From ead1e490b5dce7af54875580c174f32f0fea135b Mon Sep 17 00:00:00 2001 From: Xiaoyu Zhang <1182563586@qq.com> Date: Wed, 8 Jul 2026 02:48:13 +0800 Subject: [PATCH] [Doc] Add LongCat 2.0 FP8 cookbook (#30320) Co-authored-by: Zijie Xia --- docs_new/cards/logos/meituan.png | Bin 0 -> 19451 bytes .../autoregressive/Meituan/LongCat-2.0.mdx | 197 +++++++++++++++++ docs_new/cookbook/autoregressive/intro.mdx | 6 + docs_new/docs.json | 6 + .../longcat-2.0-benchmarks.jsx | 10 + .../configs/meituan-longcat/longcat-2.0.jsx | 203 ++++++++++++++++++ 6 files changed, 422 insertions(+) create mode 100644 docs_new/cards/logos/meituan.png create mode 100644 docs_new/cookbook/autoregressive/Meituan/LongCat-2.0.mdx create mode 100644 docs_new/src/snippets/configs/meituan-longcat/longcat-2.0-benchmarks.jsx create mode 100644 docs_new/src/snippets/configs/meituan-longcat/longcat-2.0.jsx diff --git a/docs_new/cards/logos/meituan.png b/docs_new/cards/logos/meituan.png new file mode 100644 index 0000000000000000000000000000000000000000..a7e1edd80e64dbb1e31d32b96097da239bc196e2 GIT binary patch literal 19451 zcmeFY^;=ZY8aBKENhJhH89*s%l#p(uySqCjr9lY+1?euOW9U&D>24Tl>F(}tkLR5C z{U5&Togb8IX73fxdh&km6|Sr(g^fjq1polHjI@L*0HBlt05UcP8u$(0lj8;ekZY5X z5LN$>zBiAaZf=oqapWrA3Ej_3$xIJ334O6){)mt;h6~uS176e1+v(0vT4))~Yus)^ z8|&-0iz})upFX25G^TojkA_CKLEg}?tZ=X)puxq-Ge17BHt$!meKBcTUs6)8O<4H7AMjYIzAEZ0!eskM-BC9NhyMDLMk4?~X51#>e|r zWeS*&_?EcPL{?#^W+Vm+t zw^H|sMir&;$J6u`X1vOY2mkdpR6g}%NbD0mIGPjg;Lv$)t`VzI~dR+vMy)*Wm1!zww2j zUXj;EX5*@xa`vbZwYu8Mt^&eibvDxSELG~|Aa$*5Qk{B>ripeWqjv`F5`im8Sqp1k z443YD;qTZ&T^6`2e$1ciu2D5&7|Z&#H&@GbL1uTMzu=*t#XZldP=t9qtgR`!toPwM zN5e0`*Vy8q{WtA~$$uUtJ+}w*tpWS(1QKKI;xnAAZMwj&`g`G5`1DtOW(_si2!)=R0hw5$A zV$TWL(!Gu&S<}yp35YjXC$NeWTqPoM;^l(WZws@7#$$VGYhG!VL?99Ls_*<@f6{sU zL%J>3&O*&bipD`M9@l(js_j^6>il87z;pLL>)^q(J7WGOv#`LX)Z)mprTK}LCl?+_uI-_V!nGkIBN=QHEQ=v-y2TdMd72Pr5~6-J?TBJf;x zGc0F~A4XGc(@ofoUryeW>9;VW=d^g%TYhhR;z{XUTHd1G4<*w_&um2ck03cKuWv;% zE(%R6KXn)sAoB01J;UD`|#hbiO2 z<^|*or{2vTxuYo8nDffUKQAMW;*{O-c^nlIap_UGRGStJES zT&XIMMdJLd`B6goGJR}4H@cxZ$CZzVGLFyfSz}mEy_yrV@S`%)Hyu#B(ujizXRCJY zwc1!tHLnjEJs*;UuMbL~O&3cG_u?mdpFYz~0wWR#=ON!7(D|mnk2y6>rtzj2mCb8~ zY5h(~YDz(~FNa&1kvB2H+>mj`ug4irG9lNlAdW%q*sJB(XmS2{JK$kGH0`ZDi+8$_WO96Wim83=$5N>F7?$u$!w%v%mg)e4LrDhZqn zedW8oIvdMR)I0NX@Y(JC@ve0>2F(pT&t{s14Nr~|*zVR0d_1(mW`tjVL;~>Gz;pZj zqO92V)#RrwPovtyU8Vc^*oHBiw>qz`2&a@#qbvpCJqObKy}qMtx9uC6Mu!KBbI#?w zBvs7e$$zkbN^!(C-rHu#H@HC0QgB&(?$REH8_N^rOh|%NHvbUyPXJOfa9YJ*QEAe; z$CC&4{^LGEwa0c>Vt8A4OWRoGK;SRLaX)F_j^7WyD+}OXBS;bmSPy%AXDB&d0_ znZi|LH=eQam!#HSce*jp+xHj5hF1&di`A}5yt!joz>z4pa*{$V_w`}f2I1@Ft+yHP zk5~KM@1)_dSb`=8k&=^76=S0b(hIJO_sD!o-`%&l^Ahv)L*oi zj8@m${k%7ztwC<>_?!L$kT;IYOETc}+kf!AuSvj{eS^gUujGxv=*aP@u#kJDsqVez zL!o`XVstPgOF#!ClYojva)4_nM-w3tr8g0CyZ4ZOvsok4+qou7!iTd{iv0fRBmDgF z#3nz_t;!93uY;lz?p3B>@U4I+m}r5R-<(;lb{E-hE(dIEuuFtY2tN~dKW)&)Vsy2% zO7c4w@4k!VHZiqk|IfC*PYaWnAk#-e*{*l>s#d9Vu~v)pf^(x}hE{f)Q# z>Lb9|{wcVe^iY(R`{TNTtoNoDd26vcZK2zYBjy3oa%)rqcftEf&HGh#1z|BM0Yal` zr~pfJD;8i@=+84dzoqMRH|v%4RSc6lT+D-SM4wr1mpwx4JxkrSMYS+@9>0kV+}I$E z5McMXOhyfSz4|MWPTlLy!gbiNisnMTY=p-R6)iP(=TEO#ix%(QX)jwH=>F*yPR{`m ze$cF2wf$PR-*Ai4a1`Y2Y8mqFpvb^I#EVNr#@{MGW>UFa*BZFFjF-V=dXmx)t|=U` zFdlO44XDt8jbL}W-%Nb}&COk`!zd|omTVJA<5#P>UeePEaNnN)!S7K|IM2iL$-fwn>f)A8ux$%YoVD<@x9w2ES@g9rS z?