From 392bbebfe6a4e063e3e0b7c75feb4730994c88a7 Mon Sep 17 00:00:00 2001
From: Chris Alexiuk <calexiuk@nvidia.com>
Date: Tue, 28 Oct 2025 16:42:57 -0400
Subject: [PATCH 01/12] Nano V2 VLM Blog

Signed-off-by: Chris Alexiuk <calexiuk@nvidia.com>
---
 ...imodal-reasoning-agents-nvidia-nemotron.md | 136 ++++++++++++++++++
 .../figure1.png                               | Bin 0 -> 23328 bytes
 .../figure2.png                               | Bin 0 -> 21826 bytes
 3 files changed, 136 insertions(+)
 create mode 100644 _posts/2025-10-28-run-multimodal-reasoning-agents-nvidia-nemotron.md
 create mode 100644 assets/figures/2025-multimodal-nvidia-nemotron/figure1.png
 create mode 100644 assets/figures/2025-multimodal-nvidia-nemotron/figure2.png

diff --git a/_posts/2025-10-28-run-multimodal-reasoning-agents-nvidia-nemotron.md b/_posts/2025-10-28-run-multimodal-reasoning-agents-nvidia-nemotron.md
new file mode 100644
index 0000000..3718b34
--- /dev/null
+++ b/_posts/2025-10-28-run-multimodal-reasoning-agents-nvidia-nemotron.md
@@ -0,0 +1,136 @@
+---  
+layout: post
+title: "Run Multimodal Reasoning Agents with NVIDIA Nemotron on vLLM"  
+author: "NVIDIA Nemotron Team"   
+---
+
+
+# Run Multimodal Reasoning Agents with NVIDIA Nemotron on vLLM
+
+We are excited to release [NVIDIA Nemotron Nano 2 VL](https://huggingface.co/nvidia/Nemotron-Nano-12B-v2-VL-BF16), supported by vLLM. This open vision language model ([VLM](https://www.nvidia.com/en-us/glossary/vision-language-models/)) is built for video understanding and document intelligence.
+
+Nemotron Nano 2 VL uses a hybrid Transformer–Mamba design and delivers higher throughput while maintaining state-of-the-art multimodal reasoning accuracy. The model also features Efficient Video Sampling (EVS), a new technique that reduces redundant [tokens](https://blogs.nvidia.com/blog/ai-tokens-explained/) generation for video workloads , allowing processing of more videos with higher efficiency. 
+
+In this blog post, we’ll explore how Nemotron Nano 2 VL advances video understanding and document intelligence, showcase real-world use cases and benchmark results, and guide you through getting started with vLLM for inference to unlock high-efficiency multimodal AI at scale.
+
+## About Nemotron Nano 2 VL
+
+* Architecture:  
+  * [CRADIOH-V2](https://huggingface.co/nvidia/C-RADIOv2-H) based Vision Encoder  
+  * Efficient video sampling as token compression module  
+  * Hybrid Transformer-Mamba Architecture- [Nemotron Nano 2 LLM](https://huggingface.co/nvidia/NVIDIA-Nemotron-Nano-9B-v2) backbone with reasoning.   
+* Accuracy  
+  * Leading accuracy on OCRBench v2  
+  * 74 on average score (compared to 64.2 with current top VL model) on the following benchmarks: MMMU, MathVista, AI2D, OCRBench, OCRBench-v2, OCR-Reasoning, ChartQA, DocVQA, and Video-MME  
+* Model size: 12B  
+* Context length: 128k  
+* Model input: Multi-image documents, videos, text  
+* Model output: Text  
+* Get started:   
+  * Download model weights from Hugging Face \- [BF16](https://huggingface.co/nvidia/Nemotron-Nano-12B-v2-VL-BF16), [FP8](https://huggingface.co/nvidia/Nemotron-Nano-12B-v2-VL-FP8), [FP4-QAD](https://huggingface.co/nvidia/Nemotron-Nano-12B-v2-VL-FP4-QAD)  
+  * Run with vLLM for inference  
+  * [Technical report](https://www.overleaf.com/project/68d1d48c83696e11ba669f70) to build custom, optimized models with Nemotron techniques..
+
+## Run optimized inference with vLLM
+
+Nemotron Nano 2 VL, achieves accelerated [inference](https://www.nvidia.com/en-us/glossary/ai-inference/) and serves more requests on the same GPU with BF16, FP8 and FP4 precision support. Follow these instructions to get started: 
+
+`Create a fresh conda environment`
+
+````shell
+```bash
+conda create -n nemotron-vllm-env python=3.10 -y
+conda activate nemotron-vllm-env
+```
+````
+
+`Make sure to use the main branch of the vLLM. Run the command below to install vLLM`
+
+````shell
+```bash
+!VLLM_USE_PRECOMPILED=1 pip install git+https://github.com/vllm-project/vllm.git@main 
+````
+
+`We can then serve this model via an OpenAI-compatible API`
+
+````shell
+
+```bash
+vllm serve nvidia/Nemotron-Nano-12B-v2-VL-BF16 --trust-remote-code --dtype bfloat16 --video-pruning-rate 0
+```
+
+# FP8
+```bash
+vllm serve nvidia/Nemotron-Nano-VL-12B-V2-FP8 --trust-remote-code --quantization modelopt --video-pruning-rate 0
+```
+
+# FP4
+```bash
+vllm serve nvidia/Nemotron-Nano-VL-12B-V2-FP4-QAD --trust-remote-code --quantization modelopt_fp4 --video-pruning-rate 0
+```
+
+````
+
+`Once the server is up and running, you can prompt the model using the below code snippets`
+
+```python
+from openai import OpenAI
+client = OpenAI(base_url="http://127.0.0.1:8033/v1", api_key="null")
+# Simple chat completion
+resp = client.chat.completions.create(
+    model="nvidia/Nemotron-Nano-12B-v2-VL-BF16",
+    messages=[
+        {"role": "system", "content": "/no_think"},
+        {"role": "user", "content": [
+            {"type": "text", "text": "Give me 3 interesting facts about this image."}, 
+            {"type": "image_url", "image_url": {"url": "https://blogs.nvidia.com/wp-content/uploads/2025/08/gamescom-g-assist-nv-blog-1280x680-1.jpg"}
+            }
+            ]},
+    ],
+    temperature=0.0,
+    max_tokens=1024,
+)
+print(resp.choices[0].message.content)
+```
+
+For an easier setup with vLLM, refer to our getting started cookbook, available here.
+
+## Leading multimodal model for efficient video understanding and document intelligence 
+
+NVIDIA Nemotron Nano 2 VL brings both video understanding and document intelligence capabilities together in a single, highly efficient model. Built on the hybrid Transformer–Mamba architecture, it combines the reasoning strength of Transformer models with the compute efficiency of Mamba, achieving high throughput and low latency, allowing it to process multi-image inputs faster. 
+
+Trained on NVIDIA-curated, high-quality multimodal data, [Nemotron Nano 2 VL](https://huggingface.co/blog/nvidia/nemotron-vlm-dataset-v2) leads in video understanding and document intelligence benchmarks such as MMMU, MathVista, AI2D, OCRBench, OCRBench-v2, OCR-Reasoning, ChartQA, DocVQA, and Video-MME, delivering top-tier accuracy in multimodal [reasoning](https://www.nvidia.com/en-us/glossary/ai-reasoning/), character recognition, chart reasoning, and visual question answering. This makes it ideal for building multimodal applications that automate data extraction and comprehension across videos, documents, forms, and charts with enterprise-grade precision.
+
+
+<p align="center">
+<picture>
+<img src="/assets/figures/2025-multimodal-nvidia-nemotron/figure1.png" width="100%">
+</picture>
+<br>
+Figure 1: Nemotron Nano 2 VL provides leading accuracy on various video understanding and document intelligence benchmarks
+</p>
+
+Improving Efficiency with EVS  
+With EVS, the model achieves higher throughput and faster response times without sacrificing accuracy. EVS technique prunes redundant frames, preserving semantic richness while enabling longer video processing efficiently. As a result, enterprises can analyze hours of footage, from meetings and training sessions to customer calls, in minutes, gaining actionable insights faster and at lower cost.
+
+
+<p align="center">
+<picture>
+<img src="/assets/figures/2025-multimodal-nvidia-nemotron/figure2.png" width="100%">
+</picture>
+<br>
+Figure 2: Accuracy trend of the Nemotron Nano 2 VL model across various token-drop thresholds using efficient video sampling on Video-MME and LongVideo benchmarks  
+</p>
+
+## Get Started
+
+To summarize, Nemotron Nano 2 VL helps build scalable, cost-efficient agentic AI systems that truly understand documents and video. With open weights, training datasets, and recipes, developers gain full transparency and flexibility to fine-tune and deploy the model across any environment, from on-premise to cloud, for maximum security and privacy.  
+
+Ready to build enterprise-ready agents? 
