sz21cjWv&zT5D=+
z2Qw<f&aMn(Re9}Zj=BHXX}YCb=k<N~X?Df4z7OJLf(}$^X=&eFm(^H_v2$fhq<uc0
zvU6Kz@ivvXDgIWZ#w&mO_pf=*u&vw19|D7l=Z)e=lZ@_)unP#&4IGcTrke>D{-^(>
z&CMA^JT`0&TpvGqLg|{=_3nXe@n}{-0nS&obe^Sq`h_Ezq0PlVPO*vZs+IW^40)uB
zdg4B9i<Og>Mia0f4-TX8QCxc+&eW0?lCx>X$Vi|`xi{h1VD{&mnLEWnYYb&r45gq9
zOM(*N6L>|1{Y0q!M8#YfZNSgf!LO8J6;Fzdg}nEesHmv)xn)WeLPA2kb{a7>b#y{t
zISGO=@RCHBC|l%P4=uK*>vs+g<lyGx6B2?X{7)6B3E>1G=$M&rKKiOwSQ>bFe&Tg7
z{De6iv&!dy?Qml(2tLc~>~Ql>iRG=%t}aS32`(=DXV0D$lF^TqK77Y1oi$(z@26XC
zOY%kiBPT46z)Rs<niEZ{S`p#nuWPq|P}2JCNT*-C7qD+?3c1A`p40kCItZVXG_+6h
zit4Uf+Rt|;0TBu`-k+!C`%10*7<hRfy<+Y7kS6SkPQ;=j_3YU_IIYd-@|kv_c(kz#
z3l>zO9yoF0hq_sK%8-5Z2jAX%#O1owEpqnM+<a$HT~Z03V0LyEK}$=El#`o-m2Bxq
zxcl9r;|=;YN5k2cXqu>JQDPR8FH^9TlM@%}9-kO{hun@vmUCIt+Zz07S40lAb|-UZ
zP17_L*B?@NY>qQaBfr4<ay|Ra^nAPw>*mdyUdMA`>r|yKi^^CGjJ}#iMlrA%viZXd
zVq$6QsrXYCzdtJ_OZbcI?eFhSy7nni$Eu1t%_5Q>n8eD4;VMyM6KK}%O#5yRr%)8b
z(S_A7k2qLNPfy=14!pR9iV_}PUT)K$h=i*K*a?Ug=E&HMj#oEt?QB-g{QYsZC99$m
zJv4_uMzwT*>1nS_MeJZUM5lD%<;h(!aq;#0pT*(T3yX^xzDHYxa*rD<yOQQ+t}X?Y
zA3j7wxKvXyGc!9mJI4$0l*oPm{@rUa@kxKBqY0Im7yeX@djcGpwpFbO#(Lv8dfCv~
zILXVOktFO~Pg`*yQtrLuG-_HeZir=mnDi+x4?FN;-NgCtFRa{V1_s+ZdwYAI9_OzY
zgb2ZM`JF7jBb3u=fT(ynUFRJh7pI`3LxM2<6;AM{+V$(N_;`+q?@uxKC~hN6mwVI6
zB>bw%&<Ms@m+r^FO&z{sgbFkM$ziI7o|!qjw<Z)aiJX^L6;hLbWA*FD*2{~txuqp^
zL{3i5&pRf5^!6GS3AnTp(SkF6M2H;-l9HG4IcUATy}I8_(GXCO2}RP?{)Rtqu6A9<
zzjG(-<#miyF|MT{Pb9Lit*h(ScHO>O`i0$SQHuhtKa(`__WJMaTcn(Z;nC4J?Tapm
zpG8Ks8oIiy&K4H=kj8epdHYZc5Fa1^+?2`ohHKI(w>7<QGZe~90b97oWXZsQ^ox7<
z?kTFNIb6I*MJ@63>AJD1S;4O=*{5P>8>MjRFZ2sr=i?+f-Gz_>_NV_=$@pyWVVicF
zZaq+>j%DiX?0gOJS~=}=3o=@;OvcZy2MygAWJn}oRCxH$tT+jM>p%;O&#Gz|f)0~H
zs3mVzOUr*;_(q7KL}!1wuWhb5Y@pokN!|Hg7i4CO{p3<O>SwUQu((|GHz4JHdp5|D
za8WrEAbPkpbt_SsS<~V+JEV6=pUzv8!smw*j$}ekpN8n;V-(fZ+kQn5p&`i0$#;&9
z-U%`IHo{RUEw>vr-&^QFyhux<f~*<AXe&J9w}1QcsNpID)(BFwP{4VWGsV^yArX<x
zy1Kd$Mka>L@!mAy>v(v0Z{NO6-k=q6V})?L3FRn)!R`<2=*!tz(|?;2ck~NMc~Vr?
zUbly)iFp%V$G}B;dkbe~X2K@kg{0#FNynA6#F454f;Y6of~UZ+nlSLB%S7eSq{a=a
zv9S@B`88`v=xqwVb|E8lU(7-G;o)Jgi-S=##L-NkBqdoy_rG66>Yq#+0uV3W8T%;x
zwb37e#1WsK9<w=7IlufduqdIxb9;J<g;tk{p!`X_?@=oh$#uRVY9S}YgCED4a45Gr
z?tVmSXyC_Ei-vA&JkM>Wr{oR1^n3g6T^1xufiqDRi>Tz%$)9N<8w99IJ}#846|u0e
zAkyIr1r+XwDKgKw)T*ZmYE6(%{^OU|x3F@^TaIDF5ieAW*vIv8#^QIhMW7t$?JT~!
zN6C&zYx>Qd^fp0yX$7b9r-e;+z`R?Rh}+7GuO9;_5a0aIJT}J5AN~#3kIl+r^f_3&
znXQGDs2l=C2NU%<P~Ts_e;);h#MMoVAU?)>AzopvM@iy*M;3sB629i(Qr|~B3=9m9
za3&&=bli`Adsr`0(G#HTKcpfkchqOXsA(J<Xgcrv{XoH*b=bMx(W<OlFS+!!|GZl+
z61n*+>cNljtBdVG_rt*b9zlWXim8KDo(jjASj{YXC2CxyjqhzV&7rp`XIEA*)73EU
z-o2YE`zDxPkYep86Y<(@(!^SBTRq(B$v1CeHBRr+iHZ|Dkfz5!@7SGh8)yg&gxWth
zX8rNMcy(8NLoVW$^SxQdKFudQTu~%3F_Dblme^x!@>$7DOUuo(qiuzmAO6gd1!hoc
zP}`R=xZ0on@D6IDzB$dw%wraGcdHzqoG77gPUJJX9ezqt$Pc-`c}(j*o6pPgS=B^o
zJl554+{P&XgF{45iVR=I#o--pPFR;TKrQ~$5GV;T%m@K=cD~=oZPAYJ<>jSFjjgn@
z(2=k*T1=alm&fJ3XW_Toaz`u7!n%ru5$(s1AAc(BDQi47Bj3CsblV))S$oaY7@Qv7
zE<pW1h^oqI?p4R-K$Zd{JNqqaQIEH*C8z1)zN9n$N7UyR7a1C?fSR8FgxU)E2@_5g
zpp8G1+xO$eO?$P1P?xLrP8m6*Ht8{8uc)e8=$hbhxS^-=QGy!cfKYA(62O=5FXO+N
zw_wAPq9HDK;{p|_ZpZtla+zI2%>HeOdY*Haf#E$bR23#>=HamhI9hcVhZBUzp(^8g
zA9^OH;M>tR5bYfu(q?A#2s$pV*HBX{!-a3f`<tUsji}4fl;?c?(R6;PB;13uvoj$Q
zIaIdT_3k>tf`_~{nw-&@4|%xi(}er8${3%3Af6+I#}ZpcR(5^kqhz22AfJ&_G=h|8
zV<ncYqX3{1xd<7~TGK+>CH_Od=~%aKw+!Vzf@cdl(zv*|FolQbm|@`2{Bmr#Ku3`A
zSYTf6y=zdUB8d0j+u7lYt!VG*X*t<nW{w3ClT!p~fc9w8HLl}FULKpudx4ly&Jw-#
z(P9jZ!FyU+1B#S5@zhY0EI$5+5OAS>n*Dr%#&X8S4ydt{_-$gM5;2LUQMqrS0k7HU
z`oW)C&tkcT?Zw5#<AYHXKwQ@gA4hd|Do6F+3;qeG_q*?r17MNf->-A2xyW$(ZOfq|
z-i?fmj2^sU&~<S%gZF94bGMB)u*`M}K<9|l*=8jpKR-$3l*jvyOW0mh07iszazo!<
zy}E{iOH!&|+hHflJH12jg?tb1<5CH|fGj0bJX+c5J&_=Jd3+re0CI96W0t7$0!hVm
zw8*G`)eT=p*R41IN%7xNlR!C0M4To+o<D21zvfEHaQ0s@^T;u%bWl`MdKnZH)Nu*0
z7*)C8v$daW%F91^P;52&b#Rr-tm($Tm4WN1e43V?&h4_GxG`N{+sFfp7$%?fvb2;}
zn~g+1mMQ|ucJsgh!F*esf`^AkkR>}i`*+)6HZNb_c_=gr4FMNi=D#ritqpfoC8;YZ
zDV5IL87p>O?lqfh3L)b*zaiqe6%FqpTln}XY{AN8HLsYM*f#)U`FujSlmgwyse`?6
zehLAKG>LB4J+cn;@ex7gG%Nrp$>sZrU!Vw+XJlkRdBgBI+KM^bs>MTHwxOY6f32ru
zaA@dX_z?$<f<cAdC?^gDAExKDHwsrd!zMh1@CL+fQQ!7NTboWof(9NdW2JXPajCI^
z^GrIoQLl^?Qp(84XlQE>&7S=85huN`i-4@lX;dSXt<?xX!>HCn4sLU2#K5u9l&frL
zqFfj#)&|~xMIo1fQQj1wIHXLd<;TbKam=U#I<C|*+Zgm-XS5~iPR4+Z)82y8^T{d#
z>%I>yfPn!bTnEhie5|Bp;drXn^XX8IW-|9v^vS8I-6_u*>HOip=O_EemuK7DHvM!c
z7>g%)&UJY<b2TD71NR<>!s@WbE%2h(@$oMqz~&&DbZb30P~`;pmrS<S96ZTx#Fz1`
z47RJPww6&ugaTEJVR_5+IpFq!d98a*2fwJ2Nd|~DH8;;KE)H+3zOt1(9{=I*>)H$%
z#FiT<hu6OWh4qa<$binXKha=U@;>tKC$<6zF#Y>0!V8MfH~00&b$gxce<~em0U`uJ
z`J4lI(hMBJd8MB|N1Kg_g{4CGzWo^R2%H%CAb_cKl9IFySC^uwTRvI-`1E*Z4h}a<
z+t<&ZVcE@g|F&S_;2_o1aNR_xFOTQ(3iQjdA(j4uI>7C=q5)*Jy!T|<|H!DSrpA0Y
zPsi)>WcmBkHe6J2Da2AWqCOiIz5Bf&9?+`L$NBcz+po%!OYz6(rJ>Y4c6ToaR+)w{
z-5Z9Qg1XGy<}8n*oSgQBqdj<WPoCq}WOXKNm`wgK#SyiH-yhuunGAfNdSgubUr$W@
znW?C$Tj1cyL2|4=0OC&nuIna(KA{8o_;J$i)pX#M|J>{>-BNElJCLz;j*sc-_AXmc
zyam@U_zSc@A9vjRgm0#&N~2O}(fI1-wHGT>wUx&Jlqtk3a&i<|6P3BGy2*eVx_2#~
z1CNMqa7`C+x83n9<)GxTP)xU@LCmeJu)5nIk!}ftU)67AYw0cX`u2a4{`))op;njz
zQ!suL)QjWVlODmDvUcIfsHher&nad^xL&ChR#0RA?05JvgNTSekY-K4tBWHJLUjWj
zBO?wVZ=8O9en81WoUh=`VkDnRN=nXOuB_OgNd<?5L;yn=HL&*asrmjqjz&CM*aV)5
ziG@V~Y=o#QiNlWTOL8*V(-TOX2CM1g0!#P1_1Z6;m{5g<C7}Zq{QtJ5CM>R?o)8cd
z4;6h`pqtnSl0sp^PK@ogKC+p#UgNdPki=;e3AmD=sHmv*qTJ-8h&%RsLC4lqemi_(
z;$XNM3rouk0D%W1xf-mq%gY#_tm?z%qDavYC>j8?!F+SP+zTohER?RJw~2|#(<v0u
zPx@Nn2$zat&u9qtCq=<=aq3xZM_W^ryjI;(j~|mBZqEdEE!Xe&ipCT}zMD{Wv|(T<
zQAfeZJe_>HgbtO%JH*65QM591!W~eA2993z1VmTasc?~TedN!dQdGh&HzXt^P#FPE
z2BnxWOM)3BBUJK1osImq!6Zv->!-l;QCFxs5>lAX8vvVtss>hlA7Va#X7YmzR8&@$
z0kXATxDVwe3<M*@EGjkp8&pRTS)~LPGgxp`;92#ijhU*z4R8q`eNla%dM02`^Z(Wa
zwOnoUmPiu!H%D-hv%oDYk(vaJ7E_Vy5o3k=l@8QU8FzspC2u@nXXhOY*sebUqOO7p
zpZ%pC*|u0}%0f@b->su#V`HFpTsQV!Btfy4V@OKV7KRsRM}j}5p^6GShHOk`KYX_X
zOd1VA<GaCqe0C=2bKV$)F}JnlkoM*Xz;4Z=t0?daKwzz#ts|kJsFmD|6WhZy_F2sq
zIfLT&^gk{NCp+kIeh7Y3)zoBKwgeUyR(n_1b;NzWlFw?>0j0vOOQ!9w>4p4H-AcD8
zC$b>zHFb66t6}^c7tX@JMo&+#TjO@`_wV0aKzQZXH~xIjQes?8;xO15w<|V-MFpA7
zp~S2PPS8sKT<iSV;cH$VYx)w&|CCXcWG*z2DsoG&g774y;H-nRmBK;<{In;}Zm}~F
zvWH>1X071e`>-O%6El}qRF)%e0glYCL;f=Zc7-}$J+`JdOy6+Ph<WuF@L0?|F3^1m
zSr?IWvSHr|smckM_`C;*GVk3l$Hjxj*ptUg&xO>5Y7`hv)}+F7lmHG1$VZcDT3=mV
z%v@zuRq4J854W96d&nnfU9D$n$xKK{h(Of)95ew4nOWJDy!e-}>-Fj6C}y(*qU*cI
zrar)<!rrT?bJuCuwxs)d*J9+KKAtLfZ3WVE4ur2|Z8%?5q(I;^h#f4qyEd7vEdV*0
zppX+1@#4=PT^ve*g5FC=8YyD4EVdlw3Bq^pel;k9OnpO7Pp>ss)vmVMvp7GW9hTP0
zW2*eC!Zq73iawj=qxT~>YsMe#X-w$U&dbXVvJHP3U5_dD{kQTGB_io9KVeI39o-;N
z*-R<Yh{8@M70JC$pUEODB!p7Ape_oI6_2i?0GsyX$I8Uy<(}o+*$Xb-S%b;bkx-TA
z0Hh=OX9;o2hkrQUc=72IU4H9el0Q%vl#Hk+JpWlKA(}TKIm@G0*Vs7r+c&OgYSEE`
zlCSC?1)hC={y>-!6(Ijf<<~S^OH<#xc?Ft5>V{<Cr5J>emCMtILlnIqB^~{y(u^A+
z^@QSQP6s=$j})>@+`7a03$n5U-0x^io$(J)g{%?_ArovjTsce%e7(C~=zN39?&_gd
zUj3K9pbWkjyM*V5zI%74HI_OGG>U}WAO|^6XQGQ?|0fD+x^;AQ!-0<j0j=yj0T8iC
z&Si=qBqCbh=mVgxqN1{nfxNN4$bLK5f5r<A4lyw?pAWF~c`o~LJ{)Qh;s*`qC1a(f
z_$Yzk-|8TSY$y)*m<gl_AjOoGi!Q<_exG8r4f2=j7X!$R?9)E$9Hm`XdtDD$COF>l
zjwg-u>Y?(NxbNZY&zIM8ii#X|_X*1R|BHw$->vGe(=BtAAZr2?e~3>|ZvF0oiNFts
zUd2|EzzdS2Y2OE1Kh7KyE}*agM|$r(0YvWVxJ|OV=xAfC6eMpxhkR|eXj;kXjXhD%
zExuLnnKv=fsXK}{LXZqYr6Fq4R{7~=0f04*1FyEP07BlnO|!kc90po4N=D7`%oBPz
zOe3y$=2*ZC3pGEWK-&U5iVN^74wVqTu&^+Pk6Mi!I)@8ysiw_q5qyTsP}Ou162On{
zhi$E`4D9Ur2iG$_`t(nfWgQ$glDW)evW~QHadD3~O8a*9`y}TjuTIo!$q%e<H(Z{w
zU%P&Ned7uMil)B4?d0M9zASLko+3|3h%6I;ol0#eCP2C!u&Ax)<EgOfbvfQKLv00~
z*bZW=c3UEvN|+Fio5O%>h0>{?h1HU4(rb^-x}rm%Qx3tw!Lw)E4Q;3ZJZu1dMkeZU
zUp|Hc1f8L2UG`^z5JEeh?8QT4fv!zC12!rniqAViuAmF%axf5ySp-mMT5>TPgr~1P
zO{jb&^!c<>_n^n+^xrcmjOF7wHK%T?zi&cV+q<Mw+EoE?R-Il46#*r-nV%o;qWJcZ
zEk6)?s}b$I%1nNuL)ceFM#knUf51*RFfg(J4`nv++e)9xe7@U4$Kr7N@EuxSju>(4
zcfo4}V@0szc&L=q@thF093`R#T%K4LNh=iQWmA$nZZYW@8b-%6DwyxgHlmOsv<0pM
z?5?V=*1MRzE$p^}kJ5A{ug*({d%bz@+_@umJR9T$*i1hi$g)woUiynvequXb&(y(n
zTTk1Y&qb}vm<v-%#V-C03i%y>bM}Qao-FFgDFGaZRNT=ZfjqY{n8flmyB`4sQP7IU
z?Ma~ot=dLZH!P2lo137zx?0HRUyid0D4Ww>^S1$l%Qdy<i07uOOxB^tF;lEH`nom=
z0rHAks~0x&TgeE!&(?qv+%}hr%B_o=nwnHb?!8qG!=-itd`OzImRHf=-!CHG|G?x&
z^uK>@^GDL0@0v$P?*P;lyM%y~1^H+5ynJ;3*=n#q)Kd3DfU!8R+mzaDI#D0qr+(^F
zur<Une={=gM2=|B*tsodC<GE$dxiVvI0?v+JG;A$P}-u;)IW$t;7J@K9P4&Ng9kKy