+#JQI)|yYP}ajT%&kxchhd#lOZQgkfBYt=3*6RD59)CU(Ev^g(7kktG;JS8$XN;t zCsd5b(7K_a>3-gj`xEni9^UM>F9$x_{*(~-f+=K(Wc(#J7S?!L$0790J4V~|$m?;< zH)XL!mrnngEj7cGe!fISr0gCn;A=P1Gl28@(~DB_)+bW3 zGm`*hUqLy3y;ZJd9=G3`SmEPeAC%4_^v=pR3#Cc@R2gJxpZgkv1P$HO2H!l zAo~Qqfo%>K(oho4@%ep9nqnrG*Xl_=YeA1v*8QHjVOJ_4lQ)G|j(9EqpFS|X1l>G} zjk`uh1&^oL_wm|K(u}*t+3wSt=2HI8GQGqu7_OV3pmkc4BSNe5eesYQn7L+`VzdO>afMHTlzZK%{?31PdTL3)0mKem_;CZhF){duM$3}dAjuRwx5-VHdDkyk7 zpJrn)GgDUK6b6Wt{2SYT=+w=#o<8AJyl7MRE4boLN$=)9EG7LJUbF9SQ2||SbX8!4 z%0$}%R`tQq5_+lP?)W_L+&H|+P1Ivl!U+wiA;ZZBvV$NWS^8r`Tsh=uFgmNKh0KDu zb51m)Q<4lUn=?1RA_Wq){RS06{QoAJOncdXU@zLbp8Go1)K_0MY3e= zIGO%b4r(tk65Gy@P&pC)6kaZ`kD9tV;oz@fh+~O&wol<;akcy#wyTw$A@Oo6HzRE< zqoQ~mh0gmEkj;fi`GZ^_P>wqO*Wx#9qc2FT%k%AeHGk}O0V&6#H?GG-HN1E9gc^uF!eEA>F|6PYN z-JTnV7q+$9WS+y=x|~5WA>cfLNE}5Vqy;lCo{2n^yf--xA*p%`co3%eVBK#2puM>83WegTV|Ib^pv<~CWsNiT4`m?fh zKZ6{Jbv~ZYj$LD@9GMwZ#7DGke^6l-jdGr2C^9NBS*a2gTY=zMw zdQ%gc19tU1ZvmNoccpIW&Cb!;olEvl?~edBemD(2Y#==~UdOfirQxeJm!5spKq5r7 zz^zUxmbT+;F?AX*0Apo~Dr@&guWO?H9`->UlXLPzFjCBWSvOXd&U{N+ytOI!y)27d z%UZ0JXp|+1zF+{1)e-s-%9(8w!Zo`_fmJI0ReAEVKwSF*a%dhHJNzywc zNPE(hqQ@^`T&C2N!B5TNF%wg<9L^DCmgX*0;rL0&d@fast4Vj9*3Zd~_GUGYw{Cfi zS9RM8gB##vLfB^As7J<`TX<@)Ln=;@bs?6WI>eP`YYFS?{HXDP1L*T>YAZ=lSn8AU@jXJOFm$@0ZtwKN`miRlT+mpO^GORa_r zNwv>Qg3hE#M{vcqdU&d=9p9zbm?dtHF1|JurCGyoS6C^$H=M7ief4~5!(yqiG@gmC z#q0ZG1rQj6h^ldV_kWQ%a9wi^tZJ*02kWZLoFw((Z?J8_`Z?NGLF>SOGjQ&1qX|FB zIM&%=y1IoKw|t)i9wN>iR;H=N&#ymSk@qTGcrQVf7tbm z)+ODH*OR#rLB3*c#t{qw*xw+$k15~nS)Nmi5$u2=%~G5iVyc^V?^VcusCquPfh{IK8!S`itnmB{* zkZHfc#!2aplP)V>a<#K51sa0q12?FF=LlCuR}hr%Pg?d=%hHS_vulSp*>qlI|3F2M zjWHMq24tUsP5}6v;{9{oDxFK`e*Yjb;@L+SZmDIJdjS*E!8Dr@2p!c<6WUE9 zi}X{2?h{Wf#FXlOA>vtN&==EQc;a^?ERR;;*pxR3E< zzL7W^!|AGGf4SW1oPgpPGzQa@+oMm_RrmqIM@@?vlg=U#w6Ox){W! z`(rIlx?8tdF(!HP<~T2aPi1|uFIl;;U23H>_EkhK^~&d@3)1h5FQe>w_X1{^;C_C0 z73bMI+Q+w6ijdPL01$y7W|#|?K<^p$QgoHSYV*@WRToQdi2+4LWEubbb0 z#KDyN{dt2g%F3ehd;j2HWsEZ11d38I!-tdwA|P-Hfge16f*Iv%y!R?(c`|eqCk!pM z3dfZ-Y`w0NtAa7KT1eDrX{i+5KD0dgiU|dXKmdCOPJU(8IoOv=xlvu=moT`oSM(?o?5WINUQ zCDoD3XbmM08?YOU$-vX80HB5v)W8}HDfsURTn~7};P1Opdr7~}DV~WbH*J?^?&tutbuSrw7VY@$tLZc@16b}n>*Q-&YF^(@Np=E(WQ zWiVzbmtUzNpUP?@{&GHZlv%ALBQNa6n+HY<6KrmY%a;QJ`49_7mdn+~8mM|L&H@>K z;(}W zaSWw0Z^cuK>L=EQzm%b~*J-GuvU^J?&*fb4QopGEclWxCr7^vkL|RXlMK&hQi1xe|^&#orB&eVY(eSWr z<+IeC75rxRS9W|Bkq6Na9uTq!`*7dGvPdp3ifjMemd&#RzicAyvnG7(t7IpiN#DBd z?U0v#&7V1>O%-O3m9n&Lkav_{{JO7B5^}1)Z4XNfkB@SG_Y%M8^ky=r>*&LC0O%q` z9DX3hu-&(@ZFBv$;xQ~;x}cSVR5H75j)eb-EvaUD`l^BQ=l4{2I$;f#8gug+2WZUbTc68r(dh6Pc z(kY#rWHmGkkugfSn2I`;T#U+*pVI?W{~A>PG_trK)nP#;-qh;@?E$Hv%+>&{sa732P=kxmQ{XE>crdhoi@a`chiBNt{Y}&~ zY-vn3(*tAb#(iQ+B<{U`?7@viyK;-L&`Le_PW` zj{VYLb@!=3PFLhr(6Ql!1Q5V>=TzbaZ<<~>l_)7?