+
+* Download Nemotron Nano 2 VL model weights from Hugging Face \- [BF16](https://huggingface.co/nvidia/Nemotron-Nano-12B-v2-VL-BF16), [FP8](https://huggingface.co/nvidia/Nemotron-Nano-12B-v2-VL-FP8), [FP4-QAD](https://huggingface.co/nvidia/Nemotron-Nano-12B-v2-VL-FP4-QAD)  
+* Run with vLLM for inference using this Jupyter Notebook
+
+[*Share your ideas*](http://nemotron.ideas.nvidia.com/?ncid=so-othe-692335) *and vote on what matters to help shape the future of Nemotron.* 
+
+*Stay up to date on [NVIDIA Nemotron](https://developer.nvidia.com/nemotron) by subscribing to NVIDIA news and following NVIDIA AI on [LinkedIn](https://www.linkedin.com/showcase/nvidia-ai/posts/?feedView=all), [X](https://x.com/NVIDIAAIDev), [YouTube](https://www.youtube.com/@NVIDIADeveloper)*, *and the [Nemotron channel](https://discord.com/channels/1019361803752456192/1407781691698708682) on [Discord](https://discord.com/invite/nvidiadeveloper).*
\ No newline at end of file
diff --git a/assets/figures/2025-multimodal-nvidia-nemotron/figure1.png b/assets/figures/2025-multimodal-nvidia-nemotron/figure1.png
new file mode 100644
index 0000000000000000000000000000000000000000..2e34c19753e112654cdb0a092d26e4aa23cef234
GIT binary patch
literal 23328
zcmcG!18}9y);1j5PA0a^iEZ1qZQFJ-(ZrtE=ESybJK6s{&pGcoud4p9zN&BUs$ILf
z@9NdP*0ru{b@#0#DJF)c4FsevDx{#Mz)1-A=f8syNG>pS2e=9-pCMz41SxSLQGNrC
zL;xJLx!qkJKK9X7$4hvKKLDV24*<Fk`$<6Y1{g=cmOK7<eqdO`PeUo3BUs`>{32Ly
znPWKOzX6Q+_4=fLy}yfJ0e>F9=GDpzZyEL^0h&Kuo_RlPo|{gPKlVQus{P;dZ2Zr^
z3IPrH&HT5%`TQ;Zou5um9gDwirtfRd0QLGSd<yQXkJDG%JNzS_+b^Y~J!)S)?=+vK
z0EV~Pu<hMWTQ7!d{VhQITk41L8DN#a+CKr{{5k)ueNAxczX71~xqX3tzdpF`#oPys
z0Ga?+pZS00wRlDIN%0(ikyp{pp;`JO{{%SXhv5_G(tk~V?zz-I;@|fV{k(jF|6G1c
zf6jX%*aQGRl3xH1*Ejs@0OIam{ulrw0R40Q!~DblRv!S^`n3Pj|1y8or@pQR)BzGd
zy>}2_)gR?wdbaqN{RVx>z7D^p-vMv-AN=P40ATwi@>Tum`h{S>XUYEz!2PFb`h?*w
z{<!B1(C(iGSpJg!(}dWr4*+!K5%xF$0Pi_pJ^rr%H6{yzQ2|LKXLwmCY$#}2Im7Y-
zP$FQm$O$%9LK8yLdakgN5V#PqGz!LrnZSgAq`@PcEF@fD_M%Sv)*-78nD6Kt5)i3n
z9$?}l)W9PZu`aGEf~#_x<Ug>6(1i%Ki<<FjnoPs$$r;2Wd_kNdJotn_q){`h{*zV6
zcka&Xc@1s^i0qe34~y|2g`-<C%@nr1-w9-5z~#7wHae9r<E7DZv^TjDymDu&1rJja
zh6FTfCC$Og!wzvy;2t_Z)vTNQt?temoYt{@Tn0;~tRwm&z0k}8ddgXtHxH6zoe>=Q
zI>`ua-|!*t=JS87qrYoo8dX(E-(Q|}hiLXKq>TzPni~6ui6Nl)i9fe6YFbiZ2W%l&
zovx%wEt+NO_Kl8teSuj;koDD^_qh2#s)EP(DxtV7D!>lU<;2|aWTWbb+C{qPq3u^7
zDvGU*>#BaT;q~WWO+wn)cYph(=7AYFT!)V|7)^)j(64G=iJ31{+dIh8LX+ad-@*S@
z5p>d&+#4q*wTqLGZSdX~a%SehDuX#s?sZp{E?G&9NUpR`D8vwa9<XEC6E|^rl9}Xt
z;IR~fS@sZWGu7X~B@7v$a{Ty#p}6{O(c2HQY$;!sK-daNaL79X-oH4!M^D1Qes@wK
z>Cl?by#JbF)t#2ET@F6q9ny<xFY^POTrHNWA%ln>hl}NbDg0ZjD2Xpfa6?otr4r?s
z{k6lW{Mf%iLGa3ggdrjkYJxyYhtN(36)X{^KY%}GP>p$~c!4>26AOvo#frl{nK#n~
zNi<Oax~CRyY3ffl&fJj7@2PjrcE}!_k|xI-d5A2;DJJ<l52FvubvOZU5)W+BQGCVx
z9nIZ!B0fLmr3cv?vOe(;#WCcEN6H%4(8tt%-vF%87*eaPJLvb{;iY4=G}+x<(37d<
zJpv@Gm?y`bzp$_33(j++2(_YSO=&Il>_=KnC=UJwTV^B_jhq2xA&CEh!T)3A|Gyui
z%<Xs5J%;lCVJ7~WkGd3ZYi+-u#9!$vtD=LJjVu>Ssl{AGW?&&RO-gQ_gvv4oop}&4
zaVKQvNyNgFkoiO6e|F$ix3z=5g(b@IUYlv?3(i_IZHu`w%~6e?q8++KIUa)6{a2Cz
zu3_2{F(bvshJO7wPY9ZQwGO-1dOm6h+y8HFhWx-3HudT|h~p>y7pH`_QfR~F7@4&p
znP!f*hxH`o9sF;!L&MGSemTfR^)J7;<h~<96`ac9or*57mHuyWCfn;`(?I)o^brB-
zNqSE7U2^s4#i^(hDgTwz{*G*AjzsLve>n3G7v|p=)~Kk#HE>}nEZ-@A{C6Jw)(HL~
z@1z)vfhuk|;q`B9S#5#KAg20%T3;XlS7!ZpN`=oHTOwcVgvlU4Y5Et*22Vw`DQYlJ
z9Ow$`kICQtccg}Q-wl`|FRY*4{pV>Te5<&qYl|`0dIt@#I{GK8YD+28vbWzc{+zy|
zH-E-td5_NgNZ!|gs7w0te&fJoSw<2nJ5xJ}u#L>y(E0Z>pF<O0dprE=*0GifQiOfJ
zbh@4HAHkXO!0T=Uj`QZU$!i2}|Adh&ne_$I1`2Kd6o?sITiynd4>>txO>pWTwCqm)
zpa88+_+z<-uF^7bWH>+T`d`!d+r*K-qrId07*6z2j$G>qq69`m+h$Yt_fRSh*Ry4s
zpUQuYa_Xx5*J4KAByG5ykWfUJPEb?~hC*gnobmr1w6zxYRQL7FAhdsa3@OPb5IO0x
z?H{?EbZejlA-H)=e7;mEs#8{l{GU?XCD#PVAqqcbcPm9Ve<|MtZyQFi<{S!3wc<)Q
z5|oIaP}Q@Tnv0-v+>8?Gc`Z5-ObA3ghS~=-bNX-T;yFnQ3J@Rf`D%KC1X`SATrg#*
z`SA7)fi1zCBSIZBa)vvQO65sl0?*FRc&|l2t?2y;iRh6_@ylXZmplJ&1x5T$nxBt+
zR|}~24eVuemSMw|KJUTbwh~_KBxrO{#qmqN6Bxv>^Gr%=mwVHA1XPYOaG^a!n6MSO
zCMNxjGcexM7>paOC;tD4J#+f9A8+^U7L8y~LWauN)^0-XS{01ID0k<9uGgYPWh#0Q
z;njUSR^1_F(q9CKv>FnCfGj+h!#x+>h_e6}+_5T6DbSC=?gAW6`jNo&6*rJ^t97WD
z>K(lG$BrVNY<De{q%MVz%XM+Q!C$M$WI*D*!MVRA#%N<^Iq$&}3Pxf382l1&pMwiI
z(bGA#U^KkKhJMzgg%Jq<efL7Q({xezJ=<4L>Cd8GPPe7;`~E63nDq#~ci4{CMChDM
z%$NPRQM<qYJpf{^ezI?$rs8l5|H+W4hsE=OBr4`d=7+12^QN7RL+hezu^S3D(58N2
zx3o5_B(a?Y63M{)m03j|%}5eTA-v~QPCa9zOGOJbQVXP`Gk@zJp#Fh7;+gTL8?ngo
zCtD@R@Gb{Ir<xw#)$6mAl)^1@+fRTXcWx~mS|~>F(6sHjiT?01eg%mv{F?jHxE?(L
zc?cEoSIOOv-{5wP+S&3z%Ku~-MBlR33JmyvWF|~4If<}Vu`Zb5IE$v>u@qjN7H2rw
zpOvyCQWG^v{Sc`9AQP|auS{_F9Es6zAdN5{?`>h4`n%pD_^KuSD_proG}6=*nFQ6W
zAAcd~Z<0rKDy$R1+ZtRvYaLraT{RH5fqF^JiByEoOYnO9Rg{!8xYiP=40@;!_y;0V
zo!_c74;%ibVUpUKtUp)?A4McWl(8kj7gx5g99L48+{XJpO5`8%@y7!4-|@}(;ZnD^
zg0drtv;ZA3vsC_9@#p_jQ#TD`QcUTbf0p2nt%_N7Y8tRVsfEuZqM-W^(j#oJAGn<8
zgP|{|Jt%zvNc`&f9vJYKV92Cr53+HjJZaE425qc)fMhcKoexmnm!d-obv8e%4Q<+$
z*80B$D(;L}*(y|L#Bht)Qt(|0)<O8l{;CrBKP^l5RLT!vdiqto;i%gn-E%yK!a*d$
zc&cEXaW4iG8H@zjh}E3PMA*v>OO3+ajvOVZlI}giy8Hg2h?U8C;p-8O3`SOuIwoh(
z7j-1fU_Me)B4uIIqr6_LGJ2Mmp7N%Fcy%QI#dRtVs02x%<*^6gPX1~73`gjND`JgG
z@qH6>=ho<l+Z$wjhhj2TN7fO~BS<%{s<LAsU|y)u-^^yCi?;8~V!2N((3(nSqm5jM
zMOvo?S$N_>$L2S_7m<)3_?z($a@*;p*Ek28!O=LiM$+COvhPpyAwB_!Og7>#MqVO^
z%4CJTgc&o$mKF@WG=;P4C^&jjfze@UA%~72d9-h+fP4GVoo{DoJY2I@@8^i(X+Vx6
zsP_9BSCKZk3Cbw^Ng-x*TE8zq<|PMrbA#B^gD#hSc6wO@sF0*bQb>tT;;W$By=VT$
zhWJN*LQWs=QA&xB0~{I}2%j48KXfK_&VA$$`8+`r<4iYHFdFq+TySr<-|z2ssQJ3@