zpBDOM)1!#N>EUJ+@U!<Dl3qJ>xHMwnKsv4?PLH<ZQ1}24dDVQZXaS9aeU^L-&uY#+
zk_QhSY)n-4*&et<E}Om>=_LoP*4|?*>dpfOLBY|gLap5XViU=zUdLj6cI#@t6PNm|
zp!Ke&InJr=H-Z-wp0->T>Yh%L-H+LoS9HTMZ@L)!^9T38Cgj@6biL@~V&fQ)Kjz_J
zBM>Pnp#a5{smROkLYTFVmeVrmMZJxQ&d$#MQ}0{p><cw~W$<g9+_EDo+?>ykoq=kr
zm5>2*#8Fr8BVBzNP-^|rpA=df&45NJ#lB=_UYiNL5=V7p8ZI^+QK3uWw_}z&zRoam
zNoj|5AinUX!$O0Vn8U}|Hkk8H$97$WV1hvAA38d^uzVAtrJg969DOYI3TsdYlg*k#
z=>-HxQL>=33FPASbmhbx=5n_mzjGhm^#c+KVfj3(G=P|d1Vw65pyF|%en<l-Wsl<)
zYsh)6OkrQ4<d{dLXWavUfQgN5ySqPGS0`U=T)$Q@_9wZdgeSqBUxzJG?x+z}Gncug
zdQ#tS%GoX6H>z=a85I@PahfM};f=%JJHSiawf~CifhhTQ-8st}&#hmP)p6#F?`beG
zKZbd0zT9~}c3H<nj}o*r>+9<4xq;ozm*qPCIANg)Ja>-BV#r-=^Oi~9h?{H@D*Na4
zZak{~U1FkM`PM06E1<CVMwgIP+xz+i9cpW9WplMtjRG!C4|4MIcpM~8SF)_$+1mjf
z#wR3Xa3?F2XGx%kutD|C{QSiJ`S=I5Vbkw`Mluglpe#UTXs7+z*dM(#sDb?T-pJ%B
z3K%~B^R00q!@$DAA}WqFy9Q_cIa3Rgk<<OInhxRxkv7^~ctnIeSE)Zki7d$Rii(OT
zjb#9}+K~A-t%C8!Vw1X`6aj%f`)1ii+WkSiR_$)vd!s5RG<C7v8$T`%4UW$$yY&GP
zVsU5{bBs3xHgx2u0>F>5A{+buG_0d0HOa%)sL~<5!^5|e+<@2Gdp<@GpPc;JWKxF(
z6-i)}MM>|9<dQJcqC{Boq{Xk6Uv$r&O8lN@VGy4t37F#!kH8HXKUcVGakog|$(R4_
z1*js!CLkELgWc9GwYrU%jkx>ar(JP9i!*YFHHZdXw(Fz(lN`AawF$N;?@=|Dg7mVP
zG%i~{iYUTXRFU``uaPyN>vjos`nSay$j@atVq=x-5$kaGyYQvy+E8wc1liONe^Ds=
zg)Up;F{$OW(dEGyG+u~WkIkk6y;8ukNWat7uPD-B@Gs4xQzJRH%2b{8EUqz)MASL2
z35!PhyCQpk2A6m&XnGdViCE{DCw>D<$w?D%Y!<<o8A@*2Un!Hx@DkS8mJ($A3Wurx
z;8fxh<3<3xHazJT_BuK`UP~#~8k(8|=DTHgMn0UhzDe7KfYxmYs7KZHY$J~1fmGp#
zNaT$lr)%t}wzbz<?$_6ei6K9K{v;K*_Wh(C#z6n*mXR@b<~41nJl59M7T&q}FTpHg
zSYzWA-bOaX|6xpbh^sl277r;o6o>p33HxJRcU3N1i68;W4(E)WBe$T^ES!>F&Xq|$
z^pq^ZtyEJai{;D^evX{WCqjq#>f1v$zM#h1`q^)OHtG!4EFPT)sKDibN5+L{4#TTl
zIUf=LnU(~lp+jpR79_EAAXCQ+3F<pgJV%5D?%nHPTQ0b$)CO3Q4GQpVg1ST_aQqP9
zDygGy9%uHc%ej3LeH+B`cai^0ljX@zQ6>UCIgy30@{IlIx3ci#FO;FvV&GVR^OIHK
z{2gO2>A&IaZ*fmnvBGgiej2UfDCk+cd9323bQAPC*xm?jqp$CGW5Muu;1f%~h@afK
zfhUm5yq|a~w)68ks{I8)^%4-~EPQ?w0H-;)WS2=eLN6iAzqHIhTQ5lGJv}{5HSz!q
z!gAXl6vpSf*|^V@?3VAt|9=5RMzQRu0JR}L>ps(8*>_}~ybk?ChbN?y)-;mcXQmP;
zm?KN{x}&vK;AITB7%Wht0a$E)0sw%_jaOjO@I%L8tzD(&Ig=q}BH;%OOa)aYLM>j9
zuEOfl?prXm(e<l_{cGoTa=nF=DWZ8nGd6dLX;K_kJ8rb*X<9MPSj%ZxWjb8rs7A)2
z%9L74GhgGruJR>rfmObyNd2QkGQS-u>NbD<Lb|Ou%|E{A&yan*8gDFWwmnhlSh~(Q
zk=dmx|G9zkt?{Fs-qU8*gxaPD=E5#zA`2{Gvh%I2attMoP==eL$$9C5e})q<7K~tp
z6*RtR8d9{GoSfz7Mo;k(ywS=;FV2nu4XE}GWjUG86zwt1zoZ>B!oJGhLWC$`9tL8Z
zC4TUzG=B1MiCxv%$>|032cXw04&eG{S6A1<)&Aa|KL3Y`rKLrFl==uxjnR!gfW58I
zR*9KlOYF>C<k!;DQmL3Q=xtBS-gI`uL%P{YxXI~-<L!pZ1YnCk>23_ZRQ5ubXB405
z^+|b+RRo<cq`V9!3jYDhUIB{)$iqQs*U=TJNsY3)qUB9@ME^X~pl9XAz?!oiQhGE*
zaf}hh$jM3R+}N0$wW|=o_4>zOF1Fd*f#nw#TmO;o3ah@^(~+lTd}foVsyZ<>HI*4M
zCwaEPo3IJ}f??sM$!gc*Wy!0EIEhnC6uGK(C=r*zL<wuaU{_`uBtfgP8>wNKo125~
zpU@@vOUTxt`11heD*c-~#aMGVLpH{Hy=<%_JR(+lt@lxmo{MJ4iyg9GNQ{Z8>k|gn
zQVwMlFO~F@2WiJO9;2JyzLw5wrZ$+LWgzr>fBut$)mx<7I>^}s`1mgX^bvVZI%5Do
zPxUuqV&^7i5az?5v%_s36TJ2;#rYsM^@-gz0t)i5rXi;rp=!bMchLLJv;XL#3*EFe
zm7Jot8~X7s50oJQJMX8YD|2!%L48K4Cs4h**Sd2wSigDiTm4%f#k=#s_y%IW!OaeI
z9Y>d$=Sp9-K3RFkfQU{}$w7Y8d-XRnG|<Q^Nw9AMH#1nS0v7i9%NKJHAxT@T@RCc*
zjA9)qkSxm#Vw`t!xd+0PrXw%bXBvMUU^yM+>+wpIiQga3d8m-XLbWPl7rWpVD#A&b
zGUxEPU{LITnA6Q$q-1yQm{d8L)u!&xjI8oBXJ8pdHunZ7cB?5=Srf$}6Dg3^Uu2h8
zgId^&HeGdO=SP0W>#J{CWFDSI^dx@#XmZ!A%mTX&EFM!XO}c=p3k@nMD#gXcp=tG6
zq^;1no&vC3Gql1jJf<w(gBlb8iIl9eFI_Y&IQT|s=jEnjAg=(=AxM_J+PamQ9zo(n
z9=_I6W>(kCGv~iqU6p^#Pr!JU`4pdsJp53QF^}=B?QK5&LOna;^}W%j*59vvNc9of
z7<yD$S}WW}p!qO2FRux->>Xe*^A!^g%@E%GU)5>BSb*SKU~DuA5;nN45n@Rv+R+fv
z_fVk<<1XXir=aET>>NFJHQe%s@d7p4iiIWLTsQ$%{5+RHLILV>3*Z9*MY)jp^JTL@
z@1g-q3Q9UhIE{jO=|Xh_xJ}A)1CjlnrJnby6&xMpqlE@xAi0oNepN|%rIx-X(PUC)
zL!g!>WTt6&_)7TU#Bz={Bi}vB4CfQYm?&wL-#%2+Rb;$^EcD}oWnaE)&Rcno-YgZa
z1%Hf`V~@PNg0;<L_NrMYiZon9pa7s3fCmIZUHpJ}sXK+y8F|>&zw$^`L1_B#R6M;T
z?#q`CovUO*!rgc&$YQ=~GqP}Eb`Xtz!g-O<I#c9>QCnS2ZbA#7AKH(N_LWm38$Rzv
zb=G>YvY6MozkuG2X*L-ErE^W(;b3hT#Q_z#$E!i#Na5qUCS_)pF}7%TRWvl}JEjG4
zG=GXqB9^b+fb9G9nC?kKVq5qAW^!dg!K&K)x$WZ6-+%3bXijpS!y}ZMPMi7hOc$^n
zZz#0Uvls@Eek2(FBxmJ9$g_Xo<o2XllB??SneJ`<1uH8CQBhib@!1@=pxSZTmzN)I
z*I7fDueNvh$o+VX_mkY=1BK1CiwuGN_Y0)SFNUP{r{%43cv(K%;U|2+4&0B#j}Okw
z%8H4(OB--Z57IGHZ2L4B#pa7Sulxw?e~NeqZE1y;PJ0Hi-7J)3j!w5!rQDk6>gs3_
zVw<+DiR(Ux8xi0enFG#r191**kDk==-RyV9JmfEJxX3gN4WAzt+$-Ov9y6=*hRUa}
z6#xsxGI6jN4Niq);Mw>@L{DL}UkFDbaZvZeC@4r9A|gJ!f@AIJ>E(0UY%(88dKT<)
zh;%X(d6ji5{sVDPauxrd2KRd(C`*aCr|H9EBc!XYIQRtCO5)SfVhuTFmsaC`YFzn<
zno+o&wD)$)dj!zCp)0UXP0r?GAwSi;bfs`>?@>|kaG);}xjwsru3nCQ`O>y)k4r&4
zm?fKG==YIj7W9h=#<0X)=3fy|y}RA}+uk26U&#XY6v6maSC?nPKAkwHP0v5GAQ^P$
zu{^VRsVsTifL6voAWyqUup6QN9!{z>$NfhSja7ya|7*p@NavU&8HF&=#F0NwwxPp>
zLHYG(UO~Yx;MfIP#lWgoS65p{sYTp=LTIj0xh}L5EG#YkHMu%{C+8f$Iacjzo$jCX
z&wRSnI&RlpS68=m9$5J6-2-^Wlk>#~fzLeBhsdsD4>D#rKgzsH)?l)*HFMIIS;bk|
z=Z(iNYNS?H5B_S`Q=lF;xuAGldvA_^k-9+A5WPB|mLX|bD!)t4#e$uWYmJv+Yc+QE
zv&>3{tWUQ}^lEW13uooXMGEei^Gl~EHH=&=)F<c)LZa7eHg1Rs4&3r&9S-jv8qOO+
z*jw?PPo>S#P=)>TyG~_?O*dB^u~XfAJJn|T3KS>E+9;<Z0)cV`03e2=)=KWy-lrn(
zmLoZ|<)9%VD#K!TUmd_!79~R|2d%1Edo=Pm_0fw-N^YeLr_!7<(Q6kv{11$VH4GmA
zu>y6o`*{3Kuf;+2$@F)~J5>OcyV(qkjO-1=oS`vC%7G#M#lhG(3w}lhmW8p!10Qje
zg<$dr$RE(m?`HGe{VSK?HwYzn!E>Vq%r)Tom`kz#z%9z@+nAPJzWSPc8P9Y1atA9`
zx@H8SkWJXzi@s8lK(IMOt9)e2bVuJ+zt2agmXS+@<fMkd;#b&D1NyeybapOlnLt+Y
z>jdc^96cmV+hc;p#qpiTe55``RV}lazWzPPhcU3saBqmH({8YGZ4M_pi>6v>Ao=eg
zZSPiw;_yh3PS{4UX%+j_ENyzEpZIKnTd;&6TKprwev{RR9XIN64r!#tof2d?(TxRZ
z2Guo<ak6^j2V2$4G-y)bv_x@n@Q^*r_U@7kF3x{_{Thpj%iO|ed~7*c6Sg_(Bo^iS
zVkX{?lQj_6G{RpO7(`S_{mP?5;1QU(u`jAx1`A;c-HeD{dSaObppvqo&BXser*u6(
zFON`7O|4Y~lwkV2{QMryK=3U=>nwT^ECA03PK=<wSwa_2Z<`w$8u*?w{*`cg?e)Kz
zzU)~3g|c528CIJ=9s{)<zzRJg94s>JY`cD7^lR1qqLN~sD6-k}nqJo7bff6R#m%j3
zFQrq?xUSggTCkl|W_#{YbA$H6uT$*1)nVbCb_GR*@d;!sek3kW>(ufP1c`6%Di>0@
z&nwI6zm^d?X^ncdmntr|OEYS&75+Nfg<jbeSx}kaq!kkWOT@7TdH<;P@Pz${Btnkg
z-Ac~;*4?{!1Qm5#XZMu<)qB*t69#l~)V~raDlfK10s`5ZVh3I0J^~ZSs`J5`4lJ+H
zq8tOGgrb+waDA`z7E>BU_aovaoyEI<DeFmxzap7VjcWzG@9L*5W6~qN^<TWc-gJ*u
z_U(*}x53WodNE}~--EiHCXUfZ(7AJ{-4u9g#L3BN4mKUDcP2o95p-ZNhJ~SZ5XQ|r
znasU7;spvB)foug36{mHlfFP;5m%QNES4tTi-{s_hSe@|lkTHNf8qRoXZ`Q|^1CK6
zd>50VN8FvGe9x@Bc1Fk9LlhB!jW)^^Kb%TCw8+fB&aHA>Gu{4jp(ucXA^X!VRgRbf
zvZ^ygLA$qnT{%~G%d-&oD<WQt0qJd=x$~$!#7?TNz2-#qomDt~OZmR{uShZaVAejO
z-2m*l&)9i_ZAw^GALhNHj<2oz-`c=}R(rdeg`u4Vjwxdn32@U006N+Td=Qk(K^6jQ
zPHkrtQaQOTrd1A>AD@&x$$B9o^LJA1U}uC=y@%`~y?K;Fb|?eE_HDFL%=o7vUixPl
z8~MdQZ0j=oz*z>$g)32Oal_^K=W`%j9Beb7e?0}Wl7B_-6|Inyl-o+bG>UhV4nM58
z2fkMXJ}4LF&@&`W1jE!7J0ctkSOj?fbeTT@$(%kudzYMS1LsiHj(KBI&#Z|AUqddU
zE|il%?jFOhEEf8MaC;$24%OXS=gs)q$WM!^Od}lGbdSB-jpUGpQA*y*FSrH1zF;L-
z++-pS*uDK@c~5^^jwF$>BvUNdaPBT4C7Z?@*1F;b&uuSFi<U~1^<%<wZRp+r*cQ>G
zper5;P2q>sZ+h5zB1i%vDbs0<{AH2a=8^4etOOiMd2jFE`LSn*XCCCRJIKd~6q4pV
zPkQKLSsJPOUVwc2u6!KLR6)&<Z8@lD42+DjY2v;k8!n)>5`nwoYs+nY?FI<T(FF--
z6iQC~2u59{6Ucn2Mks{IVhQkmU!=_s+|5mmjnc%#Wj!M}OZN-kl`gc5h|OySTks{D
zytP1LI^Mitnu<~OfiRwE+p}=rUv;_lTY-G6@fQ06pbE|P@pmMdT*kEijOHO9o0Pm1
z7Ws}p@XDk)q_H!q|2h-Fy1UI9KPdHbTsYn?xJBG!{053#4aL@<uH~cJl!#gAi?!uw
zW<i;dU*#bnCyQ<o?Lt&zOFLtQ-#T(UdA9mV!OU#;fbP@qHPT+`8&Y`bxK3JU&4Z~K
z<!?|)%fxZ!NB!8r1@v)QCX!U#u@TT@=wXw5{)Y?YZbr<mt=aAcTKA?kf@pvRQQe&;
zJecXzAplNrT#z5yP7l^O^p`z4-Y|zpM#{j6`EJ&Xjzh-vlWiATlFi4b8DTUf%IOtL
zpsG2#<&~@l1`5`?EMl{$f0#5C#=r;@S)b=q30p|EAilnXd0S4lYsIQH7$n$KeiBN9
zTU~lY5sEW1OwT#w<>a_WkmyYsWI@R*rd}0)_jIb2lvv&H6!K16CDPTx*4dZd76xKq
z<zZS_D+nBapiXzZ;V0BV`x4#r*MdOF$hWNdm;)2+ea&!#DjdRiqZ)8C!&O9swv4hw
zDn5KD3+?J`+%9x-jbvh+tM&udA_Vad0!77eX(u~Fy0Vjg<62DZm9l;`tKv4Vbm`(C
zw5|R>b3W0dC&$O1phcDgPz7bn)zmaI+-uYt7#IMT$FHra+PQ7V>BvTu#r$cUcqm91
zbgZo4D8Q<8bEs)d70yy&DYfyAMR}2+qXfz^Vr?lJ_5oJ32?RKFPfyRNza@9N_4Myq
z4}4|hbdY5GBxPC^c-@Y3iILuq+vCGEgu>+BUVUZzr!ILH6`a3sD*qBAq;Pel_+%*k
zs&v8^f(==(&(AtW>)@qtog(u|{?r^>R)t#7NvciFiX?!li|9n8$=H5FN4I3<ah>n)
z&+Z&JX%&=zqKkPQP;4$k{_C2=gz$DTsty}_yO9F!!%%EIymum06FRBPP1^jA*oC}<
zBg!&I(r7ca?;q6<*!6TgSS0;I&xYR>bdJ=1EbCv^DW~=}OhZT*>n`t)qNCn7@4f4q
z81wWPSsxfy(SfP`24VmJoF-^;(c{6%BJO!g>OY~Mud8jGH9*!SA|w<J{t^T201&Y=
zVYRQl@PdpLa|ZHqNLW~yMy~c>a0!tPD^;lG4`YLI*azw_bU|X!1Am|gNrP%D9ZkWn
zjKGJf5QYbRKJ|3)Ms{gyPja;@##X(G1Fs4$EH#VgeiP4m+FOJ%<jfg2-rSvT!oqja
zNp9zQicnDMbs?Brw&lqicwp{*{n2-QgV=DV5RH`Xb$!m?*Bn$#EPDS*Wq0}8`J1E%
z&+KKk-Xi{m9k<@b-h_P2J0oWgo?wfK558N%^UHF|%8^5Je+pjxTF(+S3^DO*&f?KW