+0j|_gAvQs%5N+6&;~5e|BX^4 z=d5jTQZwC9-}U~cXjr}rHK}e;e*b*~gDIuG$FU=kv~*Z-*NdpK6>s)7;Bj*;n&ud*a#noT zQTmC$dG;~TD#B@i3`)7!Ix#oaW4Hd%4xX9;v1V8z&c(UzQ*zskkt0o9mlU@Xcg!h2 z>pX~>!ThhXS5ZPqzsXp2UZz?^XJjQvo>ZwagjrsJR!)ZnT6t>vjyC?O*YQFLaW4~F z221#hb^55ERG)K8I9?bv+x*ZESfP7w&>sxX5nK8=O?q`JoKAIOnKzJH@V@9UCC@}( zk7&66^G1m2TbXa;cw}BV0*(2@>Y4;lX=R6Jw_4lf1MeU6=h}J8UVCZ{VB-CsS&NL( zVwjWTV$9YpXcgTW@Z;j%9Qtt z7)P_W^h1n9f^r{WhB#@t`JX4EXLx%U=@k}>Lshr;nSSfV&fjvT&tCkAYannLVwlG# zM5)ZM^sWK|H4$kl(bI0Ky#MP_Bs>?v_GRC1|A&!^o%V&Hs)6ITuK5M4U@tURzMU2j zdD3BS32v3-aXs@O+y0Ci#6?hon5kN`ud97QDoA1w(#05`+QTfJW#ip`WcoB!4KUA?c(o&U8J^Nuf=k!bAM zlWXa|R-l2@?KJyUdHEE1WIW*waG#0KO?Nl_Ieh`*JR;=(G7O7qN3PN?_Gw8nUl(XD zemky5Dv~RyR6359%giZ#RF3jnhR?a`m@6MLTQQKI^6&I0IWE#>QsP_@y`G$S6}yoR zr>?{K>5cyg2qgK}B5XMLjiNg0F1mMF+M-+Vn;yQ^7g{FC+xDzaeAe()h|595C$JRi zT_|RotMRWa)R$;Z2j^ zk*htj7To1mjRagWA%cVH%+XnGz(e-#!bVEqYjvvhSGqKNOpn&>4BFk;&-QE6JmXgt z9)5|DOEpZ%?rw%7`B7|SaO*f;T~DiSoyn7YA;8XuWI%84JsZ|WYwyC3%S;}7INiPz;O)m?L|Im7KWQDqTZG9-JqmTjx81et--!R4@e`?L) z3st@s8~m9Q1KZB0R%E(t-n!AbRZ!>zhwYu$9*2Oy%wL8?S5OFuRwa-vnY?DFuh;w# zxmR@{Qu}_=SUky2u3XkCj49xzO$UF~?|syAq45JdS-|953zxSbp4WQ#PkASqtlwv- zO&8zz6!xu2RF?J((B2(TI+pka1kXnHa15n0%SHEwZ$v}+o$rVpW(_FL2dxwpj{Q-A zCyy19fb8BpwI~{bAZ>4~H)y1N_8&E`?PFAF1&WQopD?Il>In#Qx^k>!FLb;jtnL!6 zSC}3OzCE;>HD60@Hkp?)N`0#@r1kv=#^MZD9f&M7o8Vwh%_{9HMqM8Pz+MO<$_~Jo zugZ^fcxLh%dJgVWQ-X->8Pa_irayq z2KD6b2R{m-;uh8Y+4h*aZfU(t_2^~wRFMZDwCiA()(CLTN|HR}!U?O0~-hXzUks-(s%!>ixKib>Ps z9Aicp#e+d~oD7;-uH6%mt~7=Av`?W6|K`dR z#Z7Q$hULQ9X!Bc8KY~F0;JLOG){*kqXX&?q-*!HcnXDiSZpWBu$B@0gsWmN3m}>Ew zbvdz;O0p9pt)|)$9f{p}u!P>4y7u?2T4bmd3 z{N+aVJ#*pWly=jvA^D8YyIK2zz)KKv2Hr`whiUd!@oG^RWhD`}c-7tRFZfxK95}bx zp53=UX5nkpXy0ouxn5m6_+@l|AvHU&G8Og{#*`tRDzAIo`U73TT?4yTc)Nc}VDC%s z@tl=HoNAU8=Hmc`ep&!45KRH7ltsF+P++ zR3WNVP`$D#9Pf`}5{kO9adjMH_|{rG$!=O`KnDUuWq+Az1VLujq=EZclr?S+Ui9Yg z(Q?>(7S_eBQ$pfYn0qbBuiB)PzTq~zhWoCvL@JEDpQj#vTo$Hj(6v!tUY11MbR={& zes@DVoA6DdjcCy+Ys8ih#gl#&HtX5z;JFXY!1x?GcX7S)csS!Fv9M2lJlEf-Hyqms zqh8fF9*Jx4fVrEF1aScv-_ak%6kZMoLty|2Mf7k;hz6<2WL(r@t4uYr8jl5Wd(egM zeqmCQC8F)f@;mbs*M07Br_`q#_PbVRfE=z;A80dcHMvD$wzb<>sv9sC=+-vaUHR8- zx?l83mHWt_Lkgw4S97ntO4yEyeseC(jLmNJTD3GAULQ|xZ)X!Ko>2CuKfC;!MwC{o z7M)L37JWeHpi>7^x_5r?Cy);FzH|TlYhnEhL4!oi(%7qwL1bVC7a_w}*dly$nt9_e zOwxy=gj$!Q=Ld0=_qH23&yesJ>3JUT66;JuiT!6&R?I@+Xs z4vZLH(xU**+PbgDWp!a3CT)XG_cdCXC#el<^ADbR_S3H$U$pkAi`%tpEFNv69uFVr zzRZL|^q!;b`sA3a!F>1P===NGQcaNa(JGc6F=F?5^Uo)SQ5?j<{Vq-F4vQPsK;(`{ z7^6fy#6V6!>64uujg}u-$`__t7~z_lk&YGLo1GK*hSx7SS$-tFmx6Jz$C)L>GH-0s z;OEv@>DM37!Q6G6Wp+T>zMJg!D6UpGiFNmzn$mWxstkNI5A0NtPKTeCM*^chTE9oXY5QJA38wn^(=e zuk_0!KSPI07vCvm2=<`&otF}5jqOz?cJ90>ImGcr_qz_|sq`Bef|AGR0gIY@S40-fIgnQcl`wqu^VT=a zccWl-twDJ4Mdm{X-q;eW)Q=$bPd0cOVMqbDf^xJ^#5sabi{|amU}SV0Hd4)?gmSH1 z#s&t-FMXnN4fo2NHXK?O%~*%|d?NI{m)z=hQ8~;htXWq8~6h~D0Lu`6CXRjeO+|M3sX{a7#_FkX~xa>?br zYBKO+-Ep39Di;?OeuyDh1tsN_$~c+S&b#YZGqxQYcbb_ew;rJ?6FFbbU1q$h+*1_OBK9KkSE8xhk!V_|>Ksym0+x zqS5{tO$nP0tZL+IPh&TRzPY^S;k@;Hb(Yg5ir1%Cu1^(7?AF5%HF#0-mT2H$(stQ7 z?bT#SD;dbYfjcpz{sT%MTUy9>o`npj({;1O5XO<9EP?q+!