z*9+OH6x`DM8A1?XCD^X(vS^?L+L``R+6}7bTg<N6vir1$D?P<yG?0QXf%K~U6K%|+
zR#pw-HLEQl61ONRZ~v2c{4eqPOEwk1yFuwkSlLrus9C!bEd9x!lE0;`*@`)SdhkzN
zTlGNWiGL)*d20O4H?0KAb9dT*Nb@YIt?KXnN(4y(dKa2b1TjLfW(V(%9NC2N=$Y-K
zr<Bww_kO7_tv6UAy!q-P7Ly^X8{YM%=k3jAC11zwN&&|oEdN~&nmN*hAtEhoDX&y5
zo$*(Rg*D0Ga;UBv+ANq7<V7kC$UAgG;*0VUW8KyL%~lTlwZ#X2CYZiR=B)X5Nr@yb
zMB_!-h9rU~)!^pcn<JeulK8h`|I?F?(N6llxtZQSJQLqKlkO8;&s-zoj=47v64`6y
z{0QvK{wk#io_>BSUY1n@C*&UUkNTS#l})q%^b!qr?hn&)39iq<CwIl#oW)?(WpN`s
zUv<8`*e&MizFa6|r&9VzEzG#B=v$hEu&)cDW>9p_rHsk}2mK{iy~O{8d&_^VuF)4%
z^QM8h^~V2Is_dC{NGXjYepk2*{Hrd(JNnpnXm*C`Lhw+1=-+H*8XX8>Lrx@8HDYMh
z{emkVImHga;7L=6;gvGz8MPZgecF$lU9YTs-`C3j8HRb$2DaBJA*Yu?!5uf$B+;-G
z-pE1wtnZ|-c`&uK=bTG@Ase%W|C^CXWDzsxlHxn_3u?7ggQLR}a$>f*l`nE~jjL;B
zTq(1HG$@|Td>cz7LW{^1XYc-#t=dT}G<Q0Jlr*Y}Q@p0Mn$fomZ~_9<i?r>SJjsQ+
zca@Pu$ut7Er*yXa>;}wK18xRe)sRS@T|8<|b-IvoppUXIYk1kVgS#w*e)c-unr2HE
z>yohF#Wh@20W4O8{i2=RHe)E}HG;_=X<C%_3qy1G<>h&(^xgWDGN92kat8FGQ~f>G
ziK<MO5{`DC*2V8xsW3PBtSU)cZtUT8ffZ4lrv4IOQ@Dk~Kq*gN*QzYf?!n4Jo0k+=
z?av~)!%?=QcY$p-`sO6f@-$;5r;`;$32v)TF9+*`_8J$?HEl#gW|pAH9pnyOTcr75
ztD^D>AWf%EIwFhSaDQ!I6mBFGGlEV#Y%@G|#)Jf~^h8rKJUAh5LKx<nSYFH8tD#0M
zB#jWjL&h6nJwfXC)QKj)5!(=w5mdtN6-pMf_f61tNg>1(ze7VNKQRv?`7;`>-}8o=
zgbqmrJ$KdVF)oN4LK%+MR7ha?LPXd)&NIA)@Gfy_?u?CqhZG}9)@vJ#B&KMHOX`LC
zOn*+3{f|<0WJminh&d4aQ`jT3rA{zX=%D}q_hiA1o;1ZnesUFa4<w?mq&|BS!kc`M
zLW(Ltn&NH91rokpzdZlO0~)>g#ysr;sYSRKNeyB$<R#6KD!he8<xX`;-oz9k;H8ol
z`fM(?5>*_E`t{tVzn*b}xM41OO$vd}rZEm3VWDCS1$eS4mSgEOR#0lGPpFT~NvU$y
zNhxhS?hl^+%c;bFDi{BK7P5lor1THL`@h{mX1)OdeSIbX{q>2P4S$>R$X#&q{QBLp
z;)+JmoG5BjpsAArDCy0<j3L#srr@yKl#P+KPtlVw+{>PS+y{zwwKY@bTy@g&BqwVy
zs~-F9R}b>|;xiC95`all^n1hF&?)1WGV_;F1Gbl*<HB`FL{e~wZn1K?gYYz=PSCC)
zMhR*=Dag>ZCoo9FLg~1G$LV!x`tvy^Aiux59{EeT`DCVR_g1`;Z`UE3?nk3v&AsZ@
z16gju&&AwnJvHl7Ek{S|*(1=S>YvtgXKQDn4&N!=SjKnb)$Ej!Fv)359`|Jhe=sVg
zy46F-52c+s%pA9gJvWl+ymNw>-Dx$Xb#Wu62TL>ak$8k#@Unf-iq*q7!X;$r7;$!U
z{-#+l^TyW>ikCZ@QD-=f?xS0cCNzxDu%eTXBJ(A@4WEGtbyw`TQe=P?#<;Y|=Y-yx
zs`R9CTa~U5p;2FzPKy4}>9Z|P&Ak7r2<CaBSL9Kni9cJhFjHxKLkvQ~beEipa)`U<
zhnI+J;-U4npBRIW$|U?sM`xq*GW;HRku8n8TbmgBBr%NJpxJaFZ;b)lUiR=QD7-%V
zt0T$#LEPJjcO00ji$D?>y{E@I>&AT~MTm=lVX&-Q6i=1vIwNz9y!0hLx;!nEG}^fM
z7)iA{2Zq$U+*nsgTOBdENUdVd--nHq8iGI|#`)C%Se`KY5nZgHe$w$n9MN)3;R1)M
za3R3%4O!E2S7EtaA2_m`2+9=6bVlCl*D%B9-K3wenUZi-2C4eE)lOAd>F!t#ueEC_
zx*IhcNvNUgV&Db_kj}^efl&eLN3{U_Gv@Cp*xP_FiNZi6_dn{;H2L$vsZ<FK29h5|
z8yta26pJd9<<0EJkJQFs+3i)mi&OB(36f;_R<!}IP@H<lQ72VpjApx$4lb;jQXXw)
zIyMmvwq&MMW7(Ug^h$n@Z<G+cIi$4Fe)emRMJgrF3e!N1J(!DAKKS7H=Pr(;afs!t
zFbI00Dh{zuDE81*2<~%koE(;AiibtUz_%l`o)aECBis$H+m0q7CTeT0V-R0*r+g-x
z=;CKAdTLeA-2*n$sD2nYRz?ZMtN0ByEsrH)upLNVJkd&!CPW(RCG!HKP00~~<dK-_
zv|GJi5ZpbxT_LP7E4-8jRyzD|WcTT9hdXG2^@gjR)N`9A2sIV6-;ZLHLKrPwMM2lo
zE{GMhgcdGX(S5>eyi>aNMpd0@&&XiXr-PU-onr0cd4|f2)1TGM(M9TpY|VWW!Spe(
zRl?zpJ3MAHP@kbS^T>g$-rQ97<XLwcR7tvHSxaeQLHiSC%NPh^5T-7Qb<E7@I+71i
z;WwCIVCbUc(`)gGuCWs^R}huQ@w@R&g-*H6+*?8~-D7@NJL4<_ND@rW3ro0=N^>~y
zC4?{3%M0i9wO7_?ncK^qe+nrTv52IPOSlg8w?p54530*%hoBQjfCJTRKZp^l8ye(^
zu+C{}CI6isc-_bhc{K97kZzpDgH4}RfP?s4+Vd5^!!(VRp!H<2-wU~>Vh1dLeZUS3
zty?F6$61{U%AKu->#|&8@mSQO5ID1aTs9a;1!?{cTV+DG8`@%;$k;y;`*d75r%`yM
z=DT>msUKF+>1Lv?0wnq`z2PD9a3I8r+zK^wRYz>IpPb4Mk{GEXNBs1t991q+tJF6;
zdFeh-dSxIdAgj`%AW;yj;N%B3#hI)}Jmw$sNn=$h5cu0*mMfueX`UsJ-*Dt~;79~L
zwG|Rgvs(!y#6>xPU<F!MM(KO+&Tlxm{ZNP<G01*!jQ-MWqD$uZ5n|mgg{pY%BF%?u
zcxMTwwjw2SoAZNRt4Dgg0DtW1rfI!19-Sy6jf?$LGca)b7R2W3H2OZxE+0*J6}c$i
z-FJ4;tR@TnPE?tzPP6*8>ntGHKJD4IFeI(!e(YY%s>y?0u$Qn44A0Ia9T#$NomK2S
zh7c)#9a@!>iYu2mL&Zagc9^SoB_@iLPk!niEvRCceVVh23FVinRL-WakX`h*{S4%3
zpX1<%J)!b}^&pR|PvM1EC`3oHBqz_OcKY|T!=U`}?84UaH_vPYSWB1&ju5WhZ87<y
zTep=Z<ApmBnyoK7Z*qJC1G6~l2tI+CK^)VD;%+5Vz#9z+=+9TVnVXD&;00d!gG=1u
z-!VNaggpV-UDAQVtfF4CeI@GN)zCCS(zsYWhOX@ex|&<!!>}=~Xgdq~`BHQ@J~w;F
zC&6aZaU}KG2L||G!8xxgq)ZLpoOxZJ8JK8ix~3n5UQe)7O3Yb41@{Fg14{3y7f08E
zw$r>?pD!H_gH9OBn7;j5^r{H0*tf43oReOeLE6a$`WWY|$UGNf$~xOuA9|;v%w$rL
z35#<`)ONUvRuxh!HA-Rlc_RIp1E2Eb4SaUDCpuR58Y5d|#>W(0Bj^~(YBe+TD~Q|0
zmopuh3hIflnbE`$0G4dH-r)8lO-!W?XhJa^M;_2#M#o|)7hv6(YdCJl`3Mu>S-2m6
zii+96k-8`3Bi=B(v|VbbN(RP+2$=y_fj+4tp7Np1il1^4jyi6Nu)fK7Ob=IC5?Ldk
zz)4H>>;xm03^0tCA}`x-(yTrf?a%C9HJO^y0w{!XZ%1dT&ZK45`aqwZ-ekgu>DKMd
zM(<OKks&ZK+0aUhHglTL%S~PnqGjpLXm1uxNBO5}_d#uZE#c+8zba=qO0qoM_MEEr
z&$@^dKx&SUCg@)WnvuD;GZVG*$PmZw<YPk#yM&9!Y6C_eB_~oI+?#Xg^$^NLrC9AA
z-YaeG3?2y{$nQuyL|}+J2s{f++RUY+Suhpx2h2}mBW$veeOYLYL4=h-?{K8q3HOU%
zHfXj;w96}v#8SHt$1d%UHDI&|H`-sQ;S)fL_eq)hHs;)jM$?#5(h*v)HZ*z8ONhpW
zS`R6`k(?O1T?iU=lguQ}oP(ZB)lEI@a#?TFQ_x`F;W&)GbkEq*#vrXZN(|S}(l@y;
zoM(8U7yUJ0qCI!2bl>`Ftm8nEQ?G^GufH%S&ak9t%^(+wM&`EinmH;tU=B8tM3J5)
z^KSHit#S@0xt$H-+T4e@yd-BJ^F1`-$NogsP^PoC;X&HC2ke|ZbAcj5p*B6=k`!sd
zL1J8t45FqzEL&gpf1gQ*EA#kzAfm2iFl_qWUr0SQxRU&%S&NB<)DutUq>06%H4Ro1
zWP{QiW=7sUrd)_lo5(l{IH1G!fG{Y)qc^gEG7~r66fFOM876SZL*H|b&DbLo1r;bE
zKBM*(>=_|73$$X3e{zk4bX25*m_RTo$z^#$(U4Tx5?u&u@7Ijkf#iC2S`oNw(4&jj
z3>Cj*?Ete84<ZRl<dMs9F1uHzxr5Tq5k!ew1Bi{7y8MpVc&1Yd**b)iH(VvZ?0}Ks
z?@TA;a={poAe_JxG|qg;v`dn(MDO3Fx8|lA>ZMIXsym6aWf6DuU{Fw6AKNg@)%`FG
z8yl3h26s9=T1qjMTW!MMUMmJ-iA(2yAorr6nxPAca@7(`HwqNe)IVooNT<0bOL<5Q
zCayh$@uDcv+4@5M^2<8@VLz{^F!Hlat)hYGs}sJ+Jukzs8{g~Nh0BLie|5cQI~2Wy
zS9rT29u6^(9c^jAUiY-fqR-Fxmm|SCbTiKlNFu+AO#+AnQr42Ow!SB$zPQ~xAn5?V
zpioG#=72_2q;UpzZ5<rf668K~GJSLfdr8iy4NthFq<syrEcdi%gu`=W44W)6-nzKo
zWplAheujAs-K~Uyl5M)9egOSyL^wFkhhW#%u;*lJUDdGyGsWmB8bhM0vwQ11sl5~?