z_Lgg&y+CL)FtqK7<%+Q@AK&D-qZ-r3))$FSgT23pSK2^4%`l=|sc&08etb4QIrb7I
zV_>GJumH1X2cJHr*j9Ty<6p4E5Cg*^aj7<}I~=#mEGa3`TmC`;Iyd*F7`X852mD>Z
z)bhH^agz2nt;FQt{RhT%e{=(&@!xs@{yGxCFXk{eV1H=cla_tplC&3%Kj&ueX42e6
z_}1{fnhvL7ki(+)P~p7?o%p--)8+CHM6lvM>Bc|q|D{vJYV2ib(%Q)SF!-LW|3__t
z%-t@XYQ<YTJ<jHoX{lf2H>nh8URzZT1`1**fX-ho3tm9M*tZnWuY(3YgER6#)O865
z3`hfo27<R9u4VM}%3C>bVwcyX-b@36vQ-q?&_JPFT1)&ao{B0wJB*#~C|t(Mqgni&
zekkvxxY5f({w(q?fzT#NDn(Vy(Au%>M(T{*l$i5||6%<cI93Jog<U3b=`Gk^wnJ9`
zv^|1Rw4c8K?j0o)j+fhMLQ5T%kl*fCA%4@b4;UW6wizzB2!@%MVzAj@Lpv|36OUGc
z>SMs^J(Q&dHP)~**NpN2qDEu5pSGHUKh*Hgj~Hin_m&UhzB?!p9O?;50!7EA3YQK-
zM~y_Va&WXj&)RusR@x98;;7y#${J+-!5bZz&NqE3(63r&uD*Wo!RATnwsoI)15MIQ
zc;V}=MJFdGDQMQBnhQ|E(ctID+daYf(Af^n(NntyuH)9?$VhDMCr{qf220~zl*dzl
z9l0Sp4>@#4%4W!crQULi(KX}N@b%tas{=QkJlXiy<dMne+W)l-F3g@AiV;yJWX@4j
z#Z)Cb+34Gb*BeL4)MH=1EU&%2%*b2WRJBNu5ZulpGPG<xv+|weC`MK*#~`bXa?jQ5
z{@hZpW;7VD!8sZYLMTE?TbuY*P!Mx$P7z>RsJWz&c#`7ud2Qm~r^HN7`Vp`Fdy&PP
zHZ?W3x{4PVV-Up1zNVc(8S%3{M(F0vYg#4oG#gf{6E-xS{F93hH4V`aVc&(g@Eulj
z!^`N*T!wzpKYH5gtjy0xL0z`Qbbsl?#2EF59wlwt6zJW$xVoYZO}g^mHSq>#54XpL
z?tT3-Oqkyx18pZXLpCfEtTNCjYC>(s#vg7JQ-@-Y!J?WP3~D(E%Ky~+o<1Rsn}ww{
z^K5DSgaHA)n~A^sSL-HMO;Al?w1==emH!sx$3}RQ?~XGP@B~Vbx{<I+97UrvT-MK~
z(z2I`%CAv2^J5OR(DC%s+8&KpM-XPLiriSg&60JXANhleT8W!G+@O?}f4`n}mYOgo
zIoav>ef%;R_I`%WF{9d}^*>F93m<BJ%te0tER;i8n1^3|Qtp!0&V&@?TkqsQsTOph
z67;M*6pSQzp)z#2z{F6ZmxAg3;RQ)3;xqYg0#<h+j|~8VC<iOpd{MI?;E$!m4v|^=
zw#Ev>7NAyt2ZdcFU1Ydh3!V5&L-g_zlbPLj#x3u5JPLI|AyJA?qofR;>yOW$J1P9x
z<6}av;n#c5>3t2or#sk~Ah;M?i)%pc73Ji)O_gtbP;)CaPkWlq&cTtwi;uK?6#QaS
zsS|tmOt+?_mQ?PDrEbOD&c4Uxc#W1k_5B@`S#M{5|2m>OUGxsn)h&)u9WAX_V7tnM
zc^J%Fw@Rzf&LGX9S{(iLzA!O^`~(HOlDG}kX+m}HKsrD*g;Bl<vGbi~dx3~d3W$Y2
z(0YQ~1;r&AWkCj&I7}<|7q#ak8n`yWom(mwfM?Z{8Um4whCp?iQF8d&^0%8DDdWjk
zt{n=W-`^oM`M5cLEWC2#37&e+Fd8QIEqPPk56`jbO<B8<zeOGYrbC|!o2T7EZ3i>5
zsm^;2)2%Wfj*R{WT|(or5o>hN@~3y-X%Pznx<y?`o2;)r3T3r-riF13YM6?H^y1&X
z6HmVA)(}+D)!f(q$<Kf>e#4!y&47h>h@9y6MSO-j|GtkrvTzR7I%&p}jE3L_2iiQ$
zmX&vet->m_fXO2c8Y+|C%js50zv+1$SeELl3Bu+g!>*mrt}fvytYolvw~fTqesaAf
zV?OQ@EHIqa|2j(8SVEEYZS%aOaJ%=|t@QU7MfbB6Y(G(cDR{uU72iRXv2&>KsTD6E
zfFfsji_~rBx#2^@`9k-;?9ufB&_%(m@)Qs$IL)ECK`b+SygLu=Z$ZdZ=-@Tq+j9ZO
zC;S1ktk}PP*_EMOB3Ddy=s+|Bo4|#JJ=nN<{_agw*rSA2E<m5qJ($$Ok->ycK9~S@
zuEV~4(#quuIQN=_U9mr1hQf5xeR<K(A~h9=>oLu$mr&YHqLDk?e|(dki4Wh_&OcmG
z=Htgm&uKp*rFA^n2RG<ELmX|0mwIl6$d8S^vR3BrtFR5kbL*g}>Y3Cp{S~Aj7P-er
zxm}g>@YQMTvh9xtE+$O2r3BM9<5sEj!4@m8H)e3(lN)$1FcA!J{B_G6aRP>mva!zd
zVf!9##KVvVm~nJK;?g?@PmNU)>mBdiAAlc<&?0WD&^}>vG4~Hin3P5G^Yg2DCt<o4
z>a7T~7V6xksZq*p$8@!KQ^u*5VfMS;gnVwE+Wu^A)i68R7*eHjRn3SW-d5dyuYjl}
zw&ufo4<6i6Q{#^s_4W0ANUc_@kxIa#t*T=Ed8eY6cydurkAb`U3G0I37b2~j$cVRf
zN%bCyfXTQ+DjdxR2k#~_VLQ&3$p)BvC(I424CwEFAZB3dsu>l8xeN<mP?qW6DrY0;
zP+liZeYQn$uV&CA7z~Y!P=;U*RvkC{PT~K}l@;0lH&-4QQZUvu@3H<i$@o~;(dEEE
z<29>aIz{cYFxC@1NdiKG2U;cA`ko@Aj}M0)sgutf0Uc$7IWvzi;fHqldLjzbHobqD
z#AK49SgdZ<TQUoSDC78$6GM~p_p7w%?KZ}W+>YuHR#mJjf;fdbR@*!&&GVQ=>IF20
zYz~tZ3bT6Q<N_7GQn`le#fb4NaG;?w@q_w*gwSl(G&G!aqc33xQ@I>;^<Shu3rD$t
z0co6}1|m{ZBcVi5p~`)JG`%R{OV^4_6_SvDLryvC>(4&-Ahm2II@t~xc}wTyD$t}p
zet0otF11A+Q}y$4I^zLL?~Zz>oPIJ!8ii2i-yyT-Mh(OdBn1g1F48lgM||(e%+G<t
zui#Tc^#l>KFtBzE{aG0GX#auY=B(`b<d#nUdCzV2dM}mQd&A{k<YlNk%ntQtC+FbG
z`S3~>QBUz<s&|H}y$@O>2yjY#H@9=@$<^`@OYlmOX?ZQx^OX<z>yIgM>qJ=9XLhM+
z6cex0BhJ?;KV0_nxj*YsmpeT$(7b#573IfK81q`XXaBL$^e0>fYT^hG7BoX?dV95|
zgP5EqpZ#wyfZ%iR)3S1MR@hLCo>xf)7farhQ)K&{#eWdHHMq#aD11e%krwgqyynUz
zUx11Ex!g@+OttV}>!)EAd`C_eR&CAd1O0EmTe40!5U`W>HNNEzn*F<hd$r<bQz*Ee
z+v9aOW*t*JTA;^h#)LeKfFs}<3~i;cUFjX01zx|a$dvvYEV6dx`xKvg)TF@8BKR>y
zx2~LwM2!YBTv_&mT2yy%|LMo&sW*NW%f__!+t}q_BFd=|3)V{%u2fHAYC0Av%2spv
zd211`?XJD=bslHSH4t7e1#*h=l_M}=kPJ15I8bV>j>=!C0WRl0@gq#&4S>y9;k1Wn
z<r0I|gsM3xl8?ZLu|F{9BG)QqES~(hJhg+#2PCv;qwB?5<7gkCZ2Y;hL{uGr{J^u!
zdueFmw=0+1{H!00sx8;?#OM&=;73Z)%LB=)^ZiE3TLPCXj08_VO~!-7$hU}-@|)l6
zzfx*FGmZ?uy-SiO?|W-z1&#FWV~Yc_&lD%9p;qS!`${YX10P>mVxOAD42}OkO*E;0
z8)J_QtDL0QoW|3CXNtuqzF{@bYC^s!&UdO=-PgC2kJIl`OTMd>dCk^{eL-f9nt`x1
zGWbA*ew4;?g^j@_l{22?3)c_p(eN*AQwzvRvX*V)w`MJW?B0G3!D{$c_C>+;jq0~e
zv5*3otNL|(-x}OBT495qfcr<$i2H;?hK~d?mhvQrPxXfIaC9fG-OfIbI+9%SiB3#k
z&6_(qy!zApGUJ-j<%xG9qJLo$%of_1Yj>CTxo_*n50(&pQ$*S4uDz&QFRDdaT4tjr
ziAk33Yq8y7_SR#t_pS@azD4+h?dyAqN)m?q`6tV)icuueH05iLy`o+p;|!P+xsbHX
zBrCPtQ~b>D{XvW%yM{ZGmnkdvnjFLQnSAeJW!;4pslZr!70lOjL;XNa1EGd220&|<
zcT@xYunMg3Az#0K-8noY1rr(Jd}|C5jM>MZ7UGX(Lnayc{#0RpMi=N}3&6zBv5_&+
zJus`tzOa`n4+hoNvFA-Fc@j0KhKa}o4v|?m*)1CLACXKIlGBh68r&iN<$LR{RJ)ED
zL)Jh=^L>KM7YIMOOx0<|w^46i-H(4m#@uwxFGQK%@l(ct0TZd-QMZ=bgF6HSW*aZu
zK&QrHpevC?ISF5PSX_IdTV@kKG*{-I&74^l@$iYaQd0Oq?Z7O0!`q}3iK03wi(lwi
zcyfhdH8j!;0)?MWZhHtaZ>TQKklx)8R8905++Ae8nc41fiqK>)=@AP_WtWhcKIZjb
zFRc8G`&Am+a44|LC*blvUyN}<A0wfaMk?ul(LODTO@!u~6{CbBzN-kwE>;dgVe&3H
zKV<mMe$hmA>z_F$rmi(=#m`AeNmhP=fP~}W*DcgSo4`Hs-CMqtz_$xbk%&>sk7F^O
zLyjuB?GGae(aI?}7qLwUl9GpAVxmU->L#~a_izKc-uG536NVJ;dZ(HUKKe-dYb2Ge
zd}}vb?&jtf*(_B7o{?q1I>kp{m^e9mUmethw;MtLr*+JX12DSpjC^iUr*l{R12KyA
z9_EMzUe*oK4}&=0`6v8_6in4OpEs3{dcR88!Hl@`Ojzz<TI%h)B!Tp9<A(1}G5A(E
zz2C61<%AiS1r)s|`8qWTtfm%>I^|oz;2sW=&MkyFSPu!2QWk+ABHc%XH(VYR|J_@J
zvF5?h#PDd$xKxZ>7d(QAq5e0&`3Rgg_w{Msj+Ef(bwwc`n<wuS4%4?Nk<evo?KxCC
z4>Fk@899vjM3)mCeqq64?ZYO(eu}sJew2kBUz4p`gTK0pr(CDv-B%4)fU|h5q4Re?
zcr{s#PLHe~m_rk%b*2_DYl=~YSmDU2noA;PxGI6(&4Xp8Mb!gs0zvuEiQm!eF>DVA
zyvZZJtUI7xb1~TL9}|NCj%zP=_V(VuuirpzQR3U>aNX_HR*7o1%#MS=uF|I9_42;w
zCQ-%0%na7v!8@+?@j>lI7r(Box3`)n3f3=PHi!zmY2<rp&oY5V_|JytCK0|3`mzG=
z%~bk7(bokM!mM)2u4km&>33RC3})QKfBr?W)Q%>#@l!5C&s$ku{M38Dae}|F@m<d*
zr&DeFw)!EpX=HmlPNG1M_DzM-cA0-e5&LMW=+Q;Jq-QGU$-=%rYq9-@FHHvOzn@#J
z7PKY*3rKJEiRK`wd3vIwDeK}eY|Myl!KV3}?}R;B$x0=D*R$blD*W0!|EM9OyDqYH
z1bFxxAFS*6C}$;ebhun=$I9j=3H(kA1b!4L2O;Vg2xQ|v)Bdij;<ssM^3bPV9K5$g
zHW`+G!o%hSL;A_Ek<QP`ZERA0wk6Ni$Md)++zaNED_)3Ru_0UxHDjdYHDpaJX=e?_
zhu<?ni1-DZ|GtQb)m$YZlKfvQr6Z0i2Wq_KA$9M@h~IrQ^#@qARJnrbgKHLS10QXh
zjPJEdW-9Oz)7o7fo13e<=JtOUmSgWxtA6*j-vTwVgR*U0dm*%5Q0h`~{e-=^j-gj<
zzdK`e0^EA6H8m_dmg_qMGfF2fl~=GMk<TpCZMGCEYn9?`SSg;8CMeBV$YgZhA<?@E
zELSKSCZ%3SzIpoakwquU=LhrEeOK?pfB#CeJT23q2)&qb#J?xmTKNRePtXc~c(rKf
zbqH>3LV_5tvPjTX!aGmJgD-#XE4QV1XBYl{y+mAC+AP+M55{+@@Af^Z{DepuKEX8S
zR&p@LtBgN3Fksi@<3o+HDzFmE7#q`|rf1UJNA$G9_+*|wWyC?hZtOJ|l3@=v-B#vz
zhCDAHM>}~dXHb5xr}WkL%naB^FA%mSANs;`oD~sqlQls)DS~o=NOapGx5vqsy6r=h
zE-_L4g5hEPwm%+t)aY@kMS_8Nz(94Uo($%sOBct=`g${X$BwIoyZ**w9^y=E5;r+K
zjt!Lh6Zew_!A6-T;GUF4`8l*>OP-EG_My?uNH^?%TVpR(*B{H|3uJW@$=omWXu`Zv
zixKd+(c<@4F$QAH49{8Xn6c>HILzC2CK&lsLL1+Ot~t<}pWq6qJiGa@6o3Ei*c{O5
zZ~BGDD_L<);DkW+bpFfWX3WAa3z#%MEA)nL19CQwj<n+r-*rD}e>n~e5V9wqBtdVF
zotd79m=djgx^HCn-6e%WzV^->mcK)v#s?n!WWq5uiiQ4(dZ#=0(6^2lC?c#*(gwF-
zG^qt<F>n!8n-vpI;Mk8~Wau>N*8B2=kBa<MFUc{CzoqRsXC@32As0_np-p>2q<Le~
zBe~4leE78I&)C>4odB#n$_f0aV@*q5^K08_kM7*2=S%+7q4Mo>H#Gz14blIHtFMlV
z>Wlszx{;J_fgz;3rMtUZS{gy5J4HaGVd#+Vk}g3S0YL=>BozcD^_}bY_gm||H~+bo
z%-nnKx%=$BKe566_uC7lbPF>+LdE!KWMU(H@8&%>VbsS3?EDR^LxJa{aM7xoynQj?
zVxf56wuOm~II6C%*SgLXKOH9K=UZ}A^9$FW;M=(IKFU=uC%R(Wdzkw?1T-`6Nn%Lf
zZr&@yXPp2Il@UEHNoKB&FA+MSk3m~J-KD*xk<Yo{x6YGl+qoCsXMxTYYVK9DWx{vs
zD-Ws7Zd45)9Pq99s2-#x$zq-YMJt|0?lsVTDU)OAoK#m;5yFN+gdX|l6!iF)D;F81
zOeHplf?f59LRXGL^dlBnGMpwf{0rwHp?h*BV#Tv|%tNe%DD}cACJX^j{?s{f$a}xS
zJhf3J1HX<baQ#NZ$Sz)c!WYpKnBt4|J_1y#+EBw)H<s67by6>iK-O0<K{uZ<&~%J*
z2OcwD?tHRTC|=ih?Sdqe=Q)VA)wx%YE|&KI1D~1zC=9khV4%AQFfuX{jyLR<_wj4w
zNosv{{lzJDa4ln@Sw+mk#V8-H>m<KV^0Ks`%TCW@X1S`KI6g{bFV6=R1}TEoXz+F3
z+rH=Py;kC5mwyx(>7*uT;9}Ma*OrkncyKU(>0RyYR8E|&O`4IAQ|_<wVS}cx2L;<@
z-t?1=({i}Ls{jIbQd!}S(D|TLNkpePt9`XCgLAd;wBxYyC}851l$9|%e*^|BMTS~N
z^;U?^eW2hNu8_K@D!u^>PTqo`gDSRm4h;G%JKbImB=x8E6klK0@!E?rk>f(BEiCLw
z@HUlar};d3)SrTWi1rI~_=()Xn>ZZ+)H(&<mef^0?U=hLx_F=UHcg~!&1_cRC~`7E
zmxR=*h`dklv1DGsxUmMVKxwQcnza`B=6Q^ty*A0f7mx9!1Ge)sU0KOXgpq4;p+hLF
z@bB&3DR}FUVIF-~j>Fqt(5vtesQ+kzWeN|l!?+BV0u*)qgBOm{>BJ|tcw%x=m$#4P
z6>pxf5z3A4MJ+jvj>gfJ;XQiFxv*CFTg3a@L(R)UofhGxqMlL`hx_i6C0he9^v?<o
ziX7@ypX_WIqWtgCT1PPNAhE2}R{x<R#=`wWMu#V2-FQ{(ccq=#gh%-fgrFap^}%SM
z0Pb2WR##~|%V6*p{hFOKZJS5!@Sd3sCFFLS?%rJ=WMgFxM1w?1sE`|uxc^Z<AXvo!
zXCWFYA-8_e2NAY?o31H_1#E+M9=tD#n%I;Rxv!+8Moj5aqoZS*glwT(9&tjWsgE}H