^(7f!SBQ+= zhpb;8L~jaBvJQUr^1UrbKgKw#mkza$}VTovD`<&X-IU#qxdyccAEC$1Mg@5dv;-ux_ZnDF`CW$T$9cjLhX zM%L}9zX+Q=W%;D&D~kSknp$qWRNVJKa&T)Z8I(+e7`P*G{ZPcMC)yjG+){z|K!i&(hlq zg8mkWgrVJmZX?248JbDR6Hnz{cp`lcm)=yiX0njuo8Yp|D@&n`R)vwa;Q-c@|H`m4 zIsu<^t<{~y#h1@vgn6o%<`+UruwCZf-$F)Mog&wdf9G5kx|7jxJ+gtrM<1~^mFq-T zYyS2f@)A5QKQLmn%Hq@MTT{lm8V#AG7YucfsUuHzzFOB}`Sy5dE5@$eh-IFBey|pz5cPU;>9F_3T(QP!jccXh`ThmSJ^nsd=lc-1 zU*c*MaCKI_cA1y4%$1GeWN_!@`#{V;h$?>cVM+*YVT?htm4c!8ex{qq0X)zdsLN;~DO)mq35H!5R;B##fmZ zrn0pX1RgL}nk64`nIING$g{^X1)&ckDe7tglr-0$P_4_@j>G z%n>~C?uN-__P{f(1UbM>qwDts+!8|KR^Ubt`m=Pl@_IK|dPS+47+dVEXgS}a%NGv4 zyNjKdTD`A(`|qOlv~r_4&4C3^Dhl=x2(*n~B4z!J3tj@urogun(pOP~h{#*sSb52vD-X+R+)giLe-QnoAb-)S|Uahp1 zSpMKF=g>4ABculaGYG~;>Nm6u^S0O)ff*({Zlch}G6N%5^(33Jvgf)6vZvuS@iNn3 zTZGkXOLBaz&DFq~Aq-(r}EcVwQRRi)goi z&5Bh-%gR#TfSUgzULmfOsxadfRU^0joksXU1DJT~Bb<*|eQ=~8!;7DD&2q<*5PZF+ z1*n#=!iSPjwra&<8B?XXfOkq`r7yhu-e({Zn$_uz^PXEBoAC~rjeIl^k1{+MIdJRF zJ~_}Hi)y-#KjYo>Pv+&Kt+vcv?8{}9YQEeXbUA28 zH{TgLHZkJ=qUh(hBsTiiwGs&!{MRga*Klshr+;;Fd}k?Lv7R1PO07muZ@w^4gIRE5 zRiGe}!OvtiB?n4FT}>_aC0R#b5J=!#z0k@ed81#_$L%RE*cdb{S6u)lem=!GUO)k- zEy(1X&Yiv&w3x2?K?64FN}$nYJ8un=wVYn{tg>n?T=!k63Vm?1gKnD7)Pv^*)sMRw?&wgi7!6wlHP88T9)2_b z9=wL(HlRSeCr-?76KNfyl9YcWruKSwQ3J{)Kjv#$yNAz&Qc~WjN@enb-9%*rm5o1N zJP`u7QB_jMQ6OG6>i;2 zT?leq>KBR+cJ0|Zi}Fl*1CyZUrjA>;m!s|LwmbM*8+ygA++z z_z8H!5<$gyw?su~?NeVB4}CLv&W{X`1%r+dTh-vP#>?tWhCrBkEF*$c7>f8--!RnV zw}?;V_g5IFyM#dY3&gC_D$lCKgSc}^CyS%1{vz0J|HzKU*vTzjAa70!?=$fuME4%i z{umMQ$4i{f5lO3J#{9&NA=dQ>MdSwp$P1E9C~O6vPNIBAI9vc?-(L}HLpyFP%mO)z z)Jzj~NX=}t5PW^74m5y=53!GIRR$1lcIcT8krR!woNzReYKWQe^*%~pAJ?i7F?arN z%$dcFJu{F8*~Xh6E}KjvnQ zRZBOyX7(V+yYv>(IRDtU@Ts;{))Zt6BG7e|y=kM~v=w!-U}Oi$24o+@i5AZmxDyod zoA^ITdh8WFV+rpm5lI`>uoqxs-sk^Ic;<7=_&@HZQTfIN@vZ+XY?nRuTgKxW^x3N1 zD7i-oL5@_Bvc7`qno8M$NB@wBRyS{&%XUw&(?! z#FAillET9tLPD-;f;!W4FXUv0_s*0f4Dw&4QD)1JX9}|aqoVsFs*{F0d$4pdw#D=N zPo)16XaBg-Vn}j3G6>N@S#q?6_=&9vVtP`)b`sXI{g3#_YL^4vm2<>{k7Y7}_Yh`+ zp$+vOt|ogDP9W}0hVZ1)EaiQn4xNJjuQ43>U8d_qo{zNLB=$sT-09JIs{z@Q2+^q6 zt^l3~<1q@4ruqia`ioz2x^yVge}v`);wY zV3i&L&Fm0V(M@`NJqm-k78au?#gxeMh-3Lj6K^`P6C64rB!^J8FXo1l_d~fq z+8=TqBoXnWN>G)ag-*`&eTqISpKhqGk=v+5GwG->&X%l${R;V%&_-QSf}ONI--h%b z?kmV~2=F@n+swk>cTvaCpzDn~;&RJgn#%!gI38^_a{QyqiJkP4s2Vvi^gq0BZug9B zfhj5Y+5GPzR9&rSy3GldYt2$P_aAQdU9w&OP+KDO3Xzii-bS4?ny=kRba{bc05=iT z%toDt(JMdw}HmzXvt< zfOFi7gT8B|gd;$@h_42U5mS3NW=bY zP5wcSTbB&Fii9<4YBINSHDq?7sk-KF_NVE;|Giz3~%C{1n+8V#m4DPZ|8O zX=*av*lnNdN=4#AC)NAjzjfLS*zD4S-N~jE>2TYXgV#I$odK3Ogv`FT82hU4!oPW? z-BHX6rt5o-tney=Mn0GP2L0#WZA&zCi>(f7@eaIa--R>I3K)Nxd|m}MUuEX*=1<;w zc+y%*cVLQVPp?Di@Dq3c80fl>?(7Th9i3fOWq!dTUB?B?BC2lz_O9X`4&5pfxzQW2 zK5h|6aCKP3wQhdI`Si|im(Jlv?yZA&zu6AhRA)B?qX0W=v8&F*4Z7aGj{n2`fdwDF zj4%dfnFs+!d(UvL{Y?