zj)d{@Vo;B!*sjz_K6)0n-lQ~gO1?>=69$G7&D|2+b=Cq9O`(4tTR^gJUaHpamPJaz
zx@=xeG{*)=-!Zv=lhfsF>ycU?Y~$UC+Di2sZF~jd=%|HTbkk1^D|>Wx5fYA-lhr8q
zPLnw`9Xv;>p#odV9j|sDZcaU$i!oFBH1%wCszCcpSyeL@lT8%q&=FrBs^lV5nxJP3
zj_<ld%?a;1$KC65pOUF)ZDYAVI7ZfkHssg&B+Nsgbol^C!FDg;<E!cUBM3AS85Cn_
zJaZ}*1veGTH|ZD%*%M3QIda2v6va@r*Hu}5b%nWgZBE&*``(wZPrsTK+aL8ZU0}Lj
z?oCc0d?M80F%uPn6aPvgj>}MW?_vaTyWftyzJKlUbMeo6Fm3PYtdYAYWe~+xI?*1y
zNr6o^XTkSDcgIte>F{cor{kep>uljjT1w8PbxI#Aa8fZox7CK@-xb0kWe_vs8Vc%c
zcBY^|CJ%T#1crc2L`+bCkg8ZTVUd~vP~moyRYlDN{&xGqId@ln_IB}67dY&5#5|7f
zIsf6`1Vnn0(Y4k=df%|nDTYm^Gjs?G`*EA=bHA53okEi8&>DVqP|vYesQk%WkV79I
zJJTUvW7AaXK=R<QvNHRb{PLS8JOE~@sdyzl`r?e}wZQ6)<)P3w?>gM0{wV!8vq46b
z^3uFPkDRTJ!3%$mQwDex!OhVPK1WZHa)BO<`8*!iJ-^4M17;;!>;4Ujm3CW3BwfGF
z2mBVLEOpy((ru4&-}kDQ_{Fk&JZ7z(X)#y&<<I~%^=+xVgpfyISmX{PMg0fQg6oYd
z5=yIOt73H`)0<<))aQ2@9y;|+r^jJZ<elq@d+=JhKEe%1t5t1Gfx%3n$hBVLTwONJ
z7PywLC0jzhVsU0i)FgvP^2<l~{n61v%hC<obeTR-o=mJa_GPOYhmwLX18v^0+YcNe
zZQxLric!awX4NCsWu@Jf@^BC&*|DelN;c#-R-FN6bRHtpbA=~P=pm8|s=arIsOYIF
zOx9gyDE(us4dRCYN_Js~ujkY#EDZRWcmyK+2zKcFZ_O&K1I~ujvZ(Fp4X}&_f$Sxs
zg@=!el3;^Lj|LUabp6q0Hu1Eu(na)==(!7u{SVe+P-VR}Tv}>^Q7IK|09VH%X4a=6
zKjePNoB{JYYB16FNWDoDMvI>TJNojwMHRVy19;Q{y#|%z1YuENYVHlTEC>Eji$HS=
zK4edVI*>F%GkqE**_9exiqWxqztKeH`CM{yimcm0X*Fd`ydT^l926|;RyNVKYc@ZR
zc^5xxU%DyZ7hq+K-~*d`Zj!rpoo-auh=VmHZgF30yQR0ien1Ao5bk_j-OYPw%L)+M
zc;f&kBt@=aYdh*vtcL){sON46?PcemgmI4w$_@mM8q%_At@A4S(t#FEgbP^$b@Elg
zdI<eAQ<;>*jJwt3st9TS1%5gNB07%6!&WyQt8F#as0#XR-1W&DkeU&GVZm7rCgN(j
zufkLJMhWvtvdZ&ac{`J}5fR7xSZ#%ffc}~y#_fKB2oLGJ1hG~v-s=I!sUyECFC!;q
zEBm5szFKo2Pjz@EX>!^%8eYK268cae%qAI=3&Trk2?MsOi99hRnFsOi)Y6cF7&{6q
zemNZE+w6lu^U4)5ya(1H+D-NZ!+vz*SV58iv}{=$GHt|5d=&b*<tuU2pCfp|=O@@Y
zNS`=FBsj|QWobpBhV5FALY%8&sw@Oa*IfVwI9GlUz*oxTmhfd5Bv@pdAyH6ljR&~;
z0H587sX)HvJt}brL)fO<EAmSm?2U!X?i`xcJ|BgAw<>U)OvhaW4m5*$!)gqC80Vs@
zpVo?GlycYw)&*<{SI&7_T-3)59Ox{m8(I8^E_d__LViF^kpQnN-A}G_I{wYGb(u+Q
z6w>jH_hwWZ&mF;6o3n4X=Om+6elvHe6n(_N?EI4`@X-j9yJ~6Y#5-H<T)J9)8h#8D
z9vA9ayX@ZIBDGy{+VuY*EeBe`S^#bOw;+_~R0`~t7(ul0%!Fv{CUsBD<hWU348hF#
zb^ofBi3o=d4*#=w(8cg`B_4?;ef}t`NW<-zY-$}3ep@vspE`>I1re&LyCMaTcF$8y
z{4&Q5f%(^_j3D-r3uNDB32lxwFwF3f<9J1bp`lc4JXl#2+>)B$oDuyzQ>b2tx*+Db
zheKex=NN{km&HK9x8SINHwc5Tiq%>CQ)vC3IF~1CiN>Ft)!fRf_Wa;Dn-oR|&-{w#
z-wX4b&lTeiC3c-ODQY9Wm*e{mABr1@1s&wEi)K3it{pXn@2B31&@Svzw6~TxEzSeA
zTEDjP(ftf=|H)XRRR_6huZ2;)K%F-95OR&47DP-X&uAwZcyH!{NN6#Ga;7QI{1)4L
zCzgUmwxn}v#}v6;FXsJDi-_0NP?LyXMS7FElISnICo@%neJ}P%9bVQ;k9^K+DyLwn
z>yuZ&)gwngWH>#k3+JK8-0e}l-)?&kb;h=A@>%0Tr>#(Vx;xkt&G4Z;n{`Y;DWk#F
za58*e6eM^5tdeoA%^!w<{cU5&6TUw@`6m3=MI>q&);OuW5td^?S4PU$U34B$v-2kr
z$n?fgOPCsb<kLDoJBFe)>B0S_i|+H{Z*aj>4pB$thkJHhj5%~OyJL=YN4qh2o9@<<
zZ%lO-&nZu0O2z$PKjNNqI#zvqRieZ_VjK%q>J7Qa=6p)h=B8>zSXU0O2iEvwbQzu?
zyGN(2n5dg?%087~s0(UozfELRwdW=&-+1^LYKp8OfJPOIotH+AHKHwOXVf056VAg=
z2F8uz2`nd34cBJb9tCG$;N2(YXg5tr^ra68wlqIY@Mya__1GQI6BFyZ1yDXuY5EP1
zDKrayPZ9T8`i^p<%#6{d$drWLXhDX8BjR#wS&)3#Bq8!*<$~b^k1Uaar!O-4EFDjK
zV1;qilMe@nO?(&nMDH6w1yKn4a$Uh?J|Gh)tO=&_D<FWEN*)kACD+GbB;P<#=u)}w
zKo#(jv9kP956q0Rj983%F#CD9IO&3m0XGd@lomuZP~d#{>khIh(R%O89aV|!nh8op
zl_jk#i5eSUZv<|jG(qkzb7*4|Oq&=S;C#AXNz~IQirHYDJINfddLF`nUZX}js>f`B
z%_gS2dI9<8E?WKu(osN~jv_dA<D|cY{y5b6$ZoSVt(fv<U%ijuEiChs@G2e52uG`_
zeLkuK9VZcotkG*VL!*{=1gc1&{r86K?J%KgTPRRX<yDAYMcSm@CJ$<a_*LhUYGXd@
zaPd<XVkWa*PD~cK?RJo0V1V5=?sqBHbxG@LM^Aj>p!(AkHoggcX7U8eDFai!q3aBK
z>UR_;K~JKDgjM<h5*p~s9e7Vdqv8i0OUGHny*F_;;7$S`J$G$-AcxP7WpA=&_4^8(
zga{6+HB|~dDbP?Df6<ClYm3mlr?(qm<H0+a9#up_R{LLwkU14+?`T~Z$A=mY6%2`S
z+tUuNhY&C!7A9S<fCRbm$>cW^?u=sdV-V?qRALO98Rq(J>&>TrnEbO4zN~85YBVjs
zp&)8i)6&j_<aCu{C}TkBDb^FyB03dM;Y!^Zh=F`le1L@9j(U(WL+)98<})Qc@VY$i
z+!}Ay8G*5GMl*Pcq;Ly$4bpj-4TCd7#h%qFu4O@;d}35AQ0hk`I1dlV2`_N<$C`YA
z#`_xJuxV~K;PML@cPxuDxua}SeO8l`(jpbedbrXuS<7Nfxf!@@=DTGOMNr1?U~e*k
zMem(8P*l=T6X36OkEKMg$@@7Yb;QAAP-A}RC#{~$0^cI1!b5&vb*%|n{}$pHSGL2L
z)4O$rx$$~N22q+i4g;zCm?`D58^Ug{mx!<T2Bh#5KNsP`*&wvSAQ}iywWCe>cVoMD
z*|tM&$?Xo;oLIi)v>!JpM9mY&;uJdX2TBWu48QO?W{l8rAsdb1o2WrpfCS_K8p)r>
z?S|YAzk3dWulYro9b7|&?l2;hXD9ivvu_oBSu=-w;E%zDg20oZh{>gn;5`OD@W^f=
zKzfx^mU#BLq!17>8S}wX-jc|@WBlPk;+&#xdGI-%OOaNN1!0FLWDmJk$o;|ztjFJT
z`X+oUy&cwu@;pc`ho@A$Ni{1<KB<4-4RKBkhP-WAM-toxJm(dDQ-ZQ@<&g;xg-<Iy
zTGJ`}9d#JOMZ(C=M|RyS_0%di#h-=uHMKn;Xqzcf+#W9DDyo__niWb99EqbvBEmg_
zyp+?BN4O-EZbc7*C6jx}r~II>Q_hSH{p51WgXF%Xs$2tzOEE2MYgnLW?o!QWnI2KG
z^IQb3o+ifb-5nqC5{+AB8Pz#UXS8{xfCEK6m)Uko$))E*wlD|pK%qsNt1FJ*x;VkR
zI`UU)0h$ubadjuz!ezz)pOKMLD!g8R&==Ht0d&%S;NGNZZ|A*&Wgz*y1=tc>U6@gy
zPN)tup48+xsKVT#O@K93I{)+)h48V29fk^G=;MdmM>83RPCh85`QRF<Xn`8M=dx9a
zWHze7qZsOx?@8h}@FKg!3Fmo(uo8+p1gjU9u>MlOzkm3KoWZ$t8L-1Uwp~?#ZWlrL
zg0|Lc1!Rpji(7?l2<%pwRt!6;h4?T>gWC6Qp4?_S5>71*Wxx@NOqB#X1cGsz?Od#~
z!E%Sa2k*?+{m$8PZjc?av(R~eIYt449(UQ!^UN)|F|(%JuW9e+e}Ej3f*X7LceJ2Z
z@VR1H3e3%r*|jr8S}?4)I_NiTiN(GT^EVlIU^Ac+Z4>%V^QEb!1+=x|AiRg2B=!9e
zx+S-eK?8MHW|FRklBm{<%3s&J)je)U$l)KK0__jb1X4N{QNs3Yap%o37oWXEMG5!@
z^bG#WiPt0~W~j+|TCa-`z$DzCoPo<RHgzGUhTrZ)Jg?gzLKzLzx@>o6xoPABRZBXf
zjUC)PO4qsKyw;}w;q5KSq0C5_&eZ!}Zz7f-1O8gF<0(&u7^OuMY2fveW2YNCTKMad
zO(w?D{Eo&+E8Q_8iXbo&c=alg`AFM9IRA1HgIxCTn^EXD>o`x752onm-VlGkHd?XI
z%meWn9&s%GfYKmGH3RY~2_E)En!qaQeBi9qdBN@n2%>*x>G#G1dJS>;I?M^2(|}y|
zWqROs4*ifijpBf4d7CBfDZMmdv)qp*!$+iIirP>xrGlz-1t39MzcAm0Lto#;_?hYd
z__pWN{t)>4+UnmwAWzykDmD={YM;F8RJpppqWo=<HZ`PkMLGXG1J&#0xsa>*RI{<X
zW_s(jR}o=~)x^PAObD^4`uc!E+$bcbMO+^2N2hKXh7C~&)I=bm4IS2lh4%upOL`&j
z>#b=s9_rZQHf9d{426fCY`IFipq(Ys)N;)C31Q|r*TSml3sa?juXhf)8kU{3nmr<r
z_rPWlkE~r1^k?iR;-<{v__ZzA73u*Qa(<oOFAu!KyaY~p{>07T%_9Y&esFc21qnsv
zEZ~W57@MA1Cgf<h03Z;`sY|@Bykr<zpwdfug`k{goUtu!Zgw~&EpRB1Fb#5=L@TYr
zgyl{&h!=(`MKDg4Ot4B)pd1QYo<HhSrNh?t<NbkDX69IZ2or>d+mymf`4_zM;c1p;
zyslSA3*p{q{wB!6{__#VuP+Mx=yt%K<7s``Ge>0Fuh1{gkH+o^kXtj7HA*ko2aEFn
zke$g!;VOUkZTSfr3C15AL0h;ys_Q5Ey8az86wwQy{lyzBa3J&Np5()=>(F)fqGG~P
z=cfY_#_R=T>SguIEc3V~l?#Nuz@B>?(WUJv2ah811=z*3^XFl#_lT-LO*(^4)9Au{
zqthsvMd=OFEP{B(>bgy<yhBOmXu=bu(R3;ZXrV+H4akX3`>-HRV*Mw1OPrh1ZVnQt
z#A5GJNK)djlh5bWS#;QsQf@FKKn84nMbSR1EpxV6vPDaTH6>#iT$TA-bn7vps^LQQ
z0#&}U4Cn{8TTJx>lLGH^c(v;Q0SWdQRqKb2knp(WF*|Hcf%FRfRQw)g@R>3x$aq-v
zh`JjbdkzGeEz>gy3{|Dc#*=s=G4MNZXU(SILRRMi2$!kMvDVL~vAEK?*ZbS+c(;jM
z$^($7H^<PfD4Ok+JRh%&*JzGhERJfNocm}S+y(O4q8U|0wgKJ&oyGG02Q#J%3hp@m
zU`GHE=EtVw_^h<=!2CrHT8L^Hx^mr6CTzPpSJc#B{lFmnyry;&&P48le_m_8kZz6h
z&!n4D+qK;~@!&w0sq&rIv(#qvHw0161&lIVIGaQbK3iE8jg`yAOuKSC%<+_IPsG^2
zvq={gWm#yMIl+y2=dx9UP{PEZmV0vX?o?XDm)GtUiF(&sBELX}#%lY(-L5PB)|^3s
z-g+DG7!F@+yv~7x?q0M2_wT>xPaPI4M7-Pe1lRx(IuZYjTF;IUrWUP?*{yTM*R)C!