zMCHj5ohA`}G%MlQ;saSE)b7k#t<6J`PF>L-^I4Bv7ads|sF~6yet_cqQL%PWR~IF4
zimF~@Alw1IJ_}k_yNXp@;oAK;q4Y^QZ&l)BHv>Le^SR3?n<T9j-P>31v2798(bWxK
z+lAXcx;F{f>(aXq2&je969#M*tJRdFkjaCLmdswb*iwFTZ@|}NxUYpfg+`#%Fo|4)
z0sk3D4>s9GX`OfZHbCYn=DJ<~)fox`tU!tV;O^K0aOa5>1_db=NRt&uUNWnp!cIrH
zLv)S+v9o4R-1}?XxBY!WKvCHO&%6R+v<<({7gw;b$Y1I3D!KgC{`R3dOS&<jD4O<m
zWj!=wNuia@34vMAAUF)})uyoMm^^hWMaMqWJ*WIm<fp*LFd@cZ(TdPUFJ>#0t^ZWO
zqQl37EuZ;*_$Tq~oHjeQ8dbS-@yThYITw-)lqe8`&EE)0mz^k3=*NS+)Cxubl@MZi
zqUCA=A2Q;QHU8ejs@f4;p`js)ie$bX3dka1!hI19Es6p=gQ1A@MIAk!Sq@`{g!%pD
zb<2lm4k+JM6!7+w*AAQs1r@{?P@-hhxp66@U$FTiI?5I1Y3G^;njGGt!gQ>xxc%UI
zufzkfHHN>3{1#8k%3_7ZfpUuLaPJ26NQXCvVik;xvQ_{q%E{fG6!6e`egyRn?s&%o
zNIRkVy_oB;SfSkpkWjPw{wYXF$@u%-PvAow!shVZo>6;hsLZ?mZ&+B47S3EBFvqbg
z1oTvuH2(A}ry=X`M}EGy0NF~#0}f~$sB{wnX-Mn9GtL{`=RTHSYuwtIY_Xn<OgRCi
z+#i|ud_S_O2>3L@;Z+_aRV+>u78ZzNCu{pMPU$xn54sN$V!ZPra(=bEpI<QKP{qYY
z8jl*aCJ>8O9~X#5GzH@0*--N^B_sjy{Y~6G|7Hd@#vCjFr*)RT?d|E2AWT+dj*C(l
z^`#c`RVGg!t<^lvI5rCm49qe5$UylH@6Kv%5|?o&slx&fq0Gx8Y2YO{$;}3SH42%a
z>GmK)@4kbsmR+N^dZWn7f<B{un6fLwcZ#}}7Qn>#eDyv6a-Z`CEGbe@)PREoge?v~
z5g${dXNgWOZUla<t%$ZZ%|$_h_}#V?6J7d4B6kfuL?iRv*Up+ZAoFR{cxpOc*XB64
zqRmQ4nX7QK{%)z+wt;C4fJLSF4wK%jTE)Jtt*v+(fzo<U@bH`+OC+Zed~g5aXlB=$
z3zQv@tm=bA8I-@P6JDN_E}3u@Q8Yi6)<hQ+RG!Q&>+Z^l?}dDZI(d#`=@cHIlWGfG
zt_Gb15<n@ZOoN~H^>y>t1g^PH^A9Mg4z7J(b^V!Hf7rYu7ogSc<l;o~kY>M6%Rl~C
zIr`iR-(21swH9IFgqSX|xfO}i;KyyR2^B^M3h;)Wn5U*W1p)Q9edXBCFUJ5P$)YE^
zPq6?=aVLnsSACK#)(#^37|cni#ZB?#{tji_pSgJFp4fANcwLRqGs|_J=if!AcXIx?
z^64i`EMa_<zmVCg%SJi8J-@h!zc~DW1KS#t{sX;ADhEJ(-3R)RKxkMc$AF*~%uPt@
zOs~)u14L);z8}%t$_1|KJLTGNdB!y5!=RfFy_6C{$tj_7hZ~ghlwNaEJk{ool_U=r
z3?5Uz_CnK+Lu4ebr_RZ5%RE*#30L+h8n68Ucn^s{6p#oUNb>j_X7k4KqMd$CTnSSA
z{{C6owimN*Oj+Q$rpRU#;o&x5Cu<;_oFtBqkAGYRgEO8PctKrVT{$09rOG!v{*i(p
z_@z%=9(#v*gjCctko`XD%$wi5P8L2Jf6a$+yU-+LhdS`3-=MW?^GwV4&47q3K2=_K
z5Cgw{|DKSRjm20t12^|QNGT4nf;-H)dM>vTbz|ipU=_~9PUi}5gSO!qFwl4ix~B3b
zV-gOISJO{IAU7_x1D<kwkaFK0I+rt0O0}W)aDK+GksA3ELdOUrQP5n`J;H&RJ|ms<
zm3dNw;~((U`eEgfDk=)*1+t7VvH@+W@lHr-C7J(?p|5N-V?CL}T2k!m=DO|^F##c=
zttBI?k>|fRSi#S}w0SQpAijge5ViOyX%BKc`T*+QQg@Kmk1?f6FzU$5`^<n^A_i0{
z+|FidYB))X+X9M&VHTZ~X2c3k%9l3zqx>@uAR=v7cXt`S-MpsMl|83f8)t969KMs0
zhb-bp$l|g{!}V0O-FK>cn{@_``q4}9=WS>OOBJOQI^r%5_r8qMY`mb#cXFwYV#_t?
zeof|hEOta~o^R@`=wOGE=bSGP(;u^Be28D7WE9a=Zdz;dDr>~rX9l1oc&(<%TF(#k
zbYo=&j-{@@F@a!#`8dxjX0ez+2j2{c-vIY97Ch-qprX$d%VT?&ED=3F_u)fqN(%n@
z`MEO9Wa&ebCRA02pb44M$vDyg1s!e6#GFY%V_wbbmvJ49K|AQE4hm13Mp9dy%jG1e
z@$-B(S&m{Tm%AJVR$BIHL{y6T1Pwh-sJrT0mx`;F(ceg1G{!ezxPDm24Lx6gUpf<0
z3g<<76}&WY1!`OopduBZOa_;y5+RB+cO+nS0Mo@nk7R1gp5Ap%jdbF=nKbG{2K*NQ
zXWauA?oO5do4Un!U-w^P!&4qRffdk5=Zbt$`8fpteT!L<wA7x~Fu_^3&W^9$GxgMY
z^uMUsNia{A<*VBjhbC*E4($6rWIZ9mwfpw-=`S$R2u{$7;4-Z<;g#=Fx(yFDxxEH=
z8V20-?Y?GjW8O$sdK(sp1PFtN<a93DWQIGf17%%h_~FWYP$dWKnt&^3oRcZ9R<<bM
zw8w_tw1=jPB{`RH`;;uYh6^{4ew==6o4q9B9h3HyJmd+fd0miv#z-aJ(*-Ub<o>}-
zk4slZo;RP7{+jZ0^3nz4>%DU1i>D_);O03{rTW4fWD`&u-8X^}yAaQp>%3@d$4D7f
zKZ#Zmz$#|*ZDy=?tyW%y81!vX)n0L?iQzI%0K>VQeD9D4cEDZ;<V~5S?v2K{?3t>F
zxVYVN$bJVZKw?k0&DwB*+Fc4<bDc?+=*u)kVx{rzu;$6w{AaLC#d~iIu|&h<ITOZD
zB)=Jc({}Rd-)D~uY~ptxL-sUwUumhKdO@LWzN2W6E&9f?o0yPO&o0v_eqIKXwwXv-
zeltMy%emXwiHA)2tW1po@DGxEgSyoSXDl{6+DEg`kri3@=%~Hc!A_Hx7qDYy#Jr4%
z8*Ub-C@`;zo^Xlw!uQwBR&)HQtVmCGMuOlS9{SOhS9x~Qe-edE*-PtEM(zE8AXSlI
zbpg)_+uBEDlf7A;uU(n`sBz$6B}K9~th$KWqaCMB{qJMipZz^<l7rELm}rwl7zFnM
zzjzuBwENr(Vu)0&h&6l@(<QC*h?7$BX@GE0v1~{@&Pg@j>y9%*!nxXvIP}eXQH1Z#
zZa2JJu?%yug2g8uidxu90DnnM(H#<C!RqS-6$?LhbDyY^lg_Q5zZd`h#i#Jbe7+3;
zf3#WV-V*`}{ftb>+_ZR1)bWh>6Pb`izp4th1W6`c&nf=ZkY&WS$|n8w=FK-WSTvG?
z5LK-q&J>Ge%Vbpnb9m7Ldf^=2%&8s?vn?hXGA4Q?rUD{vWLsGVtJM_m%_rQ%Y3z5E
zCDie8PUVR=oZ9sp;hz?}!EuT}OJh%g<xV@=rRiKtM8h*h5Su)VbL3m6s6^hP8LIA~
zb%<$oULM-&i0L3Qe_!M<+#F;oHzAH%N8%(?!LNE6BnZouJYErr_k{fA!%Rr1$d6Fv
zY~*ddhPUJYn5);s&e=vU+0u$`;}Q~1vm<pA)wFAlooTRV2#tK<)f%dTyMtwNb8{Qs
z$+9YFNj~HyQh5x>H<;0~a=+D7Rli&WRqP*e^67fS#rArNn*TUj_gN?F<Aak-HWEMW
zBr;Vay`|-NW==?`b4^5u8jK`wi)G(z=lNGzkxD6h%DcI798)>Jf?y|3o&+hCSkp+-
zH1%(RQF)jw8(p&=!rVP-h$<}dPCYKpiHK;9J~*7m)+GPA0)E*VX5z!2{E2gmU(a)r
zz)=rxH-C2Asxol{YL4Wb97;}e5wa||icVGMmLZy!AyPcXFTaPr-(#dsX1ISS9}vLW
zHFngo6fZAOu*4k)mi_+ALz~7!n=~<e1ofyyG-jANPT~-TS~k_DHB74T-hejhoH1_x
z#>TgkqtVw@suBrK5=fpNzzjGCwR0KwS+>amk20gE<jXX$LOZX(n@Ir7dB%VoAg1jq
zSc-&<OlxQryqb<-!y+Z>U+sQVvjA(~9Q<a#2PG|m*mLSHWNeS*9S#<%fV=EmUS=~B
z#XSJ)18*_Oe?_dZb!pPGFX0$J60L5L|6YH1TZ)ec0|q#r0G%pnY$QCRzA1eRLMc>)
zIk~w}L%Es|%Xj%mV`a#qI>&!XFHYBn0dl8qVtSmI1mH$pU2gd>y|VI`abrkVB8TD3
z$9XVmn_o9WbbrDCrPSzvCkCqJAYc+4fIfBOnCDsw0uq7uP>~^@8$f`1xx@c1^(G5^
zCveFq^bG`}TyXL0nav7{bz>q0sJ5H*$l*4w;eP%l0bg{oWl?%3z(!ALn!7KX9Av|5
zFhCwxbkL#DOPod!F?UJ6mnr8iFi`p4I*pa~5f2HCPtc>zWz=TN?9QLbTcA5R48rwe
zUq3&G3yJ-g?{;PnFGR&%+8Y`OK;}%I=X?c-7M1n&V-MTc!G%%j?6NB|>Y)3)ngOf}
z@ZnGVzb?>~FGrmPm^^Ygn=OB^!(Aydw~+oNyOM2RDQ~udbr*%7z;sj-v9LdyB8kGR
z5RHH3NUYAnbnYUU5ywQENKb;xNZ_VWu^#r4{oc9*%X!uTqi1f?N`qSb{4Tjf`B(m^
z{GhyMuiq;Sm8x?j7yWr@Vs@woicBXi$230o44jE=X?<8U6)9#H79yc8!zG|D7q&bH
z@6}U4_+d_9Z<l8z1a~^%Bw!1a?!bQ(x&s0Gwz`RN3FcPN-Pq<1y~5EHKNfImH3fY`
z@Q4v6jI2t0z~F`e81S0Tx9uCJ^3o=<l9D^Hqo_MFC7DVtx`<}T=^3%Sh>Q7hu~_1_
zf>eF0E8?aIkHz-HdO(CVj3AUBWIzo#ky;MnOdx;ZpT<k;T{3kddw25ZK1mEwhdKVr
z8^t7{_w`d)^t3T=C$VDfn9uuEFYoZ-lKvtA$rg+JVv$bVYV>JfmGB0(eXU(N=+iF{
z?PM|)WCkH*Q$l_@X{iqhD<j0qqct0;P_odn>&QLYQCQv(sdiF(?<~&9WqW}>pgB0~
zo<c@vs}5P4pz}l(R136Qkb%jM&Y9<$gP3JYVCDYa&itj$K{#KX?LdlWNN5|+!~~Xt
zF{k%+MQIx_TLdL}QFpiWY2WE6)#*E{KSW@w5Gg7C+Yc6+_bL9KS4gWg(b{p|r@B;$
z9CV@U!6zQ&@?tJZMQ`Zpd-jby_ajP$^AyU9IP@<qs}0UZ!UU-d1u~TH-i;9Yf3#k0
zKmok|h2oXt=*Ld>TOt{X#x=+dE4>XS%L8ubvNRu)8`Y9v87@v3RLfMwDM5{XlGe@4
z4Sud1VU|agXACYrGPwu0Y<2Ix|B`dAk67&vsT{9t@78wAUb)nq>;uT<b^=G8jGi8G
z9eAdT%gb9W3z|cqDH!`Ja4Vkm2Y6Mr2jOKG(tm$StGM<<EboR@H_)CP>k0h%ASnrY
zF&(==K8PCm=oV1Fa&*OkQV|xmSMUt~^wRj9eJQX3^ISH0JYY}Bid29;rKfuXwSEEE
zzYPSjavve&7P5tfs>=WUzQ{aIKF4R0iXhboMp7BCUa`2RD%~U|b)AU`h^Ui?BIHEB
zo}HegdPgls9E_R{8=6O7zbCi#vd&{?2@{E+GPJoU_xzg^&4)5cm8;!*K%!pyw%GiM
zl3Y5jt6>Pv$FVm&Dc-f_`1DA2*(ki9#g4wA%^t*ve+_%aer3Oi6EKb$tCZUC`|_s_
z7i3%lvnZVWUoJph&zY)#px`DVk!ujVAZnb(4X_WE&mID8OYAR@aDL3w{A2}M_k<rI
z$2!P~6L{U3h<^uS`#8X=;&(kJ1`$jr=+DjDzf{uKbE!@9Z5tkAph+UV<Bx8n76^=2
zp+Zcnvlnu{1QEW_TReS0f>{!Dpj>pry=HN^dtO{HZks&Lj|P6q{55;@(5o4*!=9;k
z(j&5-#Kw1@gv7q|_|qp#$Tr99^YyY<>H~tr5hN2N>i3y!*hv7bn4}ar7F&Fh#6&qH
zqA-fe=z&vv5{}H|3&%s<6sn+pD{>t;V~$h!g!Q~e$Zp1-Qv2hV2GK@4SJp7eA6<8z
zSG842dMONF{h4gNhRQRUBqoI5HxR}EY_bnmfM2H6&FakH_uuzOBc60@#*}`&15Nl*
z0MJC>aCOL3dos5IHx|N=zzDKQYMS5k=pFP*F}~Ckl1{hqTQax=xZd2-l3<xXw1}V>
zwgHrSfD+7=eoULN7XcttBe1(+8{s#Z(wi4HE#Uj~OPxrT#|R2ilf2n~f?xS4jDp_N
z=1k}xpZPZRy0J^JH7Nm2l*S?LM#T2ivfnA1`Q1Ah<(vCXLP5L&;)ie|2G&|Cf4n!F
zl{G&!yZS|xJ?bJVhEtc!#94g=IPt}O&4g`*!#v(n{drCLYeK>ZOL|?Zf45y`g2?W@
z7%*eudVA}A3qWtk2%3LJA`+Nz<4Sr6P<nyma~oQgS+F_UrdPk-z5R3e_6P7W*#hb&
z+O=Izks^J~<L^S5fT)%S6c;9v7|<bCRzBy{wCwNfydRhSl*Mj;z6?4N*FAp2MNg;t
zB^+!4Mg&R8V`5^8-Brto@MD<{_kIQSf*MJCdFEd<-_MnnZgv>R<A7eKQQ*Uc8q^z-
z!M$S_Hww#K6&YzoUnmlmO+~dm72!EB?&mT8rLiLU)W{%-y@KuHESZfh9Xp!Xku#F%
zbMi%ZQ23{O&u2^x;;elkwo`kGG8Qo};`K^)tAXdDEzoE_-7m|6y2uh2i+aY5wj*(O
z9vz$Lkn5)0@595x(}lu60^5Iercptz3zZtAx3_!JW1wK>HuoAKHE#E?Se|(XIPBY?
z?>}=L2*s7S(oqQq?*s71fY?n9u(V;|nV13!R~k^HDuVwGj36K1QA45j!J9x`49_07
zy0MkDwS?VWM-Y~i&YY+l$%~d;`MvAB*p{?8f(TPbl%9+1`VI2eVplebT2C)p!|f<0
z&sd)|+!u;PO9+4VhE_NtKE91G`~ywQgw5FSyL15>VAu*v9Ekw#(^#+{U}tW>RO6!X
zv&x;03H+IIC)@<wr;}<)`SmDp(Qkp1?J$%cQa$M*3=AYTk<9#eBcp>74rjn^Gz{p6
zZXo6*<j(<}t8&o32j}Wu*QtY#<Y?jD(PLnINA=wFwLETP6t0W=t36dTm{aiw<N+sK
za$XNU?80YbvjSCN=!Us$C3`}-e*JZGIC)j@H~1_19~(P~`)TRc!`WZ6Y9nc&Zu}VL
z{IOEL9jG}23b#gJuM!LlK6PV}VSp_Sy+r_j(y|7QlnX&81H*S9{YHlc|GoK47hqbx
zpec?A3>m%VKOcjA+W;)8-C|Xtm6rEG{RwhXaMkM>dJ3ap#V-MpYYi0XVA{aaH$9xm
zhd-j2Ubt4+VQWcF-nutTv@+2pTyMJ5_*>b;jJ)1#&ehO%!`1Y-Ht3|lk44J;?t(U(
z9QHA`vXyh1&2byE<9XOam>T|c#x>Xp+pQlgL4o`rO&QgroeGMEZ9XTT-_qSrM4O(T
z{trO8oX`q+dDijd`JtK0MEB?KV?Zn0=ro6Li~j#FI1^kjm}@`*?6OrrSWwSokNO89
z0KMl{43!7D9XE5!wfeM5PL6I3ouA<7)7jMVA9z7b=M-Gy6AukAg%peP>`!%{aLa|e
zO8Wmo?iJto5%o?xQ`1a`ZvBxY^`?5tR9+X5>)QhLRNNS=a2Njq@KEkMFXPs54R7r3
zrUIJal~z|WSih41ly9o%*gA0nKDA)q@80j5AReH@sHK9*@$m3E{#+k6Q}RQRfN<Ia
zq}I^K+f0*`qPdlJ8+4#zA-hc0d|#rvu7hf}_AHGwZIa%hlXoqq_93=9f=RE~TB_Ry