-ddcbwabME<~fTSYCZ5_}dweo1_xSww?pcr138yVjCV~QDB zRc*728-neH6HW1_i$8N+_5w=%U5yr^-CnT+#=fAb8K+xIc7Y9;do~Zp`X&0=2iDmb z>31;{k5N7|31i$vK_#yB7H5x#vhMsZxCM`L=Yx=dl}~EH0DBZ?MW_l*b$&tm;-d2q zWt7gyd&fHR5ZEB51eGIbQRZ(PT?Xs#8(4oGXFdk&YxF2_D`S8}>EAUZ8|}%_;91Xv zPZxgIM@f3?jV8HTD4%24V@E2P8}zXl0|MrYwQRK~w;B8Y0IipZ)A|r8K|ULsb!!QS z1>7Ca5L>xGp9ibRZ1f$O0!MhLD#=6j`z--x?S#VDXG1Je;y-ACWJItdvC@_QYMXNg z!izW6dj55HYLE;^J`C|JVwB_$nUtDUarpa#EptcVU=4i|x9wd*(4+y3WG1&`0R$aT z56r}r@z`^x`_{~_T8BgL$B$< zwLU^zo+6NFFQ&ja_+x-se)|`NZMOc$p|;eF0-ESE7w)IY$x*F!U*Z;i)g30i^;=zn zgGo&cp1Ky2@iO96khg*=hQ2mDj5gmgE#CaS^1<=8i?f+M_YFlyS&>nCW-$hBursn) znWXgI^tveHNrr=65jP2xz;HzP#3Xv8Yg4;AbaPJmVGo_c*V|}xF|+p9Qh7s)Z<}n??sMRN=AEvN4<`#v-pf6>(j@>MF{l|-j$f0R zpF(0UV}G&vK=0ubO2$ z@n_xb-}M1f_>b_n18?lL#tdrf4?HRFZnxq77qGqOS=XLV06YZf`KY<#GiL3XLijn6T8vz z7Oe=!Rshcl9Lc)sMVoiF-}H&q&_jVI#m!_~qPxK}buk8wj}f${{84*~ryj`flMBeg zgHz(uY7FEkym$W&Kde2J`JLbfRGL+zsq-PBh@2o&9RgAw-%9=s-GgTDr4T3ZqAuHD zW}E?8JTgOk`$s!H8%C%#JffKLU7q~RP(6bb%yCfo+-PyyWpJapa_M}*Uzm(X7YP_4 zN94EWncgFQZIu~Co{ioMglsL=>)n<^{j5c&UY5N${V=Y@H|iuTY+q=;w6RY-MJtlQ zoQAz<*_LXVxjNc$n!9Z3o?7eoem+}8GxiXpfs|c@$QL9Tp;nm_^G6^ATz~1FGVNIq5K%*E>6bhY|n;gt?(Qo4&i{P7O@eGYNN#tw;`L5Yj?BR$SG=S4eA zI6l8gWG;D3BO7kn_TAN8^yz}{a69^gTX)ulkY-jE-(k}*Id~0<1)*mqx;A4G3aV4ps6h_C2ot03!-`_}u%uER=nXt#0752|pq*ARzdQ7SBIOj~h|GpQKu zrMLt9Y4N+q?ZSLzFZh-drsB|+b@8_f-Dj)pXDd5lwg+;qdG}VWH)YV}kmEns_gN2X zG0l1hKC6EzDZERa+Plg+@4wDQSSot>sRuQ)k)A&W$|B9s0c#X8EONb#I@?3vm;c{H zUoh6YAKSRAXkG1{MSY&YVO(9{NMzZ|M?Y^~7CUol-pz>{fjdi<8_(QiJ$c38Ye|9;p@0N9P3KIwVu(@pZC+HnCoQxE<7S5vTH z`+e&T`>)^M{@o$<|32op_w$&i{*&FU-THR-skr!q@}K6eP`tV+`?;giv}KG9WuTJR zEL^|K_T{c9-jXGs8G(D3WuxBg+V<2cigUBh+P0fk5?jQkWY;g?yn0V~Vw;<>n#~jq z4yUzcZu|151$1f6 + + + +For all methods and hardware platforms, see the [official SGLang installation guide](../../../docs/get-started/install). LongCat-2.0 support is on SGLang `main`; use a nightly wheel or rolling nightly Docker image until the next tagged release includes it. The two paths below match the **Python / Docker** toggle in the command panel. + + + + + +```bash Command +pip install --upgrade pip +pip install uv + +# Choose the nightly wheel index for your CUDA runtime. +SGLANG_WHL_INDEX=https://docs.sglang.ai/whl/cu130 # B300 / CUDA 13 +# SGLANG_WHL_INDEX=https://docs.sglang.ai/whl/cu129 # CUDA 12.9 + +uv pip install --prerelease=allow --extra-index-url "${SGLANG_WHL_INDEX}" "sglang[all]" +``` + +Then run the **Python** output of the command panel below in that environment. + + + + + +```bash Command +# Choose the rolling nightly image for your hardware. +SGLANG_DOCKER_IMAGE=lmsysorg/sglang:dev-cu13 # B300 / CUDA 13 +# SGLANG_DOCKER_IMAGE=lmsysorg/sglang:dev # Other supported hardware + +docker pull "${SGLANG_DOCKER_IMAGE}" +``` + +For how to launch the image, see [Install -> Method 3: Using Docker](../../../docs/get-started/install#method-3-using-docker). Substitute the inner `sglang serve ...