zWQuS=DO?g8F3Os-3Fz0jO5u$Zb^WW!(6({M0oi=~GX{BLr7Q_J2F;;-G{5XZdkF}|
zZwh0ICP!bivddrsjn}2PHC;b)q@EK`s<o(kDogC5b>9MQ^SnSlj<1Re=o;}BWigva
zkMU_{HSl+Lu*m!G>4~xS(AfUGD!iMkWx|$dW(|2etcTZw=bUnyq$J0fn4V!{vemm?
zVfGgt^I&%gK8(jWr7x@NCZ4<;34Mp7GWkH@oInl7B-$&y!S9xU3Qte$eyAb~{gt~K
zBsf@WIJ!vG;?<M3EHy#ORgmbFyYM&At8c(bjB`_&i*2Z;?Hs5Pl0T1!$HlKqXC)OT
zqp4+99l#t4lHOf)25fO-C7Q5d2OCc|>2;VTav6wsT+Y%0pLNIgn~TGF{R2l`Vi~*N
zNQ3r+hjLoLcjn=Ar%R=uLjp%grv37_tix|c^~hL%Fsvjz5tHH9lyyV$Nec%~2iVh~
z>Y%~nKwuM22r{a3v;Mlg+R0^S+FM9pVpqA;SSOLlfOGK=C6K0m|1R+JS!1f`=Hj6`
zT%k|v^Gn3ZROYvR@N{_;t5?k_AaN|bExQ^)U!|q^K$+yU9HYpae5v&4Xr%UD@-Tex
zOAEl7`#foWC!^Q@9SLQ5OtFXJqkD0678)cK-ET0w9vO5Pv9$&Mn(nx@tmvok{8hkl
zqYT6P&nB1FZW7kujzza`TM5yIZ>0?b*NF<=*)B0uE7_=iK;L~@elZX$gWfYx{=l-%
za;9fqZ~MlVe{YikS4F*2pIEmtknI}#tL9@YO`_J8Y0n?AyZ4$-4l1d~DPst9&pFHw
z6e?+ORsiA|azKYa0xljVIx@*md2he}>FV&!wh44?*56#8+`yW|Q|4m&zU*lq`d5aV
zzl}@f-Jo7wcmyBT*>{MuSxW=`77@8*Zua+U9ylJa?nxunr65nZ(GC$Q65c`8Us2@o
z2&0~<JXqc?g5U7J)yE1AEU%z}K_WDMy@bRj`V2Q%C4!_K8^6g=Ui~;$ak^OTKw*L}
zS!3pMEVxGw_C<xWJ8Dw3HQn;a+8c6V@V*Y-EywtoV@jhjP`ku_6%zR``(5PKdS}fk
zsXO*uEOV_`;uv4W(nRC-(A;1abfso=?g0X1#+;w4Lx$XNYAJHP+2$!{U)7nO<J`H9
zPby?>a^h`%xs+GhX!ljEBE$i01(ZVM(-a-T!3m;@a7Y%Y<~lY|mPUb_j$uJRE;Jdn
zc0bsdZdIsw`}Okv;efYoZQT=Rf=xTMkQ4h)BOXz)Tv{xP;LyPk$@RQaC-O7Wa}mjT
zIqlX7vietCVjhR-wjOs8$|_cM{D|+Jq&GiVMgi&&TNb4=S8*5SWiIGuUeb&|uw&q4
z!9<vrL}l%zvKw5o&byKHm)UK~<$X5z&M(HMwP~bNSSco}7z^PFVLPSQ^2JuZOp0`U
zgBD-d<N{D%W`>nMm>A^O3n?TA2gBeT+P|)t2c7GlQZAl$1uRdm==~N=3CHiP-)ghJ
zthlDlfh`t6iKz|q@Y#t=?^5t0x=z>+SC3jEM@l}m9ZEFa`3h9{xqP@yJ5(fl0}<%s
zeaLn_f*F5u?0wb5xb|ScDF?D7Zb#NiLDzI~+ACFIFSOZ~j)+-jFQYqe9~fU}dM=10
zsU4jrh44X4()8cQibLS&M|irsw8_0Iy={wgDS~;Ar+t0z11a5Sh1Z)R$dz^Y!qYIT
zQkG5qXQ-)6Q7?GVD>3j;DsM@67H}gU*zI1MEj0r*x>}4q!h&lxcvJ@zAp>&_S$ovs
zF%qdk(XO#r^jDI-SP^!ZqfhZ9)3(UntxY3z)Ita}!-`{#1Jvz&Q(E={>2lAUTg%4%
z*Gn_v&iT1+%IEfj(N!L!n5t1n*$KJFt57fMrTAh(!w*GgXPa2O3pC*<ARFxRqBn!A
zS1>?2VIGH6KIyiX><2!po|K=e*o|c8MV)=N8nbbCT#g>O#~?*jA=P%ZY*a&E*D8L$
z#BgGW4<-SN!b^~>7K}cQaCLDZ@t$uGen>Ma;pWsdBXgNrLEG<>Tl4ww6&er|ilJbG
z8{*mUT3{uDwE{~FIYAf`vrStp5S`ndbdjAUtp5TH8Fd1MPc$5`9i5Z_6?*%i3b0U|
zirWC2d`NzCOtk4Za7x&0hX%GirfQm94;)CI-RJ2(3T(=$sX?@4k$5t#oI!%WlTc6o
zuy9(x74RuY21RTeyqH*v`u$><j@Sj{w@xpIKwF~uwH4CYn6zHCSTn=Bptdg<JB>%#
zkh(YkPp+_V`*i(T?MbvzjWQ+}_M_%{&R$Hh1*K#Rd)bcl^XkMKEnhv$z%(JUpoeNX
zx4(>q$#G#c9AJialDRYnI?yblM5}wfr9v$OdhumBY|z4z_*D1Xd1_Y$ea}=qLaNzt
zZPiEzDWw=?ozLB`4Rq`-ehbSYk`rqd?%I3&bfk;f^x)S4LueEh)*_Ffjkk^w8j2Kd
z({~^Ks_jji!zm1$gMEh8CVoxkz9O{~Nc-YkKM<@=`2;vvReKaZ2EgJP6+Qrz01|dK
z)0d#e0m$qZH^nh)>Rh~h?+2K9LC@$R|D%VLnm;7$q;V@`S-+QwFFk*1>iRg<x1XCq
z)oCR6*nO+0#d}mRLm^uUOGU{lW5V{IrIlZ@`<fC0Bk}LM`U=(I^@ORv{X7_lqi<{C
ze|zSIwfe(zB&&Hp^czt~)l#33C_YS_mw3vU{~@~~u@^c6p=SKSQe%nfQQM$>Jc6`8
zFGM+f;n<T`J)edQ%BAXQ+4PfS{HT59NRgZY@tGhRZjq;zc?@p`&Hk8k>E;Yg3Ff~(
zmzGbzo^0cwh<-cbeK((l(6?yl4=-kzEw}4?Gh{+7R!>XJOBlBGilO+SNYQzIKS1J<
zeC_yA8{^8tMCmZ~tT?G{m_n2#mdb?zMq_v(<;J=iqSh_<-8^O34=2%z)qde%1ds<(
zN2HBtq+I{g{D8f0HN6z!a`5dN!<%92y*N;=Do$eOC|lZIZqz<pSd+VMjOLc7V{NOM
zjLWd8n6#Ppm~+{p&fX)3K&3r`uL5l=%ezlX4aHragQoy2PdWOJDxD^_V#FU`-)bm+
zA2vXY<;oX~zO>(3g6dfpb&jJPR1bM+NGw{TPy}02K75%T8JU{4u0P71?dJ-g;(qf5
zD|kNTwsWOa`|(%@f?#}hY9eoShBR^#a=CvP2<LOi-1a60H#GlYbMlsBJCD?2Ktl)*
zgFQX>deJsME-te>AMHAj=bp^v7r)n_Mu)9Jn+iO$J@@4<dxjr&QYkanIfaP>TU`Ms
zMe{wlnNz85bfdJ;bSk2(d?o7*LU-nzHbe_c6Ekg01ya~w`t>NbpT`@Y`!u>Y_?rY#
z1+k)C%R^6YgZeDjZvq+_yv+Ff>BHhD0&Ru-1JUZ@wDS<TR3l>uM&>@&nPs2KY4>{)
zrjV#B*6-cG)>iM7bW)XpIq+w}Nq;{=RiBBdK;Vb6)<44%GR-^tN?&t3>Q;f;$Ysi-
z4~N$o>7c-h=JFCAtX0ruF0Vr4o^Sd5iS4x;4sWfeks^xZA-djRWMGpR_=u<mkglH2
z#I|zOW6f`;0nK)_x53!Lz38VUqi!ng1yM>aX^nDZG~Y*X!d8wEXonP0Q^(K~bRsO7
z3e=6n4GOvG_87Ij-aB}bDYEj&OAn%u+K6ug4=1>iZT?yM)7)TBa7z^__-1LeA7pC8
z-Lbsxqa_my8KAJ4ar7&%ld<VS6<29SLgbXuZQgX0&Ihkc{UduJ@+vWw+++{3EE<Fe
z1%FWLI2Pai_(lIw40T!wzu<#o)~^5L{GVMF$a4{H!=5tMMM`vHYA?EY1|r15E0H?4