zZ2W8=KNk+$_|X;V892?Rq!=UV5v9R(*_tlxCj{)^LDir}6Jrmu?|+F9u!TW_aa#3J
z1ERop&$9{E3cp*nioA%W8Y4RZ)XIG<c)EoJa(?I|;QYT=Wp!{8x&Ws-<dp&&>T)hM
z1QiCE<^UW_OQWjyNR|G6BD1^BWM+EqvOrBngV@n5?)u?*_okL0CJ(?#sbZIGX#7xF
zRhVc~x98QkX;jy>#P%QD0-R#RxguYv*gwT;?;QBq-Nk+2`?nxEVy`QJ6E#>FTv0uH
z2@riPpI<DMy6ah_o`z%4_5&VcYG0oejQnx+;(D4$NTkGX)g%}!9R+`=e`e$3<}byu
z4at*$q577gG_^$D_g<0lH(x$m0306d^_cqL9S&lTLx8@4SWey*dj3u4GPo~k<bQS(
z%KvQtY^KKvN#Qc|c89z%vJAIlp|=<SNTZK}!9WKO*Xm~NLpi16tBGQXI`k~%ucKn&
zOwKkwB(g%DPqIEH&%VA<-1NgJ4#=>1`A$YR#?Q}h6NHrks9jX-P9^LC`MR(y6_|64
z0k%eZ%Q0wo7V7?+K_f5F&Z7mLbc{JB-iYx_V9mS@ItkQ`ci>FXJiG)q-UD*cr=XPc
z1PQJkNck^8@XctM0aDLcVoebSrc=js(;AJg!niS|6ptxFmt2&#f$;KY!JZhSaRNs%
zthO;j<@t59PSgO4?FZug1d!plJAG__d*==XbUl~fzy#BO{tncRG!RLoVv&VV<2_)P
zSWFd4(}4t-|JN^Y;F=y(K-FM5;ZG7oXLu*5-R6$?Lc6pe?0<0YC&UB*K-Htik6ERF
z7o5)TUO-dVHZ#M?Ru|ghWk3a+vN7PhB_KUNaZ+7uoXQy4xL{!-SxG(#zv-QWBVgwg
zQd*2@VZqgUjgt>OSNvAGyoAFvFE^xyIsGfsXQDt*iJHdXAhVmo0uJI@x?6vi>O)8Y
z;aJb(gFW9S#0>yWv{Eg{kEL%Y+T2$8cG}{dY)#F~yuc%`8hZV5jP{w^@Hh&z)F9m_
zn8uqj0QcjV<;=||BrIG~RfPqV%PO7U0TE8Ic0q(9%fjDet%R|RMcB;x<&vG!_x>1V
zg3S)%YxVoz5FWLrjTF1PLi(Cse{@0Kd3AEJvV?Pf;yFasI3(J`NcnT%*r<fl9^J@p
z3J<(&RM&xkc6xbHADRsCB#_WS!vVk{v_Jd#_uoI?|JrVJ`E`KlIsEfdgk^is4iMb$
zJ)c*Dn2`>*!LpbjZ`uSTWk?y|3G_4o@dW@O9n{AA6u3tNP}RTvzyNV3BR>Ki4H@ZZ
zqnoVoYU(i`Ei}Jqhjy1u-Ur2p6vX*ehqc5cfZy4q4xGZ@G?-DZG(NZHw0na2M;lp4
z4PDK54s(qd87<pM%{6jpYk2K=17^DE75nAK>lo|=!g?wOJBprc+60w!LooaXkQhjx
zKYtDp93lFLW37AhYEpSQ%i_|K4Ui_oUKqwAGfvh5xh^FjIjr;q`po|V1QsOFE)eGG
zyR=o=0uEr(mLXRTpfMTR4{*)7^>v9B|FhE+YSD<UE-{evYWZY<`1Ann|GEWct)z<-
z6*jbE)!H>Nk;W=~(uail8eaDFum`gE-;6v43&L*_5a{xEhW0Kzc8l(V*_wisTOr1J
zi(Gb={IA}q6?I9X#`-5cjdYl;TT9Y}X_J&Ni17%zFh=q9l6nQ8_VMni;uWj+66K2D
zeBUu@XA)QjV{qW^pcoPX#1E+H$Ko!1+_Escww5!;eZq2>AMj}a*r<XHp9qvcgsFu9
zyi-{y2a7+q2-?*oU<(X^aO4S`AysMc38!`b+kB}Y(UD=A(9qN;Ws5lpz%c)z$hQ<k
zf@DR`*GyWDi<gp-OXEq7e&Tzx`D+9QI}M|fTKUr8$qQzm?6-kg3Lumb9)9lwmJ}Tz
zN`OYdD%kxdz;$&CfOtbQA;0*65(vJ3`tRMo`S|%kLW*NI)Sm>yW<;cbSv!EApcg{#
zbqzsT8+boK&4{R~QaR9$iyIwbwA0ib+C42&7W8ayC*_JIo5O9%!|c>)Rl}+o!uCU7
zJFJM^(1B6Fipaz_Jmd)?XK2FOYB}vj)WKK2Zk~~%x+i%)8wXrWpNw&boZ?_HQf`3h
zI-Y=y2?_az{~;`6|J?5oh*fsVWGMCQIBI(5IfN%{4LB75B9a$yM_0mlXxU@Dr5xu}
zXICdHAjW_S-fsvPfD$vSy91+BaGqEkG=OU!X0)r*{rKe~Q|^ORr`+!pKI43b;fi@j
zM%0Su#9L@w`eou>JZYaiTxw932L`gbdn>(-255u^Z1!wHJl5^H+RX)1tpscV;pB75
zTVSj~Brnhg(Gq4Pe&PbWuCEvZBIK;B3Zr!9+#n*(<|{Lg0kXH33aH58#m98d&(orG
z2njXn#ZNj=pdmNWWR`G0xV$`g;4%Bh$0NjA65&4YMKOdoe2KSC?LH&(KW0Ffk1}qP
zl$wO*3Y4uT7rpf@nq5g;PQGu<V&<eS|E1Bik<Ah{wH<d&;ND@VTq4t}gHlxvYX9l}
zd`-8F1}v`{l(s3nEHLRsB?#n%UTe-3%<gIrh(K)*nQhAJBp-uIk6GA9g+V1hC}_zZ
zq{KNQfG;)z^aAl?QG$AD!Qc@HpbH8LD#cHZQ|5z8_#YtJY=PsZ1%Nb!NY-_-K1Pgb
z{b1F@fl(W|k+~O|rsRdbdBBDrK;JXP1Q-!mHfdh-TXADCSrm40BYal=`eqWC(bYPY
zX-MBj>ykN&g~bWu%0DYIZpme@MqSu2Rij9)v^dj};}lIUU|S0Z8-xt&9(j0+Sd({E
zgiqT%n#|TE<{gfebU)2lg0@v96m!5M80ra}cyi<J?hfbX&VpB;8h||yUQ2YC#IF@Q
zA|<&_Ao4gF0<Pbla|&cOAohyvm!<=qjVEZB*#Yx4py6p>WB|=m!a4ZsnptybP?Qbu
zqlYwNK;n`L5dZDcbj?AiG6()}q<?>h{>@5P@MA>ye*Fg$e(jQ9x%UK)P9HAM7!J@9
zPdM!Fd&zG!eI>H7vNdRmy_7DUM0bZqt~PAVrIOyB>Zd$CTo~pB`6~-JGUc*;(@09$
zc!ufz8Mc<ljh-;60MhE<?iW0-hV}x6UfSB*NPLYJCV%vRM4rB(=C3)5eNM13C(G+|
z#n<lsuN7NJBWDmJYc#ztc8r*aypy~)bWkOb<k1Vqitb%>i^b;4$I9lCI+jp+j~5G=
z+6eoIJRsCz0cWIS2cR)1xz&Tdr|{b=4e&-<J$b?ny!xSF;Op;iJBLsJZ=I+Ma2W<r
zu$@#LuoPWg6+N&6NqTZ}@+x*Rl)KLNn@2k~16)q7p6;1S6*$tY!5@A*IH0^Z)rLob
zVMlJ!O7a5CoP6aB2SSg&C&|=h$l=RyX$eYkhsz^ESDrE2#BiU)3hun<6Bv>G(P8d+
z6S1-(sg0CTRm6W#Y*162!Ag*-K#0|Dt0hdoKsvbTJ><N=ly5*jq$#O>0;kWClX{90
zHY^h^m?nell7Uv@nXA}&_JWutk&Y!4llCY4vHGQrLER%*g55Ezq;TYn1$8`XQ5y!q
z#cS5f2jHnY0iH|(nCfx6uNA7gR;LXoz>h6naw)@dqlT1|s;a8cc!YHjia<M6^q@S~
z158Ol(Udrq&jU;_<Gx@~uxd~-FyJO+2WelseD^EInjrUW`3I7_>;k!Wpi4Uj%{_SX
zNdgjNLK-$GJb$|gGWQ%n`{$HAu<yGCJU(bIp5CX3hi5g8i`X#};9zXtLnTNVCP>6n
z|4^+j3m`Ptrhb>k{!*jzf~E|B<|Tm>lngzR!I>}+X58t`{O3okG&IeENI|Fzq>SXr
ziov`5g#(~KeO5J8OmP6_R}BJck>c|t1z?T^sy*=hcALPy&IJF?N%KD3$YznjZ=H+a
zD^URc<#A5vqT5OQkcX$a&$OV*$K2dpNJH`kA~49OC3^q@##2yG%+1duGoXUTPbvO?
zLn@oX{w0P}{M;8@P*ss@4pRA!{qkHzw|byI1#^nvkL0(-0mm2W8S*=WSry=&-T|i%
zNdGxGfo7U;`VpuNt!}r-(%UXT&eL^RH`Mq1=0QRNwD00}#dg`j7xQBzLg};5EU-k?
zF*3sFlnC7U1p1ah=Bx><{h`mlvjSX#PXqKw1eBhGNmZkA_as4VhdMgIEdva(`|tBF
zG0KjBmIfhu1fqx{Q0m7WL*qzHO-;dwoBuxlZcQCvdSLfZ{|0F<)Lm6u`vtQ3A;;!T
zAk75RTEJC->!SJhmFuJb{oeqH8Vh39Yhmsyy8*E3P((5Kf3Gv+qw&AtpuoTI3x@RN
zK-{rqJ(-r&RVlEq|M%gA8e)?U|JH~Q_64wm1{BCF1Z<y~m{33|AQ<rR?_*mBCRt|G
z{P%6Jek|6!FxJ9gBGocJ=NmAJ&ANFNQfn)dP!Q_G4Bs`dWj0)I>ncIhFNwjX09lT3
z<G(#D_g-Wh7c}-9hJQgCVX)!ap-XYl22fUjFhU3HSCBnZ{(BsqvH!M}syFba*#>V4
zW#b*I^cBQl5(SrqaEp&c!HlxijRg}t<TLJrKk0lBqfH-&dGnoc+WFrd&0zwTDCjgP
z%nLD~LcJysi&RNS2^|!X8()w)I6hq5s~vUmp$NvRvV(3@^_~wYVx8dks*PnYHpil=
zF|}pGcgDH(VQ9q(lRf8ESO@lg0!wFB_wPqX@+vA=Q3ON^it_1MG6G*>HcF%Ycf{O2
zbz<~l(<D^qHH_=gX+YP819svLdf7;HH6sm4@3RKG1+m)2Y8$Sf^yCir6cL?K|L&7=
zDeLX617vy4)CT5cqJ)C<8%&RhV_{8+dy6epLixDvQ`nnsvAG(aqK+mc_eRWHeBun<
zSSU0IAb~`{J8T{`RyxAi423-rPy5{8c5nB>*suIPOE(qLeX96f)PIZXq~5;njRWMN
zFWW4@nGKZ(%n5PcvJJQ{WfTRKi{Z1ETbMU_4F|2pjk{cn?f!D*lS(QY#A|jomwATY
zL?fVj%@GbQ_oKS@!F?R2<xf#LFM})X^H*B^Pz(#JV&)E_q^A_9;vM<_`$co0(eIy~
zfuKlm&PD_NkQzCm%0t>jM6-AT=7;rh9p0!ODnMaz5czFJ)%ZL>7y$Yi(${=z;yF5I
zjxn!B6u`sDL4%eNzjHaMY0c5M<!zGEJDlP)zY@P2MRUIszP!#LlpgbzG$%vbf8%dQ
zeL;n`9{h9qpXYQ;3_2*e)we30+^~j;^Ikf>SCuC`U$M~zEOCp!=kkSrWC__1ObHUv
zzgV^ty*oR5!RnU>7FM33-`t_xrjC#G>kh+{Rn=M_41(m=qB;bo)ur#>wI7FuF2zDw
z1H?%Y^{9x|$p0SX0+i-K6$$Li=~EDs$-v|)=6EH7A0jrwRK<pqX<6+A6`baAJrc|<
zia1A_bT1#5tI!__$+>$ovFnUypv7a}8VHduK?L(RD@&_5(Z%GMc<ylux0Ov4ox6-+
zFHVb18F*9skm^#})SNL9xy#8p@Zu&|0xfOQMa}!DW&2q1ep{wFw|{YkF!=!ZGe>?r
zMu&mYXUL;njf#LKJ&y59i)_7R_!2%H!i7JNRyrrgY<7M<sa)|M^?hbWOY!$}2R1(y
zU%&`{uG*|N5M%Wcx>VJNmvc!q)FFb<me&nZmpv%-ub3-6T$DJEA{VML9xE_I#b8{g
zDn3Z5$Mg<t{H_U7T1y(cP_Ni2!~)z$VwN7uy0;AO-;1mfBAPbOE6oGDNfDZj0W9^g
zVV4I+Zxb-+g0;8eoRk++`D$o#&5-1rADUJl@uZNKJ4US9Mh=lie|C<3o{#nF1*z8)
zNihbw;*s&K1@-$l17<rB@0wAb1tH0xYuHejbFm(ki(Y1`FXGg!hFgn$R)0YB4INfN
zt7=3UissGCc|s7}CVJTej)xtvpGA{8v6GeRqDMBP%WAP13UM^E$i&Us?Y47{JzcSN
zcD)Cjg{QX|eMo3)DBiaaW%*7Nw*?WLn?64Mso0tFU#Oa2U5&4)>%xwgM+SZrEFl*)
zRyKcvL+r<~bm(f!TUS0+J@T?3kjnMHz1uN0-gtDmrC9r^$IrTi@IEo?Odfu`G5LTD
z!KNqb+*4%j>f$9T8$L7ZWv$x)0Q~{umQVTLj?iFez6y6E3LnRxJmKS_{WWli$B6te
z$q~nGd9K1HSqT{vH3Cz?7H~SsK4B&<VK@7qlVcQCSc^m$Q)^dW@R)EmyuR4i_*uO;
zWZMX96kst^l_ywlq?*l{zo8jl;&Hta1Ts3EI_tWZL$7P5WW3}fAEC-J$OF<$ag=FI
zwB?hC5!(ncGrGtfm)A$MsK*AFH`QOf0~?<I#&<JqN1=U;bi_rL=g-e_%PT>s(0}-d
zo;ubX{E-XQU2`yFf;N4aHBXWg^^2O;*<8*B(<D1{ALO|rx^QjGO(*IwBGVEI#~QS9
zEJ+pBZ`1dU3Eq7XX!Nd}HswlvG!Mqk$;+bzJ$p6+4ygyQIN-5l($dzZN;5yTu)q{F
z+M`jn!J<<a3A1N>+vRRM=D!$+Oo0kR(&%)`mL-?PoHH<j1IfaZ0$aHSdq5Vu`ye$r
zZQE%7C(2?P(X~QP6856%OhHGb6AbYoAl+d`RLZcDyui}1X>~eLt?w*oQ`9kWBd9L!
zARW{;XQNyzKa0qTqn$I<gX{Ap!{J<{%xLULGUL7@h`Q%1mJeGI0#?lS*vaGE?1Vcz
zq==qU?QiyX^=2A6$;#NBH42NbzT#x6Gx(w@AWD1win7m+qpxuj`epZx{&GpXhbiax
zY|;Yi5VZ3cO@xWQsF~9LeDmq|=#&W{jmFF@Gp$ihVn_vWMAa1;XS^~3?nmA_rsi;5
z_Uy1hdR3}5{@_oIzVU4_syc4k)ZHEieGrZKc)yRUn^K`llZ7OPQAgMGfp5Z}o~oiI
z6Kb_?F?l_Qv{^QLSqC`DNZAeo(W6hY_j^xR`)lIaXo^kpuee{M*r4db%ATahPE=C}
zROp01q@0@8M-(AcO0$;<r%nGmc3<RrbUA;MyQ>em;HVm!2$ZyNM>MwSX58Pw$ya11
zX_`wPEX}Pp7El+#*(U^Iv_IQ^=_)<q@*0o+UXhL@+g3YJ-KQ9vo1<6ty;BJcl)I^%
z@zXNPhcAyuAu4UwqgpE7Jkr0vA7#s9nXURYOFyx^h5s1WxrPKiTQNfbU;g6)Jc0C%
zcjBx@p_>5FTW9P9G;Z2+JX?uj1-stGR2}FV&gtI>WzSod5*U-!oq(og`xL`hFkTlg
z#`61gc8~dgc?jn&J2AhJPiD5ubwZLPqO3~>4l8D>CY+`zIbLAS;lPQKbtE8NFugb?
zJWgIh*f)H`z=Y!>D!uJf(rVQNNNS=GMM0H%cF$UEitsYw&PRlZ8Wa6pjwo3pMi)hM
z-`L8B)Ns#9Y<Z1mp}88QSF*)Car0}h2m{TYX@tMP^MyUErMXafki-99-o+3_0E+4H
zS&S>X+b<ZpvFRVha^y8s{xqaHjCt@y%U0@F^LkJgb5s&&5KrfIQ_9cD=qL$KaMoAp
zGsCjswfRBPYG7UGxm>VG?&2LcjN=`%oa#79*eOu$M-X09A7m5ZAE*=c(ySPA9!ywt
z%gZ&&!m(XP8W-B*us5vT&~;2rI0g9f;qKHN{r7cvpFV72)cK&xk)Y34%;8RAM>D6-
zcR3P)_^TGKZUY^{PKvg5d`LbLM%h*6HRFIK&MpuZRwvt%bkXfmPt1%`f{m9+Or4!W
zD2XU(l>N$)B8jox*M=^pfIp-4>>vR>A&ejafvuBPPf$I0z+2y0QRnFtYN62VRLz5#
zU9G*C<R_z)JWh8@gXi1xAg4yeSri<No2j9<Okl_w5)5i3ob>*-jAxoOz>1ReH>SDx
zpE7v~Jr_jARm@(~csEY7G|X#)qm#cTH6Jvq@yg8cW$q<r*9h9HF&al#=Wm47k~<h)
zd=ppb^;~ZSOpBrya#FiEiubx}$R8<5ns$euSa__|@`_k+Wtg;>1m-6*>}O5DVC3T}
z@{jb{cVoxy83zkdcO%Gc8wU!_8n5cWk`Dba9_?@(FV;t1K`9cB<MV;3Cf;L6MYE$V