` with what the command generator below produces. + + + + + + + +Pick your hardware + recipe to generate the launch command. LongCat-2.0 currently exposes one model-card-aligned serving strategy: + +- **Balanced** - the validated B300 recipe and the 2-node H200/B200/H20 topology use TP/EP parallelism with LongCat sparse attention prefill. + +import { Deployment } from "/src/snippets/_deployment.jsx"; +import { config } from "/src/snippets/configs/meituan-longcat/longcat-2.0.jsx"; +import { benchmarks } from "/src/snippets/configs/meituan-longcat/longcat-2.0-benchmarks.jsx"; + + + + + The B300 single-node recipe was validated end-to-end with CUDA graph capture enabled. H200, B200, and H20 are shown as 2-node recipes because LongCat-2.0-FP8 needs 16 ranks for those GPU memory profiles. + + +## Playground + +The Playground is where you experiment with **SGLang features beyond the verified matrix**. The Deploy panel above only emits combinations the SGLang team has signed off on; the Playground lets you turn on additional knobs on top of whichever cell the Deploy panel is currently showing. + +import { Playground } from "/src/snippets/_playground.jsx"; + + + +## 1. Model Introduction + +[LongCat-2.0-FP8](https://huggingface.co/meituan-longcat/LongCat-2.0-FP8) is the FP8 checkpoint of Meituan LongCat-2.0, a large sparse Mixture-of-Experts language model with 1.6T total parameters and about 48B activated parameters per token. It combines LongCat Sparse Attention (LSA), expert parallel MoE layers, and an n-gram/token-table embedding path for serving long-context workloads efficiently. + + + + + + + + + + + + + + + + +
ModelArchitectureServing precision
LongCat-2.0-FP8Sparse MoE · LongCat Sparse Attention · n-gram embeddingFP8 weights, BF16 KV cache
+ +**Resources:** [LongCat-2.0-FP8](https://huggingface.co/meituan-longcat/LongCat-2.0-FP8). + +## 2. Configuration Tips + +- **Remote code.** Use `--trust-remote-code` for the Hugging Face checkpoint. +- **Topology.** The 8x B300 recipe uses TP=8 and EP=8. H200, B200, and H20 use a 2-node 16 GPU layout with TP=16 and EP=16; the command panel injects the multi-node rank flags for you. +- **LongCat sparse attention.** Keep `--nsa-prefill-backend fa3` with `--chunked-prefill-size 2048` for the model-card-aligned prefill path. +- **Memory.** The recipe uses `--kv-cache-dtype bfloat16` and starts at `--mem-fraction-static 0.92`. Tune memory only after the generated command launches cleanly on your cluster. +- **Weight loading.** `--model-loader-extra-config '{"enable_multithread_load":true,"num_threads":12}'` loads checkpoint shards in parallel and reduces startup time. +- **FP8 backend selection.** Do not pass `--fp8-gemm-runner-backend` manually. SGLang selects the correct backend for the LongCat FP8 scale layout. +- **Host, port, and ranks.** Use the command panel environment fields for `HOST_IP`, `PORT`, `NODE0_IP`, and `NODE_RANK` instead of hardcoding them in the recipe. + +## 3. Advanced Usage + +### 3.1 Test the deployment + + + +```bash Command +curl http://localhost:30000/v1/chat/completions \ + -H "Content-Type: application/json" \ + -d '{ + "model": "meituan-longcat/LongCat-2.0-FP8", + "messages": [ + {"role": "user", "content": "A shop has 17 apples and sells 8. Then it buys 6 more. How many apples are there? Answer with only the final number."