zMN^N-)!~Po%VK}h??Jn+{d;eONO(O;+gXU~!|%KbK(bfjcOdzQx`7pA`gs@ZR5kri
z45B?6SrML_$_Mw6NjbWSb_wh!YFsnCBP;*Wd0j{oNt&+PUt)O@KsRvg*NF4DKD-gB
z(>OfW(>Fb5e-i7byyx;YTDI$j%V8;d^NsGZ6j&?a&tbN^Y*&HN9vR1*Ct`J8g}5}s
z%fe7glpzD0ls@ga2+yc9RGe<Z+q-gwt%L^0HthS#kN6eQW7tibKgrk5M!aAWNa;pg
zG^&9N^{3KR=drl&sydh7HlN!WKz^YT3brF*C;&6Azf9EaLJ~PWrE=^xF-$;jNg?AA
zTIz!pa59V_v2}{Z($4%6@xry!2G&<yomW@hHm>)JTzMO1jDD<L9yzhmh%RZ8Dw_Mf
zqv+*mB^W;@rB|XI)Ee&+hjHKt<QpGj0c~Y~aQyPJgl`TdE$<|Ish}#Lam<hwA$4gt
zyNN0R5+-&uoii85(#fzx81Voe6oAQuzFCkV`+7|*idMyE)NDa{SI&3HmwTs(RM^Th
z(GRld-W?CWjxtok(uQp@A|t+{daep#1pb)$3kEfUqSUOlqdslUAQw2)(9&<&oAtUZ
z@{lXfenBb7*^~m<jiA=BdW}Xd)w`6dPrMSWwj!GSwtXe<aUsFbTaCSdNkk*`p31fs
zprC1Wz$NDm(qqt7{*;~!T2xEq>iQTeL>L)Q>M~iL66NCfhKN1Z{8YmzEnK+m@qqbS
z7+*)$_AJq3{M!rz3ynsB<W8^?1ubgW7L_lC;1nS9Evfco`@GhHc@&c}Z3OnR*9T*{
zcLG72B-3%2Yy9SQ{C5)u_UQA%>!!~`vN43$lYb8WNK(YdYA3#PzkNijsE?6=3E?}1
zOWE|bd3wwxAvj2%5R0bn9I!TNstdl#pX(M$KD0CTix&jN+@0ob2m>e1&k8t6LDE<^
zQ){n@`pH;Sb5_9*=Jm6D&{%H69YH`!FBiDngp|*?%|^=>F9g2G&xG=&RAuL=1T+}H
zY)*Xg&8p5&D(0&tM$vq6odgSe<d0C`2+=xv(ZR(EFlH$GEc0Gjq?oVJ5u#}>SBKzT
zRYVBPc^P1cWe?KLfQ8d<`IuAOVr*1uc7x6lO@Z?5buTPt0p9@Kno@4_@50-;vKtM#
zoHD0zJQDkO?$g9tXOU}5cUK<wc1E^8dp1z_sS71CmgwFL$yIXLct66ryk+{B&d;-(
z+Ds~tTY;8){$9o{1t0UQEl*IjEw}4(Yq+9$W5ExpgaZ|J9$rLhX5*OG=TFbR4WLUE
zN`#S-%IC>hTzcQNK#(Z?<s^UGgKRqiXP9LIo*!sf1;k_ghm#l#<dr6xH@S1GYF}CA
zJq+anD#&L0C*XOsB$Aya=sEgZ^OyR^A=dhg*xM>$O5#WJ^1WxWJcY0sMBpRJX=pPd
zmb&Kfgh06Z>`47U)H_B?ynJT<o7NGSr0Uo7b)PstBL`8t>4m9dl%o(mexLGQ&7^P7
z3qe6HKl>muljMSxa!?5WkVP-0;_AC&yYj@JnycjqnXqJ)9~eLIfSWMuKw1P~XGXgI
zq$$Yf4Y4s@S0^-mj%LZOYSGM|9FA+Zr-Nvb%&f9Ol6%6P?IMW}KU*-P3I>dbC$T!5
z#dIrXUY@QDH(BCrU`1|N1x=6-sb8?J$<smtWJmm_!!$H!)%=Owr87f51bWowDwj1>
zMY1B^D=YUa%F?*EbW-C;Lcc4@2nry&m}FBZbhS#B6r*OOHabYN5ZQYP&+KgPOfvMJ
zBcXe<bM9S+scQlnCietanQb@OVu!y<h_hJKE-yDdh!R`r0>gny7lQb?<h}xdt~SY<
zSaZACuL=WN;fKn)_d{n4SmWDk*^}y1yf%?Bf@h91qbck4)P-R1svkjc47vPQwAI81
z9$@V&zWaRG4^RqrS$snpZGhV%<0vs^zz;?L&R$-POF8FvuJG;9j6Q{$+7W;~xWQn4
z=-^b=C)kz!U}fqy-*11TZfmUe2Af_o`sf!tp3YG#3P=Ac6A;*~l69xkX)#)5{*5on
z7Qc&+0wzM4%q!p`(=ETfLPCE6X<0d;FNP~6`Io7PE@RNhtJutWs$-W8{v3rq!{#LC
zl&MQSGAe=sDDnR0JX-Pkz5|F3qkc~}Myp-ZBF(W32ge&(z&u7<2^MnWZo-lFE1;%(
zYnnaIPthWU98T2MW1Gko8wGqj6}HQ2c)@mqp5#;nb|}@O>Yy&>UbsiQ0!PA$_?Bag
z=SW*iC^L;GRkOsN*EvZ3N<Y#V!>jT#Glu&A3rQEY=z#bodmjQMnMKJ}auKl3i)5Fm
z<a9(gMOCfJ<F=Qs7-F9kV@=jGgEy{dyJA-U^FfRMEk+gri+BSVb7$94Qw>*ZFE=Sv
zg!hEAnJSm^NH&O8c4IzFVtyXrO?}!b?2Y9X*L7fQcYM0?ed>Jhs@!Bm^js1XTP?B{
z%33}2YZP2)pk3DDy#XaPxqp7eA`nRSu!MWtW9GDsjwLM=dna%wgmWlgx@e3vML2iP
zPmIhvA^+?XS)U7u8HuItKPxRW;EtS>4+V%Fbqc4~R^7HpeovCc5^(@3?O_mXZI@#;
zYMh5jem{7sBH6~=(3+r6GXu{YBIy-WlLB?MGXWDY?vg@17gAKkbR3b@W{<>^16b&)
zwS~Dsgb_OZR9{STx*oK%w4HnU!Mf@WG1cYvC3Iv}+9m`i@4c9gz!Wj@LnB60c1)he
zu+=~&0ndJ%e&tZW<B*8)UEFSL0IMU{XJ0@KVynD)bjKu)O)gsp{A~P&ZFm~Mn$@AO
zoFv@TD+<ONK_~!!fZP_huZ;`+&I2qxkj$zz^RB`Y+6JI}R@OYHkBS2&y)o|Z;M89N
zZ`;cfrQ+TtCdo%40%Sncl{ewW1clCXo#qI5%xp8BEkWS=lqsmYab)ZYJSHDGjf{vD
z&mq*dwl^GGMX__KkWz*(4iCLZBKEsRyQ4WVy_wK%CbvaVXnEFDm8sA~weekd0}@m8
z;moP&KZx?)%f(YM9`d6O|A6c{2p5{pE0}6UvTkI#7>)#Dx>8x5*ff3TS4=tNL6MVU
zBaYVJRD8mR<~6X?ZM~Rv%1HN}bw{K<LZs}qbSA~IX=U0jt5^|6wgu+F=TJ%cQ8(B^
zC-9opbTWezGtcQNt(50<y4PAhIR!5(Q;n7=k3Seb9H-GE9E{k3TeT>jd<nrae&3T1
zaMGEoEU3I5M4%hWm)u?X8m#Bu-$i~$_t};Y9tc~4*@{Z9Ouhfb+gC+o8`mI#qr@-@
zcXMOct3YfbUkx6(*U#7iq;JT+SF?_Q(sxNAhwBRi?MPxC)+z2)@M2p!RI0(qG>(Jb
z7&<QtHrNqPP*pv7p>@xn4iRl9?u!UF=*~-Y6R?!M)pIg#Z!B_+UM#+8Aq|GCI?~#y
z96Ok*(L<jO_wnV1MxmdiJw#R|`H8!;>BbpYVMIy7DezTFZwb-GyZ>LvNI)}|WBQGm
zXu|M%PpG*f1meVN32admzLGJlFi@nG8zm|-bPut`DvT>ZN<|7@4XVam7!jR3$W9*3
z_;G<DbDZaSg2#8pj#f|mPUGaEnE(uvdEF<&rWcJl=dlxV2>W~kI|LK^Lq6r?L+BF8
z&Mmimg_N}kVvs+PuNPPiAZB$Oy8r;~i7;ll<OU%{*m0q#$YE9Drr4}~qi-#h3W6Ps
z;1nH`QyCnR*a&F9xA%07EI7_i5nT#b4>HZA9ase4Gd`8>gy+#uC<#r@a@BZGbz4P=
zVg8-G>B%AE#XAv~pCEA}g!N1Dz8^qW#~{uJosa3JB4X^L_wPKxIN#~Ea|zJ`1>6l!