zrtTT^kZ0uWM@;vRfKp;keb}7YV#)xXN#4W7{R_ska?F!*8iCm{+0fehH;bK4+-4#L
zVy)rSRlivg{L!ULuKK0ra7^OC*Hd}~cjSz(n;GYQx$h9DAw2koLy6}b`Olo4NxS;q
z4O*O9EJ<?)>B&(EnGiRcVP$ec*qQ)Ws*}*a6v`<DC}w_pLfJP6wBec+Lr|9~@aL`y
zjUj%*;N$lGK0m;5lnJS#^2l<J!0FXGn@Yn(KjRp-<+Uby*EP1~C1V$@B7VJnDnp^}
zPiN~~D^sX^uGBnH8zr3o&WLgo8+DUhoICF&C@ulYlzR2_$&>G|Q(IeFzT~*pQT+5O
znTvyVy7LI(DMkYZEQl#i9x2JcS5@+ylEVL32hlKz_4D<c37eK1f;IyhwWfawhS??`
z$!{{MH`gjXL@X?{KNv8Hdv<N!ep4pjJh2N9WR|)-T01p;<XZ_}P4@5;y|p}ijiyM?
z`$O!z@UfK}SaO~B;Owhg&9A}5Hp;eslRz)E$)BWA*E&SzJ(RaoF$rJ^(B+tZo!Z{g
zl618Ke5j*=>v-zbB`_CE1$2U+RgrO^*1p(NZ1g1YN-EHDN)Cdzoyf`>VSnf-{K*fe
z4gb6yQB(akof%enrZ)f0o+4uAErw~XgYk8yj0m|@TOuc&L5Q&W_HD}cP{PY^=<2Of
zB4g~p{+TnTI7xx4uuro{UNjC$IS`_?6%`nUHKc&J5DjPrfQ@|$S`VfN4c^+gA-NQ;
zhreM7$oiMTm6Se<hW@QxH3aaVgM0()7VZKi<4&>8Nn<;v#8bP*K?<WznBu0}v#+Ok
z$_Zo5omE}mP7uj{sj4W*S!@L3&9`Zv0pEd^j|pQKg*}EHocD(<G*=$7e2We3ENSVd
zN*<!2Bj3nMfC!5?YS}oC05O@1+rMpjyH4<;a-Ot!YA0gBamuujmncQX!{6p<#hEcY
zZ2Sw20`I!VyUV@Q7{U|ko{~pZ73?EfB7{+ck!*d)rf&p7UBQw6;NdR&e$HLzWaC{e
zkC|*-MOt&I5~u{CoGVARl7&CL_I*FUxF~S44Wus5g7BEP;_d}coz%7ua0Uz*I7Y9Q
zYFtay2SgDxk0CkuE52m`YOEUX0Z`j-^up*P%lq*w-l27y<X_*O!Wq42I1RTOl5J3B
z$Ha0d%?mPUEQbb)aWR@A*q`c>7!hyIHs&eU?7<at7XC%$oS{Adgl+*m5u!W&0C2fX
z#exdWJSJ%CLlcFierC_mb8|nD9b>VzK}C5KHAE=CjU`hRWB4R$!Y$mSFl3j?%glyK
zd96kcN#uOtqn3{AqFng$AKJluGn?kh&7a-Y*)l?IzL%SHdtRNN@Ka|B49&a>(JE?_
z`%QOZ#g~MU7QTX?9R^(s;F{ZDJ%9{pCBXo5DbPdV0c|2MOFsGOUEH6GiC7N+tcU;+
zErf)CJ^*TOZ!hryTfHz9!^bX^URcoYqyAAp@djXnRTeqt6tU{4ls%36Mq(Oy+td5u
zlmjj5J9Esy8Ytbf`|a;$-c952*e0}b&V%x?@0YKy*XOo_5?Beaqhf=uT#oRK^~}{I
zNcy#<ZRIGatGeQqA~8>BFHIaHM3pS<z+)BYeO*)fnCmNhB+xGQyqZvCe*<($dBJBB
z5cvQA#j$6Z0F1PO8>wVA{|{6pC;ASuM=K~OIFHoMwVsD{|ANT~gPDc>Id7<7@FHGv
z8$J^oj^+UueO3*+9K3jAOsOnluJ7;v>_2CSEN{T&$|Ck(Ky9Kd!wc_t7-;q&Zq}Mr
z$%KL0VU#eI8Hb6-Hw^iHtfaymPq;O`FDENswy}L_r$*NfJ;=nARi&>cv|{e-fLWzC
zW%!TPg>p4+0ZZ#<82`ZK{?Z6~F)T_wlRH6mQUTZnR>HXn70|?r!$5*jqu^si>E3FP
z7)*g$L|mp}$9Y}gE~$Z#_j8+j21Dk{<RN(#mvA69w>ZyG&fiqmC5z-FANW(Cz*o_U
zSP(7Y>*s}2;cP_`1K~tp%b@K;qKAH@2<3pIV7`B89nrDLb6$bO`Qi@<)eNCBDMXdL
zAV~>JGMR>vd+`fADXS|%#Pdnsdka7)hh$Uv-l+yGe}>JJ!Rs1re_z2@*c(r;u<7-$
z$`A}{DhgW9N-DV~$k<oET#*cF>oZWAmKQMK9aKIfpJcU16GSlWAPF*terq*@yXeKt
zb5075c_WCb{``vbM*W`#8gMxo20bF+;qoNpXRdB6e|cM*5Y(|e{R}1-?$a&K(H{kM
zj_Z-jUoCu(8+!g6@x9~Yrf?z22xX_XDU~kCvo85agH<bPw&yz31zQ*DBN$ruv6h@v
ztSt)w`Gq*pyOk%$eNe}QgR(hqym?$CWvD(;New54dAj88{ntVL)tBMv7x8_4LtE32
zL)lf&48|<)sg^Vt2*tIz^<b`XMBDrjBN;(3?1-AmN1v>aVU8qM;eR9Yi*733o77%?
zRosv9ufvz|^K}i6QL)<rF`dM-P~FEPa~^>%xlZHe8(QZrf^n+3I`E%o@}Fe9xZZxP
z*jf$1PXq(Or}CB0<3?>qqR!$Ax)3(`)HQkm-fnDUIX0p?lO(ZDuk8LMM=4i8u*C(4
z{~3Sot)oW*R@RF}hU9%9q#=b3fBy5UT_Dwoj$EajAllFF!sv)-Iy3rCf!-t`s;Ia8
z?c)Z7@}dLJKu#X0Kg>7<wn^(_Htchdbxc3L3@7wqY8-zZQ-D9WyiV^@!yzB&Rm@Xq
zKk=4HK}Rv&pCL&8nO-vdp*D@$3sNy1eH~W56*dv)jN;lNp^L-A4$A?yzbRcV3Y)X8
z>-2n<IQG!x!Yg=<Z0ckeDr$Na1eTWKfWlZuzlbnlJ4CAs*Z1<!rHYtTq07$w9jFF?
zM$^IlI&-rGcj28$u4?Q|&&w`;C!bD7r%(kiro>%=y?@<6fmmwbr5wDkW(ABNfbd1(
z*|TnF1&rQaXF^XI_$~Xdjhzc*{`w8iNai1ODH@-(iee+w*lq!opU;IN8DF)xX6GKb
za3m=7v5d6jg~99xwLfK?e$7`H-M)ycOQ|ENPve?1G-AYUvcY=7g3kZv@2?ZOw&d%-
z+m|&~2Cdn)=ou)eyqsPYpPqzu;Rs^hd4G<k#_<b{8zoitBXfD7y|JIww3zwDD4;rR
zw!R!i5I33sLTpLBSJUqWJsH*j8di|az5?b&?folp0k-|ZVSgf4_v>KgH1CQ0J1_57
zw_U&YJ{87xgzUK8jE~PskBqKpKl?hq`YSW#d4yQ!YM;pRXOjUnGb+iBzsbM27hLnM
zLVlXwEPivkS<F=0rNM*2?!PctvnM^DNj^NLqS4IA{lML>m2|7;xiEfZ@Hg~2OGT_>
zPjYeEBIdWhCB@;>{^xIWzkmF@d$IqEJ8$vn;H?@AjHa6PX!is2n(qKN4a{@yhGu3l
zr%z1NF3VZ;1{nYjhj#bIGBDKbmU>?!o|9Snigm`yu|i(8s`z#3qvVW)NzNOmH1i5J
zw(_KWot{IT^XxR8I{oy+*sj349IgWCvJ$(JHA#u06%_Gy0lziFy8CW}GKIKlMrjTb
zWfg_bO()ze-u*2jR&}VUPX6B2?S8&r$~P~XNn<t8`cq6ytTuePmMdSXQc^`OJ6Hiv
z+H{9Wct>EDtdCA#EdAhV)JM-H)5{QezM=O>@AD=0#GA1e;jG@re;W$Yop%kaj(;$-
zc~olyQ~D75PjfJsjD2SJIJ<E`cwR)+@WRZ&epHpN&z0o1qT-0DLJ3Mhn_c|V)IGxo
zZ*_E}?BJbyMVeMQ2Cmk|B}y}greEH%?PUlp|IqCs5TbtAH~iXp;^-V1+2~|y3(46*
z>}yn|j*h#(joqvvSF>-Z);5LSvTCTkH4b7N6V^|>1#sP^;o+zU`+%_f0ic?V6HGxJ
z`}@wl7t3Y|2=UA)Ful_6xf?l`AzitEhY7nO)mg+1I}w(ef%IZBRe0X5ssfRfMm;lI
zneUi=ljv3q8;6;8m1)*tbb^Qa;n`{lC5<KJ=ANLa<$c}?I^?A@G>e?>X2CtNGjVDg
za-6ClcnOz_uj@;mX7(jXv;{Ih+s|BhZM)|iWU&|5YB5u8Dk&fATHgokaG@U<o4#7z
z4=c|b9i&r2Q4J95uc-?1D}F%HoH4`Ii%eK)=}+R|=0H2$&IE%Qqclw-H(@X5ir3xl
zvrVh%x~gjOn+N`lDz=w>X*+4W&y|>zCQ(sao4u%3Pq$3=dAdrl!_2ON%SG&mKDUDd
z%krgI7xmb@N(vjt8%@FKalfxz9;JM$^&;<;HOuNoqWJUPeNNY4Y_DBQVt98#o)Pvi
zni2f2R@lUZ3XOD#`OA!VJLdq_@;Hm51!FOMf_5^E&OtJu{1Q;R40><R-dWHUPQDI!
zJl^KWz5jLKL+j7g3KC&mRWni28oZ6pc#b;xu@}T}TxMt0G0U}b_m9h)9JE<ByF0$)
zel@R@+0r?xP6Bhtsjm0-_*1CgYRz29|6DpZ3uuoy&NEC&FV5@YU@PAkiI{&*yMcH*
zsoS`m@%NoWzG#b4WMiPlb077;pPb<iWCq`xRyuqQ8|Gh$eZqY@Q){hf7>i^5XlQP6
z@HdK~=JNdSN+W|+4Y=93e~&f{2BYP+CvCzuDGQaIS=f?zgqH-*+V}JpJoI*C#&a~7
z^1uE{Gg(9CP=R}V`&;SS(M7XypjXUyw<+Y*!T{%$E|n?D+xnlXU5W2G-oJd$k)32H
zHE3~mIn?B%wqOOXFxPA7E*exsll(&XD2)&n_V5HO*&?XyD*U_O01U<;Q`Mn4_B;#_
z%7A1?-OkPqj1hQy@<$?f`q#OLVqNXor*MprPdT}BOKMios$S?avUML-R~%6%vbE7#
zQ7w<d*}8R?iU&sAL$6+`8%LH-*{dfU&#$bydcWhCIlpifC1u$YY-q1_k?Pcm?BOVw
zf?GMmf7E|FN+ZM@*%jmu)O}LnW6?sNe1i3*yH<VAWFWb$IpR%$DHAh)`Jc|)&8j@A
ztU|Ubd&wOB3nW<o!3%?NcKggfs-^09DwNcs#fj5h-V>z$_!gu-4%1G`vDG^%H1dO{
zhPxkG1x5MmS7zf<@_HGioho0>z&k1EK0Y@}9!WTmF*O{s1*XPEuPAxG@2eAwq3s;T
zob3e1GBbv~iG{YFnK5|`B@N9|Vdx*WAszu<-eTjUfh8WHcfg9G8jxL#p(&AoUlcI&
z9nf&fAus$fC3Uv%0=~ayE%CU1^O-g0+`<<|#IQyOaknhQ=8{Mh4o^whoeWQU<y<Lr
zn7EnJEbgv-PLtdY2NKB-cjiPBh0=@N0?oU3s&r~5yig>MK9dHeuwS)I-KEcuvx!~=
z$5J#scP!215I>7_6$`ahx4Sam>n2*t*simY98GDCU>=%u`92s=iiVT5JXp@&R%UBm
zV><0c2Hg&($qoYui%<3ulQ%H}d&~MY!e1q8BWTdPT@_Xx6<K>T@HGVV_qXJgOMkK_
z(agvH{6KP<GTHh5SmlPzP2}u;Uuw2LSg`pfG1?b~%a)ZXN^^r7!!N&+*gG7mrhL%1
zJr0lvZ+r3!>}r|NTTl;AIyM8sc43R%7(STS_z0lA>yG!-9ua&4zB1Jx4!b-zBRoF$
z-Ob)vK06ka1XBWfr5~;9*Uf9!QIETJd0QB1v{xyX6jybQyi`&%QroWHJ8a6D)>J9e
z<8>ZQKU7%zBE@}nkg`kEmLw?Q&Y1Q>fSZk(zf%)sA;0y{k&f1{a>GelJg??TMgFqW
zt&X#2nl69)i(=AMDB?RvqOpUu_<iQTc$kgESD*Q-QZ-Sl2osezjZ}B#M$$Rk>@$qS
z_)G@mky7t_j0m==U73?fM!pO*a4l%MTIe!uvlk-5&gsi|Uvu)xLB|FaX8n%7=7qs^
zrh%ulod26Dj?9PzW;1c=m*}y&sw~6_Y6-|zpQdVeF=7N?K0G{oYMJdJ9l+|IS?w>4
z@i~2@RksYuB(TP3fIop`<UwocBaXL&H|{#-*-a(<j|`(HRCcCYRL{Of#~kU(E%kKT
z{6;}Zed&Je-`%?!3D*1YJ806(fs~0w<r)mnx&T%KzmGfDj@!Wc0=lusKHzuf6c&B}
zg9u1a!$WS4h3{^E&g2vnECUe<_vhc;K~uo04PH1CAR3`;L%`<o!0=L=ZnB0nWBPen
z)vE`1`{ROEJEUR=)XbIw$zTH%J2F!JH+Z<XeUrOt*YFV4w;j<2ZD#TM&BSZct1tcY
zXMLMHIs;}Tu@Ewr&1^IGNEOpaw8nW}s%DqJOjURAx<K?Bl(3cY?VqgCDfF$!u$J+u
z{{BC^q_4?!r6jhpF=fdmui0#gvcLkXZB3CquzXBSZmA3QJs&R6(j^jZlc+J%>!eR^
z>-jqi_qaLNDJh?Rqu0<nZ*klFkYlfMjWS;BEV7Lcw9z!4drzp6KSH+)xVSgR5qK~4
zR^Wb-yJzK&t;)T>FBaI3+dZ%H+04G*ZgH+<&o=0t^=$sFc4f^??Q;=dyq1;8&7QQy
zTYcGVt#wn<{&rnjd3cwU{ME|+>)yDrwr|inZLz}pr0}e7VUjxz&wBHu&iCB$E#G_A
zyUn|O02m<~exA@et-KmozMHQ(32YgNUCuH4dXQZ{=bOQ`C%}bjZ?v{}H)j28($igY
z)ecsBGnQ#A?mK--b+c9!^QAvax{e-mzJB|4JFu?Yrt-5->fkCl$9F$=s6AhLV_nlJ
zODAifJ!bO2o8MDJ84f&d1D9DoJ7cZ^Pd_lPmdKuTwrsoCoF}VaZhx1@d)spDg+PxW
zjpgsWx-!c*E;zUPh-%98V_U7&&elx~U%1im;@+1#W=&smtz2vRoq75aEPin}FB@Mv
z5H~gJv61}$eZV`<7;Z3~(9%`}`N@8ZH!DLzcjnfpS+i$f1|~4GYK_(34sq*e0FUmD
zeIvPImglPlvoA#~T~%~0rcjo3f138*`k;&VjEkgyu8aO{;+?s)KNi^JTJL8nBl>Gr
zi|y1ibFIv_?yy}k(YWtbtge#$$<<F&pYOZWQ}(jsa)9OGM*=|mD@-PYPIC`qWU%Se
zFk)h8I388<w#BHrPy2*Oh?M>bKl>dCwJ$7J&9z&X9KWOM@Av<ikN&vd`dp%az9{zj
z=ISr@_s?WSD=l6u{Jd1@^73tVj-fN3IM!a5y7J=x{OkMP^SyqsGtzdq`SQj8t6grM
zIV`rMUt~w=T)R4-r{4Q+>A$-?vpaq2?VnS&eNCTogTZmL-tvdqZ(eUck~(AB8=p<x
z+|EEJzF>`XRa%{sI8lB1wF%<izU$qpP+GYq_w*7qUYkh}N5?(6Y8odKo?7JfnYmfY
zPc2pDQILq%2ivn}JezmtigrCdb(^1i-`x+Prz20*nl7Hkw&iy~<kOFHF1?;}fAOxh
zM{f6>5nd<d8eHg_b*)QFOnaBf?MX?^QeIL<JkD)`)0-UTUB7ZB##`#f(%Q$WQ$J0c
z6w)JC>?=Cy|I*8wr9Xsh4D@X&y*bAi=-AhNYn*@=Cmouln*3jNtJV8mX@}j{pVVG^
zr$FMsvjxEREW@@wjS1ItPJMQ64OBTQY}B(PXxYjqyUtYgTv_uZMSS1I9Pgl<fZcD`
zY}%%)ovf9qa;?TZBH+CC`&+;(W$sq{Donb+;&frGX^g+4fL70vmc$bt%TjMnThpW2
zd|+QHOP+7(_3(Igw&RbDf~IW>>#~+)WMKH!3361HlE2^CiQ+D{>(Acowk|hHyS_Jr
zFQGOdg#{SKfg&a$x?-mPu7;ky`+fI~0}VnSpI$1C=?n=AUB1MGQ5xvA16H8CX9irJ
z!NAaPmCM%jRq54Nr8|A&>kO9v<lCE+)ciZ-bJnIve<q+}hy4>m85pD+KpE)3t}YSX
z+uCaDQq}pGL2~gEfpV|AL2^*gFz-+HW_B14*uVPd1Z-NtfP+6MnqlAtC&&dbV7m|$
zs8Fy#T?FJ`DEQKXVguNtP^k;TpyCS#s_1IjC;9Ljr`y|4M$3Zi_jL7hS?83{1ON+r
Bny3H(