} + ], + "max_tokens": 32, + "chat_template_kwargs": {"enable_thinking": false} + }' +``` + + + + + +```text Output +15 +``` + + + + + +```python Example +from openai import OpenAI + +client = OpenAI( + base_url="http://localhost:30000/v1", + api_key="EMPTY", +) + +response = client.chat.completions.create( + model="meituan-longcat/LongCat-2.0-FP8", + messages=[ + { + "role": "user", + "content": "Solve: A shop has 17 apples and sells 8, then buys 6 more. Answer with only the final number.", + } + ], + max_tokens=32, + extra_body={"chat_template_kwargs": {"enable_thinking": False}}, +) + +print(response.choices[0].message.content) +``` + + + + + +```text Output +15 +``` + + + +## 4. Validation + +The B300 recipe was validated with `meituan-longcat/LongCat-2.0-FP8` on 8x B300 using the command generated above. + + + + + + + + + + + + + + + + + + + + + +
EvaluationExamplesAccuracy
GSM8K20098.0%
GSM8K131495.8904109589041%
+ +CUDA graph was enabled, and decode CUDA graph capture completed successfully during serving validation. diff --git a/docs_new/cookbook/autoregressive/intro.mdx b/docs_new/cookbook/autoregressive/intro.mdx index 03737967d..783331a60 100644 --- a/docs_new/cookbook/autoregressive/intro.mdx +++ b/docs_new/cookbook/autoregressive/intro.mdx @@ -31,6 +31,12 @@ metatags: href="/cookbook/autoregressive/GLM/GLM-5.2" img="/cards/logos/glm.png" /> + " }, + NODE_RANK: { target: "command", label: "This node rank", default: "" }, + HF_TOKEN: { target: "command", label: "HF token (Docker)", default: "" }, + CURL_HOST: { target: "curl", label: "Server host", default: "localhost" }, + CURL_PORT: { target: "curl", label: "Server port", default: "30000" }, + }, + + curl: `curl http://{{CURL_HOST}}:{{CURL_PORT}}/v1/chat/completions \\ +-H 'Content-Type: application/json' \\ +-d '{ "model": "{{MODEL_NAME}}", "messages": [{"role":"user","content":"Hello"}] }'`, + + // Reproduce commands for the Benchmark card's "Reproduce" modal. + benchmarkCommands: { + speed: +`python3 -m sglang.bench_serving \\ + --backend sglang \\ + --host {{CURL_HOST}} --port {{CURL_PORT}} \\ + --model {{MODEL_NAME}} \\ + --dataset-name {{DATASET}} \\ + --random-input-len {{ISL}} --random-output-len {{OSL}} \\ + --random-range-ratio 1.0 \\ + --num-prompts {{NUM_PROMPTS}} --max-concurrency {{MAX_CONCURRENCY}} \\ + --warmup-requests 64 --flush-cache`, + accuracy: { + gsm8k_pct: +`# To install sgl-eval: pip install git+https://github.com/sgl-project/sgl-eval +sgl-eval run gsm8k \\ + --base-url http://{{CURL_HOST}}:{{CURL_PORT}}/v1 \\ + --num-threads 32`, + }, + numPromptsByConc: { 1: 8, 16: 64, 64: 128, 256: 512, 1024: 2048 }, + }, + + accuracyLabels: [ + ["gsm8k_pct", "GSM8K", "%"], + ], + + dockerImages: { + b300: "lmsysorg/sglang:dev-cu13", + b200: "lmsysorg/sglang:dev", + h200: "lmsysorg/sglang:dev", + h20: "lmsysorg/sglang:dev", + }, + + github: { + cookbookModel: "meituan-longcat/LongCat-2.0-FP8", + }, + + playgroundFeatures: { + attention: { + knobs: [ + { id: "tp", label: "TP", values: [ + null, + 8, + { value: 16, disable: { nodes: ["single"] }, + disableReason: "TP=16 requires 16 ranks - switch the Deploy panel's Nodes to Multi-Nodes first." }, + ]}, + ], + }, + + moe: { + backend: { + options: [ + { id: null, label: "Inherited" }, + { id: "deepep", label: "DeepEP", flags: ["--moe-a2a-backend deepep"] }, + ], + }, + ep: { label: "EP", values: [ + null, + 8, + { value: 16, disable: { nodes: ["single"] }, + disableReason: "EP=16 requires 16 ranks - switch the Deploy panel's Nodes to Multi-Nodes first." }, + ]}, + }, + + hicache: { + backends: [ + { id: null, label: "Auto" }, + { id: "file", label: "File" }, + { id: "mooncake", label: "Mooncake" }, + ], + writePolicies: [ + { id: "auto", label: "Auto" }, + { id: "write_through", label: "Write-through" }, + { id: "write_back", label: "Write-back" }, + ], + }, + }, + + cells: [ + { + match: { hw: "b300", variant: "default", quant: "fp8", strategy: "balanced", nodes: "single" }, + verified: true, + env: [], + flags: [ + "--trust-remote-code", + "--model-path {{MODEL_NAME}}", + "--tp 8", + "--ep 8", + "--max-running-requests 64", + "--mem-fraction-static 0.92", + "--chunked-prefill-size 2048", + "--nsa-prefill-backend fa3", + "--kv-cache-dtype bfloat16", + "--model-loader-extra-config '{\"enable_multithread_load\":true,\"num_threads\":12}'", + "--host {{HOST_IP}}", + "--port {{PORT}}", + ], + }, + + { + match: { hw: "b200", variant: "default", quant: "fp8", strategy: "balanced", nodes: "multi-2" }, + verified: false, + env: [], + flags: [ + "--trust-remote-code", + "--model-path {{MODEL_NAME}}", + "--tp 16", + "--ep 16", + "--max-running-requests 64", + "--mem-fraction-static 0.92", + "--chunked-prefill-size 2048", + "--nsa-prefill-backend fa3", + "--kv-cache-dtype bfloat16", + "--model-loader-extra-config '{\"enable_multithread_load\":true,\"num_threads\":12}'", + "--host {{HOST_IP}}", + "--port {{PORT}}", + ], + }, + { + match: { hw: "h200", variant: "default", quant: "fp8", strategy: "balanced", nodes: "multi-2" }, + verified: false, + env: [], + flags: [ + "--trust-remote-code", + "--model-path {{MODEL_NAME}}", + "--tp 16", + "--ep 16", + "--max-running-requests 64", + "--mem-fraction-static 0.92", + "--chunked-prefill-size 2048", + "--nsa-prefill-backend fa3", + "--kv-cache-dtype bfloat16", + "--model-loader-extra-config '{\"enable_multithread_load\":true,\"num_threads\":12}'", + "--host {{HOST_IP}}", + "--port {{PORT}}", + ], + }, + { + match: { hw: "h20", variant: "default", quant: "fp8", strategy: "balanced", nodes: "multi-2" }, + verified: false, + env: [], + flags: [ + "--trust-remote-code", + "--model-path {{MODEL_NAME}}", + "--tp 16", + "--ep 16", + "--max-running-requests 64", + "--mem-fraction-static 0.92", + "--chunked-prefill-size 2048", + "--nsa-prefill-backend fa3", + "--kv-cache-dtype bfloat16", + "--model-loader-extra-config '{\"enable_multithread_load\":true,\"num_threads\":12}'", + "--host {{HOST_IP}}", + "--port {{PORT}}", + ], + }, + ], +};