zQ!$sokCvzFM4fz-vs#k9Yb7Ya!ApgCR>LoVMCW1jvBe!*C#*NgEh*<cQNhJQ&=47>
zXYD2AML$h`^{F#Q?aSIWiWCm<9?HF4)~vAXAE&o+t4m{%Thz$r@HB2<#7tXGA`)e4
zc)+jC4Gs1&XKBvk>cB2i`z^tS>w?DVu|ogjRvqpQkxx^cj<SV?F%Z}@!Hp|OI7xE9
z3glIj@lUE5KSz!La{xkoN@5RIPP=g#kj#{)pw-ZQoxw1Mc|Du2#{i#;V9W8*cvtSS
z2*Y=-rjMtvDG8I&wl+EfffV-(d;WY(UFt)_Q!Mzz1ux#gKbCI~nu~iptzO7Xjia(1
zs!mR%Mrje~YEKIZF_?Pof?jl(BgYS=4vrmg8OE2Mq@!MD8{r$ZfAhPIS<d}+BI4F7
zQ(s^lP@3x-Gp`=NWWcyQNGr+XoWN7&z6<XH1-QcziWSE+3IE~^YfP<Ey{CWf9ix`%
zWS_fB0hrhiWQTZnY*?VJ#uGE7NmSw00Yf4E$vDqio^?^@CF}MUZ7Q2Y{4D)LEKKKG
zZD&=sK!K6s+VCqe5n%Q;;A8|&!MH8b0`N8?Ht3+Uj8BNiQw0LOoCkBdi2P9Z^-_%I
z)bHrME<yu7d)^+_aGB>!$P;)*!WY>i|N3fbDg;Y#b-C&k`JMExQ$JPa6HvJFCFa)M
z`HK18<j=Y0yw_5=M}vK~J|a{!`)~_HUnZEnO&0-U5@-_OrgoM$sbb^7yG*T53lPQc
z-vjz-o~%r}dQvmQM9=Edc{Jj_3BH`{7rvFmg&?PhY<(1~SyN5y!!cDt<by}69p=V|
z=jQ44`Hf?e%2a%*lVgGQYrfryEDDRO;A|cDzdAOcSaJx>b-$HG&h^MlpU<sjpIVRW
zw|}4<m>0osadbWnfGa@c6W(Zg%&^KJ=#OChzLDvCdRVue=kx1UePA|$lGub4q<C=q
zf29h=v(PIc<0H06IVe{2k+?NaRfNe*R+Ug;B@W~q&t}+faF)QWve78o&wO}s{uE$4
zkY$>J5;%B&u=C{SWrK6S6?U*rt*yRcNT+_*G$Q^}B%)%Sh4p$2PTwa3g{CkU6&`&@
zH1+v?M^JO$bme6i7O_HJBySKh5m0lulCSw6gCbdp=#)c@AO*J7oh+TDo(SK)?3K6=
z(p(m=@7GcUdo08e<V?746dQ!3gDH}=%Ojg_r)%^&zBgT@JgQ%&QpI-**1eH0z<tR|
zf^*$sj4SQMX?1QnN7Zi)EYoW;FS~);As^%DD9K%{Bki2q0CtHKVYiE|CV|f`U%9!v
z18f9|9h>%fYR@w^bneSuq(yDMl$q+Sub!KCYUZ0ZFPtJv&wp|(>;MiRJ$DY3vrF#4
z26yKPV%ktvt1Rj=#go7~z=m#lpp)N4zCKtvDW|<zZdbt!fe~_c30V${Hs(ykFJcl8
zYH9HRX355M`PeKIJmH>CLWm&}$OtRxZH{lcEXRq^p2k~BHN3z78;Ma{IBHfu`8y6F
zD9z^MNhN{wliAJO#=5f`FB%qiq}0p&1DUXxK3@F!*H(27kgS2GDYgI8U(7DpqNk@o
zBL$l3aIjQ9ACQ$lpr+e*R&5H{T$a}(Xr)$q8>q{n+*X*tTrM~itYYr&(Z<l1hXVk&
z1CFG#!$LxeTaA@`$x-iv8kEXVGtRIO-$@TW;WKgc^Blg8l%#EK1X>IDdGBStW6f12
z$$Nhlk;e__W$^&cT|!wIw_krrG{@4<RuDVofGyzTE(go1ddw0Hi!<g($8Wy80;%jV
z?Z-$XlI1tUgD>wf$xB^B{C?Ni2LZWm`588g2CH|!Vq1W`r#~Ot)mQbHWv+gKR#MU}
zIP%n~q9``vUqmS1D{0m>ULCAL%_8O(up|n=wWg%XxW!Fj9Zu~9D%=r7<>lAJe~!^t
zd?3nLe~tH)LL)L;%Q}fS02FmlnWtC;H=K&df$PPd@#v)pyy{qwn?O#}di=)d0-((~
zBx`9#hDbP#z}Pf}wKsxfp8x96^849cD#|nT)Lomdj;h9X*Wm4q1JV;Bx;T%az>e>J
z8_kqz{(sx{AweIQ(4ZUVAqp7w&yM`XU@Q`eo>*%ym7#N03v{&H<iDD);^8_&ib^g7
z>6!Bcbf-T`TV%*(U|II0@Ie85l!@;rAY!{AdWCwryacea9sj{WJxDvCAYOgg3#s_K
zEnjgsb{{2+6c`=xB2@T95EfS51|~&mb#v8peKnL|ygT^yh5W}J@3tMkK(ctKMjCiS
z6hx8638Q8CXxkZ0+*3DHRW}`$v@U!zmLu-BPcD4_r?Y>XGEm5<4L}%IbohTA!?xGi
z|5VT8W^U2zDSL5IyLn8$SNn}z<cEqo<OglR<?t&b2Xo&Q7xoRWN2mSjx)5`>1z#>z
z2mY`?jY1Q0fST_JtxDJBkX;}dkmaQ`Xq-JTLU3-FSRm+V_{}y;SLq7!X&X{yxI8IZ
zZ3zo3HN4pDw=Ic(8LGg&D-pGtz~+X^11fraf2c+b4-bPAjy!4}ODDI_P5C;AOLd4W
zlad&owlCMJmf5!vRK-C$0w+%(%+l`sGO`W67%71aDLPoLXjAYxk4GV8@6ByGCEhDn
zGrP7pP;p^Pj^6JoB&NuF;FCya;eOIQPXBK^&%-p?+R+T~e>-8C)#S7g;CIkl{M-pP
z_#8hl;ydEAC^#V^aS|IN-)lwwv8&w?VU!)qgks#<p@azkB*@VoNTc7VdYw3C-hcko
z`+6J-`ubfvZ@cd!U`ln6vxDJ`9!fNM>w=%yzr*Q4-Apm_-I-@3&4Iokk7p*&745~X
z>Fzy~4?PmsMaBHl^0GT&o|l<U!z;KUI$At7_P=?f`w?zqhSF;B5Jju+{yRBbx19S>
z!OE$c;PwJJ3pjs-kYUBg&t7pH$QhWhovAY;T25B;C_EYR_FTlcf&Rp4{GVz*bz<W0
zcT1L`u-;OxF8<BzR>E0B2`T@2F+R^vy(*5?&JPVfg8b~09dHUia0A$2=)r&}cm5e{
zTg|ry{9vvO;WjS`>U`~)uKa#-!<)8kBBL_FM>6I_6Hn$3GK*2cP>^|v(3-?znZAI?
zECr9X3c5$mH-1G##O*nFsHOB-6S>``$p&cwo(>jWCrsq>R(sy^?g|gmoF>e#IS~i#
z-R>CXE+Yu}4uyfM$Jye2gXJ`Dz<xMNu%}1DT*4^Lu7B9g28vlZ#bpE{a3|p-Y?LB!
z6;p;9aO8<1*`ORiq1iwFOu<FbsX6(nD-`ED7o@j`G9=V><N&jo%q?Vbqz{Y_|1p(s
z#r&P{6{+42T*~_t8};TuU%&b`*r(3yHq;hcjC(Q<{R~m2h6+3bKotD`QzPmSdGA!X
zY6YwAbGp!;hQ?gkewEb)<H+`dKSXK6_7uBW$bg~{7OrrL0D^f}9ddgX8V?9cGOKz-
z*sE_GL)1D)GZOYbO(@qlTP)y9QrzQ9^QCaoPp$hz@2d+`Q8(+I!L6vGF--C`p>r%%
zIQ)HxT0{aCYOJ@oN(Zb2FUW*+Le(qf3iM>w1AWOgh&!pCZx)4iLpjM=1`O<4R}<V>
zXbi6WW5QY}2W&uD2S_<)nz&z7g^Ch~yJPq9*#TU$dHkG$LR1V3IV)&tz7JI3|GnR;
z9DoxH3(xaD4pvGZH&(65^eiFsWtO`Nz$|ULo2c#$zqo|F7K7j2od^>$4|q<R$<9(;
zE}x)|?cs~?-XO<8#p=msGFDfZ9GXq`sF1$yeAF72gb|(nJF>bt>zo2F*RY$&f6%2+
z?1T$@Y(flA@{0VdihzkJYi>N@ITay3f5K|mLj8+Hq&{;imrCK7Zgm38+{c^Mg@F(v
z$k=Ci!`f#sx&vb$)W9z9=4?xmURcj^;cWQ~CpjJLN_W1Z|J396*Q~9X>%`<d{pb4>
z%A$IUf~HbN$zF3&r%)`-%z3?7S8MPlD`3OQAmv=$JmOFjajaQ#qyu+82GJ`zNtb?p
z=e{F-Dws~B|NMW{DvaKv&oZS9`tu!#De=jQZjJ2*fCX@2xp&9Ui-^TbKj$&($8%O0
zk>sX#X<oC8<T(x);0S788y@Kh!EHj3#m8H&;AGf{nge$EYVw9gYj2dj=w>nRdaB9!
z%_iUw!fXIKcl(-tg~7<p8V;Za>2_u#7AG#=`d>68*aC}RQZsFP+9(Tz<g}}(!3`5;
zTlsKu*cCxbv^81P<1=lqC)7CY_xbp$vud-PeHR~AL?E{I+TY~nCHO;uT5&j~6G-A!
zUc+7_av1W|{S>tx^IY}xR7UE$_Y_#<XAN8l9tlHJ!H|62X;Py)F-FBqbxXJB`|NcE
zl&8LHHes_EcVGu_Ki1}=1!S<9mgcLd30DZ0-I&ZilM4*f=Q&SE^w8-bEES?&Be&eo
zig)=&)@l9dt1BGem;6<V8sSJckWG1XVl6<;b8{^|8(j5V<!JFfxH!~mp3e@W9V;?N
znqN#kt;zT0ldmPbf|oV$9x@01l(hnWyj(UwJTto9L3GW8ti!aNLX1H*B`|x<nM$P4
zA`%b}c?kIqD<81x$yI?3FiO4UE{jEXP*uEU6hP1!j@7D)7X^2bWiE9?ao5^m{Wqbu
zyJQMa^i8RS_$qP-9}FPe2EokuW<vPhZ6RIMB)tH1m*vCq>~VWpp<LuP7?nkdhw`W6
zYfCs!NPpIs^&3kb^{wqY$J@asU~)Ne<JSTano9-PIu`;H)Ry5)zMrye-=ZN;C^BNd
zy~$Y55xe4B)D&pqgwu8z*m<T}sy}hr(a_H#6E$PnkKF6SN=hoMBLI@IeY8zsYR&@y
znif=!imO$71%F-GKREd^%@H@;*D43IWt|UX>Q9;hSw8Ulk0=<3fkGjk7@m<RvP{T?