literal 0
HcmV?d00001

diff --git a/docs/losses.md b/docs/losses.md
index c4126f0d..a85c00f6 100644
--- a/docs/losses.md
+++ b/docs/losses.md
@@ -1249,6 +1249,35 @@ losses.SupConLoss(temperature=0.1, **kwargs)
 
 * **loss**: The loss per element in the batch. If an element has only negative pairs or no pairs, it's ignored thanks to `AvgNonZeroReducer`. Reduction type is ```"element"```.
 
+## ThresholdConsistentMarginLoss
+[Threshold-Consistent Margin Loss for Open-World Deep Metric Learning](https://arxiv.org/pdf/2307.04047){target=_blank}
+
+```python
+losses.ThresholdConsistentMarginLoss(
+    base_loss, 
+    lambda_plus=1.0, 
+    lambda_minus=1.0, 
+    margin_plus=0.9, 
+    margin_minus=0.5,
+    **kwargs
+)
+```
+**Equation**:
+![threshold_consistent_margin_loss](imgs/tcm_loss_equation.png)
+
+**Parameters**:
+
+* **base_loss**: The final loss is calculated as `base_loss + tcm_loss`. 
+* **lambda_plus**: The scaling coefficient for the anchor-positive part of the loss. This is $\lambda^+$ in the above equation. 
+* **lambda_minus**: The scaling coefficient for the anchor-negative part of the loss. This is $\lambda^-$ in the above equation. 
+* **margin_plus**: The minimum anchor-positive similarity to be included in the loss. This is $m^+$ in the above equation.
+* **margin_minus**: The maximum anchor-negative similarity to be included in the loss. This is $m^-$ in the above equation.
+
+
+**Default distance**: 
+
+ - [```CosineSimilarity()```](distances.md#cosinesimilarity)
+     - This is the only compatible distance.
 
 ## TripletMarginLoss
 
diff --git a/src/pytorch_metric_learning/losses/__init__.py b/src/pytorch_metric_learning/losses/__init__.py
index ba653cda..543dea04 100644
--- a/src/pytorch_metric_learning/losses/__init__.py
+++ b/src/pytorch_metric_learning/losses/__init__.py
@@ -37,3 +37,4 @@
 from .triplet_margin_loss import TripletMarginLoss
 from .tuplet_margin_loss import TupletMarginLoss
 from .vicreg_loss import VICRegLoss
+from .tcm_loss import ThresholdConsistentMarginLoss
diff --git a/src/pytorch_metric_learning/losses/tcm_loss.py b/src/pytorch_metric_learning/losses/tcm_loss.py
new file mode 100644
index 00000000..045be5fe
--- /dev/null
+++ b/src/pytorch_metric_learning/losses/tcm_loss.py
@@ -0,0 +1,60 @@
+import torch.nn.functional as F
+
+from ..utils.loss_and_miner_utils import convert_to_pairs
+from .base_metric_loss_function import BaseMetricLossFunction
+from ..distances import CosineSimilarity
+from ..utils import common_functions as c_f
+
+class ThresholdConsistentMarginLoss(BaseMetricLossFunction):
+    """
+        Implements the TCM loss from: https://arxiv.org/abs/2307.04047
+    """
+    
+    def __init__(
+        self, base_loss, lambda_plus=1.0, lambda_minus=1.0, margin_plus=0.9, margin_minus=0.5, **kwargs
+    ):
+        super().__init__(**kwargs)
+        c_f.assert_distance_type(self, CosineSimilarity)
+        self.base_loss = base_loss
+        self.lambda_plus = lambda_plus
+        self.lambda_minus = lambda_minus
+        self.margin_plus = margin_plus
+        self.margin_minus = margin_minus
+
+
+    def get_default_distance(self):
+        return CosineSimilarity()
+
+    def compute_loss(self, embeddings, labels, indices_tuple, ref_emb, ref_labels):
+        ap, p, an, n = convert_to_pairs(indices_tuple, labels, ref_labels)
+
+        # calculate the similarities for positive and negative pairs
+        ap, p = embeddings[ap], embeddings[p]
+        an, n = embeddings[an], embeddings[n]
+
+        pos_sims = F.cosine_similarity(ap, p)
+        neg_sims = F.cosine_similarity(an, n)
+
+        # calculate the positive part
+        s_lte_m = (pos_sims <= self.margin_plus)
+        tcm_pos_num = ((self.margin_plus - pos_sims) * s_lte_m).sum()
+        tcm_pos_denom = s_lte_m.sum()
+        pos_tcm = 0 if s_lte_m.sum() == 0 else tcm_pos_num / tcm_pos_denom
+
+        # calculate the negative part
+        s_gte_m = (neg_sims >= self.margin_minus)
+        tcm_neg_num = ((neg_sims - self.margin_minus) * s_gte_m).sum()
+        tcm_neg_denom = s_gte_m.sum() 
+        neg_tcm = 0 if s_gte_m.sum() == 0 else tcm_neg_num / tcm_neg_denom
+
+        # add the components for final loss
+        tcm_loss = self.lambda_plus * pos_tcm + self.lambda_minus * neg_tcm
+        base_loss = self.base_loss(embeddings, labels, indices_tuple, ref_emb, ref_labels)
+        total_loss = base_loss + tcm_loss
+        return {
+            "loss": {
+                "losses": total_loss,
+                "indices": None,
+                "reduction_type": "already_reduced",
+            }
+        }
\ No newline at end of file
diff --git a/tests/losses/test_tcm_loss.py b/tests/losses/test_tcm_loss.py
new file mode 100644
index 00000000..cc1acfe0
--- /dev/null
+++ b/tests/losses/test_tcm_loss.py
@@ -0,0 +1,46 @@
+import unittest
+
+import torch
+import torch.nn.functional as F
+
+from pytorch_metric_learning.losses import ThresholdConsistentMarginLoss, ContrastiveLoss
+from pytorch_metric_learning.distances import CosineSimilarity
+
+from .. import TEST_DEVICE, TEST_DTYPES
+
+class TestThresholdConsistentMarginLoss(unittest.TestCase):
+    def test_tcm_loss(self):
+        torch.manual_seed(3459)
+        for dtype in TEST_DTYPES:
+            loss_func = ThresholdConsistentMarginLoss(
+                base_loss=ContrastiveLoss(
+                    distance=CosineSimilarity(),
+                    pos_margin=0.9,
+                    neg_margin=0.4,
+                )
+            ) 
+            embs = torch.tensor([
+                [0.00, 1.00],
+                [0.43, 0.90],
+                [1.00, 0.00],
+                [0.50, 0.50],
+            ], device=TEST_DEVICE, dtype=dtype, requires_grad=True)
+            labels = torch.tensor([0, 0, 1, 1])
+
+            # Contrastive loss = 0.4866
+            #
+            # TCM loss part:
+            # Only pair (2, 3) is taken into account for positive part
+            # Positive part = 1 * ( 0.9 - 0.7071 ) / ( 1 ) = 0.1929
+            #
+            # Only pairs (1, 2) and (1, 3) are taken into account for negative part
+            # Negative part = 1 * ( 0.7071 - 0.5 + 0.9429 - 0.5  ) / ( 2 ) = 0.325
+            #
+            # Sum of these losses -> 0.4866 + 0.518 = 1.0046
+            correct_loss = torch.tensor(1.0045).to(dtype)
+            
+            with torch.no_grad():
+                res = loss_func.compute_loss(embs, labels, None, embs, labels)
+                rtol = 1e-2 if dtype == torch.float16 else 1e-5
+                atol = 1e-4
+                self.assertTrue(torch.isclose(res["loss"]["losses"], correct_loss, rtol=rtol, atol=atol))

From 19ecba5fe65eea9c708467141f86b0bd708dd4c2 Mon Sep 17 00:00:00 2001
From: ir2718 <irep2718@gmail.com>
Date: Wed, 16 Oct 2024 23:40:06 +0200
Subject: [PATCH 05/16] run formatting

---
 .../losses/__init__.py                        |  2 +-
 .../losses/tcm_loss.py                        | 30 ++++++++++-------
 tests/losses/test_tcm_loss.py                 | 33 +++++++++++++------
 3 files changed, 43 insertions(+), 22 deletions(-)

diff --git a/src/pytorch_metric_learning/losses/__init__.py b/src/pytorch_metric_learning/losses/__init__.py
index 543dea04..6bd679a7 100644
--- a/src/pytorch_metric_learning/losses/__init__.py
+++ b/src/pytorch_metric_learning/losses/__init__.py
@@ -34,7 +34,7 @@
 from .sphereface_loss import SphereFaceLoss
 from .subcenter_arcface_loss import SubCenterArcFaceLoss
 from .supcon_loss import SupConLoss
+from .tcm_loss import ThresholdConsistentMarginLoss
 from .triplet_margin_loss import TripletMarginLoss
 from .tuplet_margin_loss import TupletMarginLoss
 from .vicreg_loss import VICRegLoss
-from .tcm_loss import ThresholdConsistentMarginLoss
diff --git a/src/pytorch_metric_learning/losses/tcm_loss.py b/src/pytorch_metric_learning/losses/tcm_loss.py
index 045be5fe..bf7a6aef 100644
--- a/src/pytorch_metric_learning/losses/tcm_loss.py
+++ b/src/pytorch_metric_learning/losses/tcm_loss.py
@@ -1,17 +1,24 @@
 import torch.nn.functional as F
 
-from ..utils.loss_and_miner_utils import convert_to_pairs
-from .base_metric_loss_function import BaseMetricLossFunction
 from ..distances import CosineSimilarity
 from ..utils import common_functions as c_f
+from ..utils.loss_and_miner_utils import convert_to_pairs
+from .base_metric_loss_function import BaseMetricLossFunction
+
 
 class ThresholdConsistentMarginLoss(BaseMetricLossFunction):
     """
-        Implements the TCM loss from: https://arxiv.org/abs/2307.04047
+    Implements the TCM loss from: https://arxiv.org/abs/2307.04047
     """
-    
+
     def __init__(
-        self, base_loss, lambda_plus=1.0, lambda_minus=1.0, margin_plus=0.9, margin_minus=0.5, **kwargs
+        self,
+        base_loss,
+        lambda_plus=1.0,
+        lambda_minus=1.0,
+        margin_plus=0.9,
+        margin_minus=0.5,
+        **kwargs
     ):
         super().__init__(**kwargs)
         c_f.assert_distance_type(self, CosineSimilarity)
@@ -21,7 +28,6 @@ def __init__(
         self.margin_plus = margin_plus
         self.margin_minus = margin_minus
 
-
     def get_default_distance(self):
         return CosineSimilarity()
 
@@ -36,20 +42,22 @@ def compute_loss(self, embeddings, labels, indices_tuple, ref_emb, ref_labels):
         neg_sims = F.cosine_similarity(an, n)
 
         # calculate the positive part
-        s_lte_m = (pos_sims <= self.margin_plus)
+        s_lte_m = pos_sims <= self.margin_plus
         tcm_pos_num = ((self.margin_plus - pos_sims) * s_lte_m).sum()
         tcm_pos_denom = s_lte_m.sum()
         pos_tcm = 0 if s_lte_m.sum() == 0 else tcm_pos_num / tcm_pos_denom
 
         # calculate the negative part
-        s_gte_m = (neg_sims >= self.margin_minus)
+        s_gte_m = neg_sims >= self.margin_minus
         tcm_neg_num = ((neg_sims - self.margin_minus) * s_gte_m).sum()
-        tcm_neg_denom = s_gte_m.sum() 
+        tcm_neg_denom = s_gte_m.sum()
         neg_tcm = 0 if s_gte_m.sum() == 0 else tcm_neg_num / tcm_neg_denom
 
         # add the components for final loss
         tcm_loss = self.lambda_plus * pos_tcm + self.lambda_minus * neg_tcm
-        base_loss = self.base_loss(embeddings, labels, indices_tuple, ref_emb, ref_labels)
+        base_loss = self.base_loss(
+            embeddings, labels, indices_tuple, ref_emb, ref_labels
+        )
         total_loss = base_loss + tcm_loss
         return {
             "loss": {
@@ -57,4 +65,4 @@ def compute_loss(self, embeddings, labels, indices_tuple, ref_emb, ref_labels):
                 "indices": None,
                 "reduction_type": "already_reduced",
             }
-        }
\ No newline at end of file
+        }
diff --git a/tests/losses/test_tcm_loss.py b/tests/losses/test_tcm_loss.py
index cc1acfe0..f0349a02 100644
--- a/tests/losses/test_tcm_loss.py
+++ b/tests/losses/test_tcm_loss.py
@@ -3,11 +3,15 @@
 import torch
 import torch.nn.functional as F
 
-from pytorch_metric_learning.losses import ThresholdConsistentMarginLoss, ContrastiveLoss
 from pytorch_metric_learning.distances import CosineSimilarity
+from pytorch_metric_learning.losses import (
+    ContrastiveLoss,
+    ThresholdConsistentMarginLoss,
+)
 
 from .. import TEST_DEVICE, TEST_DTYPES
 
+
 class TestThresholdConsistentMarginLoss(unittest.TestCase):
     def test_tcm_loss(self):
         torch.manual_seed(3459)
@@ -18,13 +22,18 @@ def test_tcm_loss(self):
                     pos_margin=0.9,
                     neg_margin=0.4,
                 )
-            ) 
-            embs = torch.tensor([
-                [0.00, 1.00],
-                [0.43, 0.90],
-                [1.00, 0.00],
-                [0.50, 0.50],
-            ], device=TEST_DEVICE, dtype=dtype, requires_grad=True)
+            )
+            embs = torch.tensor(
+                [
+                    [0.00, 1.00],
+                    [0.43, 0.90],
+                    [1.00, 0.00],
+                    [0.50, 0.50],
+                ],
+                device=TEST_DEVICE,
+                dtype=dtype,
+                requires_grad=True,
+            )
             labels = torch.tensor([0, 0, 1, 1])
 
             # Contrastive loss = 0.4866
@@ -38,9 +47,13 @@ def test_tcm_loss(self):
             #
             # Sum of these losses -> 0.4866 + 0.518 = 1.0046
             correct_loss = torch.tensor(1.0045).to(dtype)
-            
+
             with torch.no_grad():
                 res = loss_func.compute_loss(embs, labels, None, embs, labels)
                 rtol = 1e-2 if dtype == torch.float16 else 1e-5
                 atol = 1e-4
-                self.assertTrue(torch.isclose(res["loss"]["losses"], correct_loss, rtol=rtol, atol=atol))
+                self.assertTrue(
+                    torch.isclose(
+                        res["loss"]["losses"], correct_loss, rtol=rtol, atol=atol
+                    )
+                )

From 285cea0c89a7e12f5021bd67059d45901eefe4af Mon Sep 17 00:00:00 2001
From: ir2718 <irep2718@gmail.com>
Date: Mon, 28 Oct 2024 16:35:31 +0100
Subject: [PATCH 06/16] remove base loss

---
 src/pytorch_metric_learning/losses/tcm_loss.py | 8 +-------
 1 file changed, 1 insertion(+), 7 deletions(-)

diff --git a/src/pytorch_metric_learning/losses/tcm_loss.py b/src/pytorch_metric_learning/losses/tcm_loss.py
index bf7a6aef..9038c670 100644
--- a/src/pytorch_metric_learning/losses/tcm_loss.py
+++ b/src/pytorch_metric_learning/losses/tcm_loss.py
@@ -13,7 +13,6 @@ class ThresholdConsistentMarginLoss(BaseMetricLossFunction):
 
     def __init__(
         self,
-        base_loss,
         lambda_plus=1.0,
         lambda_minus=1.0,
         margin_plus=0.9,
@@ -22,7 +21,6 @@ def __init__(
     ):
         super().__init__(**kwargs)
         c_f.assert_distance_type(self, CosineSimilarity)
-        self.base_loss = base_loss
         self.lambda_plus = lambda_plus
         self.lambda_minus = lambda_minus
         self.margin_plus = margin_plus
@@ -55,13 +53,9 @@ def compute_loss(self, embeddings, labels, indices_tuple, ref_emb, ref_labels):
 
         # add the components for final loss
         tcm_loss = self.lambda_plus * pos_tcm + self.lambda_minus * neg_tcm
-        base_loss = self.base_loss(
-            embeddings, labels, indices_tuple, ref_emb, ref_labels
-        )
-        total_loss = base_loss + tcm_loss
         return {
             "loss": {
-                "losses": total_loss,
+                "losses": tcm_loss,
                 "indices": None,
                 "reduction_type": "already_reduced",
             }

From 31eeaaad1a4313380cbb96b83c1b1560e5f666fe Mon Sep 17 00:00:00 2001
From: ir2718 <irep2718@gmail.com>
Date: Mon, 28 Oct 2024 16:41:40 +0100
Subject: [PATCH 07/16] remove base loss from docs

---
 docs/losses.md | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/docs/losses.md b/docs/losses.md
index a85c00f6..b8ced202 100644
--- a/docs/losses.md
+++ b/docs/losses.md
@@ -1252,9 +1252,10 @@ losses.SupConLoss(temperature=0.1, **kwargs)
 ## ThresholdConsistentMarginLoss
 [Threshold-Consistent Margin Loss for Open-World Deep Metric Learning](https://arxiv.org/pdf/2307.04047){target=_blank}
 
+This loss acts as a form of regularization and is usually combined with another metric loss function.
+
 ```python
 losses.ThresholdConsistentMarginLoss(
-    base_loss, 
     lambda_plus=1.0, 
     lambda_minus=1.0, 
     margin_plus=0.9, 
@@ -1267,7 +1268,6 @@ losses.ThresholdConsistentMarginLoss(
 
 **Parameters**:
 
-* **base_loss**: The final loss is calculated as `base_loss + tcm_loss`. 
 * **lambda_plus**: The scaling coefficient for the anchor-positive part of the loss. This is $\lambda^+$ in the above equation. 
 * **lambda_minus**: The scaling coefficient for the anchor-negative part of the loss. This is $\lambda^-$ in the above equation. 
 * **margin_plus**: The minimum anchor-positive similarity to be included in the loss. This is $m^+$ in the above equation.

From 60c0583e4c0c94d654347983ecbdabff2383d576 Mon Sep 17 00:00:00 2001
From: ir2718 <irep2718@gmail.com>
Date: Mon, 28 Oct 2024 17:47:18 +0100
Subject: [PATCH 08/16] fix test

---
 tests/losses/test_tcm_loss.py | 20 ++++++++++++--------
 1 file changed, 12 insertions(+), 8 deletions(-)

diff --git a/tests/losses/test_tcm_loss.py b/tests/losses/test_tcm_loss.py
index f0349a02..586f9107 100644
--- a/tests/losses/test_tcm_loss.py
+++ b/tests/losses/test_tcm_loss.py
@@ -5,6 +5,7 @@
 
 from pytorch_metric_learning.distances import CosineSimilarity
 from pytorch_metric_learning.losses import (
+    MultipleLosses,
     ContrastiveLoss,
     ThresholdConsistentMarginLoss,
 )
@@ -16,12 +17,15 @@ class TestThresholdConsistentMarginLoss(unittest.TestCase):
     def test_tcm_loss(self):
         torch.manual_seed(3459)
         for dtype in TEST_DTYPES:
-            loss_func = ThresholdConsistentMarginLoss(
-                base_loss=ContrastiveLoss(
-                    distance=CosineSimilarity(),
-                    pos_margin=0.9,
-                    neg_margin=0.4,
-                )
+            loss_func = MultipleLosses(
+                losses=[
+                    ContrastiveLoss(
+                        distance=CosineSimilarity(),
+                        pos_margin=0.9,
+                        neg_margin=0.4,
+                    ),
+                    ThresholdConsistentMarginLoss()
+                ]
             )
             embs = torch.tensor(
                 [
@@ -49,11 +53,11 @@ def test_tcm_loss(self):
             correct_loss = torch.tensor(1.0045).to(dtype)
 
             with torch.no_grad():
-                res = loss_func.compute_loss(embs, labels, None, embs, labels)
+                res = loss_func.forward(embs, labels)
                 rtol = 1e-2 if dtype == torch.float16 else 1e-5
                 atol = 1e-4
                 self.assertTrue(
                     torch.isclose(
-                        res["loss"]["losses"], correct_loss, rtol=rtol, atol=atol
+                        res, correct_loss, rtol=rtol, atol=atol
                     )
                 )

From c6d5b525baa956e4f1f03720a9df68767501a7b5 Mon Sep 17 00:00:00 2001
From: KevinMusgrave <tkm45@cornell.edu>
Date: Sat, 2 Nov 2024 20:01:41 +0000
Subject: [PATCH 09/16] bump version

---
 src/pytorch_metric_learning/__init__.py |  2 +-
 tests/losses/test_tcm_loss.py           | 10 +++-------
 2 files changed, 4 insertions(+), 8 deletions(-)

diff --git a/src/pytorch_metric_learning/__init__.py b/src/pytorch_metric_learning/__init__.py
index fab833f3..2614ce9d 100644
--- a/src/pytorch_metric_learning/__init__.py
+++ b/src/pytorch_metric_learning/__init__.py
@@ -1 +1 @@
-__version__ = "2.6.1"
+__version__ = "2.7.0"
diff --git a/tests/losses/test_tcm_loss.py b/tests/losses/test_tcm_loss.py
index 586f9107..e961bab1 100644
--- a/tests/losses/test_tcm_loss.py
+++ b/tests/losses/test_tcm_loss.py
@@ -5,8 +5,8 @@
 
 from pytorch_metric_learning.distances import CosineSimilarity
 from pytorch_metric_learning.losses import (
-    MultipleLosses,
     ContrastiveLoss,
+    MultipleLosses,
     ThresholdConsistentMarginLoss,
 )
 
@@ -24,7 +24,7 @@ def test_tcm_loss(self):
                         pos_margin=0.9,
                         neg_margin=0.4,
                     ),
-                    ThresholdConsistentMarginLoss()
+                    ThresholdConsistentMarginLoss(),
                 ]
             )
             embs = torch.tensor(
@@ -56,8 +56,4 @@ def test_tcm_loss(self):
                 res = loss_func.forward(embs, labels)
                 rtol = 1e-2 if dtype == torch.float16 else 1e-5
                 atol = 1e-4
-                self.assertTrue(
-                    torch.isclose(
-                        res, correct_loss, rtol=rtol, atol=atol
-                    )
-                )
+                self.assertTrue(torch.isclose(res, correct_loss, rtol=rtol, atol=atol))

From 3b371662276f58ea05b4c31b7561aa20251dad8f Mon Sep 17 00:00:00 2001
From: KevinMusgrave <tkm45@cornell.edu>
Date: Sat, 2 Nov 2024 20:04:43 +0000
Subject: [PATCH 10/16] remove version restriction for numpy

---
 setup.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/setup.py b/setup.py
index 0818062c..b8ca0431 100644
--- a/setup.py
+++ b/setup.py
@@ -39,7 +39,7 @@
     ],
     python_requires=">=3.0",
     install_requires=[
-        "numpy < 2.0",
+        "numpy",
         "scikit-learn <= 1.4.1.post1",
         "tqdm",
         "torch >= 1.6.0",

From c1aba4c5a72d7b0c1cc214f0f2c0b2d47c027d7f Mon Sep 17 00:00:00 2001
From: KevinMusgrave <tkm45@cornell.edu>
Date: Sat, 2 Nov 2024 20:07:04 +0000
Subject: [PATCH 11/16] update versions in base_test_workflow

---
 .github/workflows/base_test_workflow.yml | 6 +++---
 setup.py                                 | 2 +-
 2 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/.github/workflows/base_test_workflow.yml b/.github/workflows/base_test_workflow.yml
index 02367536..c697a94c 100644
--- a/.github/workflows/base_test_workflow.yml
+++ b/.github/workflows/base_test_workflow.yml
@@ -18,8 +18,8 @@ jobs:
             pytorch-version: 1.6
             torchvision-version: 0.7
           - python-version: 3.9
-            pytorch-version: 2.3
-            torchvision-version: 0.18
+            pytorch-version: 2.5
+            torchvision-version: 0.20
 
     steps:
     - uses: actions/checkout@v2
@@ -30,7 +30,7 @@ jobs:
     - name: Install dependencies
       run: |
         pip install .[with-hooks-cpu]
-        pip install "numpy<2.0" torch==${{ matrix.pytorch-version }} torchvision==${{ matrix.torchvision-version }} --force-reinstall
+        pip install torch==${{ matrix.pytorch-version }} torchvision==${{ matrix.torchvision-version }} --force-reinstall
         pip install --upgrade protobuf==3.20.1
         pip install six
         pip install packaging
diff --git a/setup.py b/setup.py
index b8ca0431..66326d46 100644
--- a/setup.py
+++ b/setup.py
@@ -40,7 +40,7 @@
     python_requires=">=3.0",
     install_requires=[
         "numpy",
-        "scikit-learn <= 1.4.1.post1",
+        "scikit-learn",
         "tqdm",
         "torch >= 1.6.0",
     ],

From 892f8f724decf3d3576c97864c47d56ef36d3935 Mon Sep 17 00:00:00 2001
From: KevinMusgrave <tkm45@cornell.edu>
Date: Sat, 2 Nov 2024 20:29:29 +0000
Subject: [PATCH 12/16] minor change to tests

---
 tests/samplers/test_tuples_to_weights_sampler.py | 2 +-
 tests/utils/test_inference.py                    | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/tests/samplers/test_tuples_to_weights_sampler.py b/tests/samplers/test_tuples_to_weights_sampler.py
index abe9a56f..96eecf7a 100644
--- a/tests/samplers/test_tuples_to_weights_sampler.py
+++ b/tests/samplers/test_tuples_to_weights_sampler.py
@@ -22,7 +22,7 @@ def test_tuplestoweights_sampler(self):
 
         eval_transform = transforms.Compose(
             [
-                transforms.Resize(128),
+                transforms.Resize(256),
                 transforms.ToTensor(),
                 transforms.Normalize(
                     mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]
diff --git a/tests/utils/test_inference.py b/tests/utils/test_inference.py
index 7a6f4ee1..4424d0c3 100644
--- a/tests/utils/test_inference.py
+++ b/tests/utils/test_inference.py
@@ -39,7 +39,7 @@ def setUpClass(cls):
 
         transform = transforms.Compose(
             [
-                transforms.Resize(64),
+                transforms.Resize(256),
                 transforms.ToTensor(),
                 transforms.Normalize(
                     mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]

From e5117dbeb4d3eeb586778b8dfb22ef1fd51634b0 Mon Sep 17 00:00:00 2001
From: Kevin Musgrave <tkm45@cornell.edu>
Date: Sat, 2 Nov 2024 17:27:38 -0400
Subject: [PATCH 13/16] Update test_metric_loss_only.py

---
 tests/trainers/test_metric_loss_only.py | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/tests/trainers/test_metric_loss_only.py b/tests/trainers/test_metric_loss_only.py
index 10f85242..b9f25cd3 100644
--- a/tests/trainers/test_metric_loss_only.py
+++ b/tests/trainers/test_metric_loss_only.py
@@ -66,6 +66,8 @@ def test_metric_loss_only(self):
             dataset_folder, train=True, download=True, transform=train_transform
         )
 
+        train_targets = np.array(train_dataset.targets)[subset_idx]
+
         train_dataset_for_eval = datasets.CIFAR100(
             dataset_folder, train=True, download=True, transform=eval_transform
         )
@@ -79,6 +81,7 @@ def test_metric_loss_only(self):
             train_dataset_for_eval, subset_idx
         )
         val_dataset = torch.utils.data.Subset(val_dataset, subset_idx)
+        
 
         for dtype in TEST_DTYPES:
             for splits_to_eval in [
@@ -114,7 +117,7 @@ def test_metric_loss_only(self):
                 optimizer_dict = {"trunk_optimizer": optimizer}
                 loss_fn_dict = {"metric_loss": loss_fn}
                 sampler = MPerClassSampler(
-                    np.array(train_dataset.dataset.targets)[subset_idx],
+                    train_targets,
                     m=4,
                     batch_size=32,
                     length_before_new_iter=len(train_dataset),

From 5472652e271769c500e5998939383b301161c312 Mon Sep 17 00:00:00 2001
From: Kevin Musgrave <tkm45@cornell.edu>
Date: Sat, 2 Nov 2024 17:30:41 -0400
Subject: [PATCH 14/16] Update base_test_workflow.yml

---
 .github/workflows/base_test_workflow.yml | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/.github/workflows/base_test_workflow.yml b/.github/workflows/base_test_workflow.yml
index c697a94c..d23f6859 100644
--- a/.github/workflows/base_test_workflow.yml
+++ b/.github/workflows/base_test_workflow.yml
@@ -14,12 +14,12 @@ jobs:
     strategy:
       matrix:
         include:
-          - python-version: 3.8
-            pytorch-version: 1.6
-            torchvision-version: 0.7
-          - python-version: 3.9
-            pytorch-version: 2.5
-            torchvision-version: 0.20
+          - python-version: "3.8"
+            pytorch-version: "1.6"
+            torchvision-version: "0.7"
+          - python-version: "3.9"
+            pytorch-version: "2.5"
+            torchvision-version: "0.20"
 
     steps:
     - uses: actions/checkout@v2

From 206e8396c72113b0137e101500444c2ac8f72ceb Mon Sep 17 00:00:00 2001
From: KevinMusgrave <tkm45@cornell.edu>
Date: Sat, 2 Nov 2024 21:41:12 +0000
Subject: [PATCH 15/16] Revert "minor change to tests"

This reverts commit 892f8f724decf3d3576c97864c47d56ef36d3935.
---
 tests/samplers/test_tuples_to_weights_sampler.py | 2 +-
 tests/utils/test_inference.py                    | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/tests/samplers/test_tuples_to_weights_sampler.py b/tests/samplers/test_tuples_to_weights_sampler.py
index 96eecf7a..abe9a56f 100644
--- a/tests/samplers/test_tuples_to_weights_sampler.py
+++ b/tests/samplers/test_tuples_to_weights_sampler.py
@@ -22,7 +22,7 @@ def test_tuplestoweights_sampler(self):
 
         eval_transform = transforms.Compose(
             [
-                transforms.Resize(256),
+                transforms.Resize(128),
                 transforms.ToTensor(),
                 transforms.Normalize(
                     mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]
diff --git a/tests/utils/test_inference.py b/tests/utils/test_inference.py
index 4424d0c3..7a6f4ee1 100644
--- a/tests/utils/test_inference.py
+++ b/tests/utils/test_inference.py
@@ -39,7 +39,7 @@ def setUpClass(cls):
 
         transform = transforms.Compose(
             [
-                transforms.Resize(256),
+                transforms.Resize(64),
                 transforms.ToTensor(),
                 transforms.Normalize(
                     mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]

From 63d27dd23b0cf2eb0c8b3933a7ee6782376570fa Mon Sep 17 00:00:00 2001
From: KevinMusgrave <tkm45@cornell.edu>
Date: Sat, 2 Nov 2024 21:41:21 +0000
Subject: [PATCH 16/16] Revert "Update test_metric_loss_only.py"

This reverts commit e5117dbeb4d3eeb586778b8dfb22ef1fd51634b0.
---
 tests/trainers/test_metric_loss_only.py | 5 +----
 1 file changed, 1 insertion(+), 4 deletions(-)

diff --git a/tests/trainers/test_metric_loss_only.py b/tests/trainers/test_metric_loss_only.py
index b9f25cd3..10f85242 100644
--- a/tests/trainers/test_metric_loss_only.py
+++ b/tests/trainers/test_metric_loss_only.py
@@ -66,8 +66,6 @@ def test_metric_loss_only(self):
             dataset_folder, train=True, download=True, transform=train_transform
         )
 
-        train_targets = np.array(train_dataset.targets)[subset_idx]
-
         train_dataset_for_eval = datasets.CIFAR100(
             dataset_folder, train=True, download=True, transform=eval_transform
         )
@@ -81,7 +79,6 @@ def test_metric_loss_only(self):
             train_dataset_for_eval, subset_idx
         )
         val_dataset = torch.utils.data.Subset(val_dataset, subset_idx)
-        
 
         for dtype in TEST_DTYPES:
             for splits_to_eval in [
@@ -117,7 +114,7 @@ def test_metric_loss_only(self):
                 optimizer_dict = {"trunk_optimizer": optimizer}
                 loss_fn_dict = {"metric_loss": loss_fn}
                 sampler = MPerClassSampler(
-                    train_targets,
+                    np.array(train_dataset.dataset.targets)[subset_idx],
                     m=4,
                     batch_size=32,
                     length_before_new_iter=len(train_dataset),