zx$P=j(z16w`-~nuPzcHU61*3DAfz>T$Dr!;uot&ifqt=BrY42ZCY*9$G(b5*T5W_X
z9SHR_O>%sJ!C_CwqVDB$A^-#`xRF_J8%h8TL93jb3GVF$s3vkS9?cDbAq8(=<4@rL
zFVu{$2vHC<{~T-+qdqB>aV$tg;2aYH?a2`=+XEl*jqGEDR+%BtzX;LS0u|=)T7<04
zH_I=EN3I0P2W|2V5JyN>r{K^);2fIzuglS|Ikx5hPcTbp8G6RE%MHpwTz+7rHp#~W
zDE5!B6*A<+)_*nYvz4VMD2eJ`VBWFzcq(z%H41Cj>)!8@T5&wJFpq3{uU))TMXoOx
z0RC)B{)e9b_!1K^?JE<P=>@ds!7V<1iLOv<QYCtpbPtYb9V}A@A&Um?YM=RO{kE^1
zzzj+#K*q@tw<%}=^_sL0Ctg?8Fx`f&9?~Fvl=PAXAK-9o-<~ef0zcziAAg^Ut2V1S
z$ORd0=Q>9_|Gi5wJ~q=^TAL1M3090r^Td2^yvl&9&$4_pChutZVfd$!$#h0&Lfk!U
z80Jm7p~XYzKFUl?^dx48wv>4Vf;0h_cxARcU#g)W0|XRbLyW9NIEJCrINwOWq&<C~
z6!tj!iy|?lYe&9}68uiixX<fMlw%5MSq)QS=5>B%pe_wl)M-?y4LbkGDeZqw!Ie&o
zMW`9Gjm1teTMEdN?g=4HzwpzJ77Uza!y8Fq!oQSJ{*D{frj5dkVE?x*GB{B0hw)NS
z(7mKuGqoR><7h6Ku$7p0lZa6ZJoP@fZ^xz$dH!d5k}Dva)n=Wt7)OI#)tq5v?<sU~
zx0@xv`j$k)HeM`c_k)R`x}tJj3)S{T59z%Pwc8+4hdQhamei(UQ0o`>IhV^O8H!tG
zzXZ=pRMH{?M2D&c*W<Wr0nEX}F_t~j2iFQY(QDsy#M%G{)oRd}E-T8N<-$t_%J1j`
zW8|<Xzf`&lbgTU&+frPyV;X!Y)su1*=<}vQ+&m(5;&wd;^N__fCYzMDM#BPeRY&eS
zIyxETLT0RcY>O<xHk<90pT4=k?c;jER-+iVkiEo~?Bx2b_n9`Nf|osnhOZd(9bT3K
z_UiC2)+;o^_7la#;7k51Hb?yp#$Pxe0|8V0#^4_-n#rBt&`xPGoYuMT_AFo2)Ec?T
zu%7PFT7qXI0qoG&5*P~*YD&VhMY`8;Bia;ylFS7eZnJ&sNp<H;k5yg;@6c!FudX!v
z!2K#`uWs0m_%3i+?{8s{{B^ng!@%j!CZ3J)$gi_04(3v)h4k~Z)nX_ue?8hQ@>QvX
zvrdmIkzOrjiSI9X{>3f1FC8Z}gxHDAXf{l*tg&9LscMO${a8DZTg``%V2r!??uYl2
zO7K-LJCEX>Z{2!Pj33#FX-TsS68ae?YNqDLUp~#ESP<L%9y6kB2pmdf7eFz1TQEdG
R000000000000000005hd@1+0$

literal 0
HcmV?d00001

diff --git a/assets/figures/2025-multimodal-nvidia-nemotron/figure2.png b/assets/figures/2025-multimodal-nvidia-nemotron/figure2.png
new file mode 100644
index 0000000000000000000000000000000000000000..e2b20bcb86f61269a8e8c0a6d4537710407c0834
GIT binary patch
literal 21826
zcmbSxbCjiDvSr$pwr$(CZJU+0S!uh{m9}l$wr%^)uiy0a>esK|%v*E+x+~&-XT^!w
zv3K0>9AznSac(srAPq5LMRi3^Li)eH-yQ&G15>ktFM;qI(I-oi5)%^<DP!dYz(HI6
z{N(UuMWZLZA(?)nE%}JM5;*a8JM$;{#`s`3*y-rX|1_w^lfbJ=07&J8JiC6ae4V`6
zI}i-?wD=Ew*L(mTf!}!mDDNdNH$FR8zS915fKY(@3+OJw-ORQB2Vl<o$qNl&cMkrQ
z3&5zkDTM^YDI(AzY}AD2UJ%^)Kl`Tv8g8TjBMq>xkbv)Yzc_&Mr_M{)_R6|{^Y@^?
z17PaHejh;hRr0~`y7MTo3%K*I{Al_VcvDD5u=uk1`k2i<+%e=o`G)+Ad(XY>Y0h=(
z2>{4^GrXuw?M(Xn0=V8pzE%!l?`jSS)_W@a)&QPgJ@3u0H&0zd*aB~Wm)x{mApPzy
zgfD`Zn=Si?oi+k9{-tk}PmL@4HiIF66@cw+<m=@U@-6p);f`R--wDv+4*-nZ*?m`h
zFeDr7`fvEReeisMd>edGd>OpNJ^Fw2Ec!C}QvpOid){|GdbWE2fZNx$cgR<RH~(XS
zW&eACz+V=UH;c!aCxfk?Y5xlV6952k-`W8f*aKPw0ADjbCIYXK!T@a5M+B@Qa6Zss
zD<K8JyG%uuSzhyG)PHY;-;spGJc+d?Dn=6l_0;jz3N{jz0kO1wBzV9|5AXYKCArQ1
zSZIL@x6w4((()id4(T%knXcfJ*NUY3bC{);DmL>~BCT0@x3>&3+Wa9P$fe|F-~=Yt
zt^w*hO|!dse^*SR;0S21Bxtbrx4w`Xkj6r;ZzF2|_`^}95?L+sWzSctbbt~lLdH<q
zf}hV#Q%*d-_D{h}`(pjrHTNj*JFGvixWpDTXmdqQ=DoZ9qJTtnjLqV^m3!4;fa#q|
zqJ38?6Dkw`o(B=^uPP9yrHEey83*R#oJ`<KSn<fjlSFcvUHF6BSYrG>8lO*~--Y%m
z=JR@&FB19g8t-%W<1(C`-l6*4e{g7JBk-J61ZpmUY}js~GmWr8bBV{9;i6ZkjR=i6
zQx`<N$Ssydf1ve8FB*$BR=1yG_=F8t9B0#wMk5?bLx?JxHofU-Z6&N`wV-+!jcoqy
z^Ta2)1ZDr-a61U9FJJT7Z3#R*W5+G`Q>@LH)<70sjsfW`OY#@<G?w9`x>Jul>!D3O
zy>6U%czK+{<`uZoaBQFbst%C@#~yB{XK6xw^fAepp=!R`etX+kxU6>xalcR(=JGnE
zjel_cUFCd72P*%ul#tC~l!!t}<4Ao;p2rj=)hRxUG>jKy^n6fG-~nr1NiLXf2rb(U
zNC5y_`8plx;2XG;;jZYP@ba*y5Q%?wcpU%yKf#m2iEOW*nGtDV=h{;|Gk-mV#IoS%
z`zRKH7AVODp{cJaTwG;f!vEW~R5w@GUy|nzO$_t*K>uy@|1d&__MH8{QT|syfc5V@
z=KcFr{A02IogX9n_y23l{G+)TTfY8lUjJpWza=|Grc7ahuv=HUg0r&Wiv%~j9@lJ5
zD!PM;Db0+W`Lb|WFFAdpJ_*ZB*@S*v_W$W3APR)SXOWT<RYF()H--H#0r+o~M$!DY
z1n`gDC<yjkFN&o!{3n$Bm&N`DojijV>;K%PiYmiMnpir2^Zg6fybEN{Ze?`OtWkFG
z-~U*=yGi+7*%$eoiknsCB<SFzgdh^m4&z0?!s=3*3fOH8@n2KDI4X2R`fh4$-5eoS
zV4RlZA=#P37Tzrxb}%IK9`*1Nq|PC4V$KAZTIn}YfU^0PBwO*H0zsd4!d0!xrP<bv
z5$P!uvhN}O=`HOUPWcB7DjYK#e(^U8TN@S#A3MiK@|G(@jVqq+{YSoSYE5yVnQ5^g
z+G)eYg=&+l`x0=;XlhL*{xZVMHC9UIp%~%M-F$0T%N+)yGN^0~bP@@-89bC~+585Y
z3Hg?;NgPwV*P`y0A=4N$Eq|j;j0UeIEH-o`g?7~a><}CXNVGVLq5h}-`+K7JC4fBN
z(n*I~t}lacX*yF@U-cZ4)gL!<(fr>1?&BvMZ14Me|4+~6xh>oy**5){xo>V<`fmsx
zY<mDa0-aJTW;)S3i#O=+Y5>;g0>^6;7rGJ%6ZL8J-w1+~R#{Ejz5g5sZ3Otw-N?TO
z>#WlMOii9e_|M$%FIu9ikMLWm9hN06V+)V5=5Lgu`Usm;^7#8s;cT|P`M;N;T<UIp
zBYai5=Wo6;|NoGxA*He?)JG@U{4y>a-L{`J(B3CA1aG!Z?wU7$7I~$a4Sn~zJKoI^
zf|<XseuyxCk3V3SSk}K*Pq_RbshG|e`5!dVGE<nav1Z^R>rj)!z>R{?mYi)z858p?
zPZH8wzaQvYl)c*mC?rdaBWqGT@=I2HrA@)mFy^U(B%C6;tdH5uYu{ciiaW{oFxIUo
zrUga;-NSS%VQ!erY~u2*3L)XtYMQpe`c|!rUrLUb33p$&?|T20U9j3=g(tdPW~P1V
zBBtJUQ^fCmQ}n`PsA9E06!;O$u*%M0D%5jHaiSUEWn4j2DVAd+1x4h{=NG21q9q26
z$rlQL@#&q&@@>y>+@tD0iMg_XuT5>NZIEFjn+wf_J1m9qqUa`cid$7cFf2VBIoJM9
zgBPt%Y`^6FzwoJ)we`PP{?_>o&bHc!yJ#{AyJWx0sJg_}kM_cIeyPYB|K}<$4^sZy
zy6%a%;2g#^@}?;BEm~!*naGxS)LF)O(oOt)gP@c*vY>+bJGDOix4^K@BDkA8Xp@=J
zH<MA|T?J}_L4#Vx%M4tmg))*XUJKW2HT^#m0;Qn@gu3{DQmubR<~4IVO~e#jT(`h#
zJg8A1q-lAV$3WJM?6koD*22hhk~9WL*QYmfMBcC(lB7$R{}}=QowmEeo6_qcV0S~4
z`n;OGCX<nA|1NSi$M7JiTG}H||D>k>9M^x0WISdneX{JIm=Pa;X9a<i|8^sPdtk_5
z1$Q6u6d&$?6D;$Thw{HE-#_F%=Mt~&z27c+<aPdUoJ>|${#SAT;r&;Z4%{zHqV`FA
zpwPY~x-9!^RQUJc_^0P9S#*2`D9AUy2>wQ~tt&;}LYvGfdy<c%plDi5pz#W&lm}-2
z(OFe#u59(!#m#z)%M<OHYwjvK^yfDJ7ngO^-$XH>1>-NM@WcGdrn`qVqX%k8RSA0(
zTc!GVi8)Df*gt3E_kce-I(6!iILXOSz*gRO#yHSQ`UE$VJ;rV9C1cg>=+3f|iJ|sz
z=5gMsf@u%Jr?PN{4=S{`Jg1>{%z_*-^2}@)lRf6N4C+yFK<%l_Q_h9z4?3G&@RoKj
zJH-^LHmYQ~M@`F3D_SI7Nd$ml3&WEpkc(ieK)OKh<{zvyG>45ZjVIatnCwjXg@&j#
z){!C#IcerSY3?5bFQJN#VAJY>SdTxRSGGSyv8z0_YAkme8|iS1zl8aOKPh|inkt`C
zG0=AM`D5bcO6R{I1>{=bS{HgDl(CfgC_ullNa4=@k@JcHBl{Up4bWIusI-0%!F`i5
zdyM?=`TXxwzmYdG9dy`%e%nkpB82@Z&njoc5M%LL9wcQkwqzKpvnU`v6XWFx2+2ov
zdQC7=?*btBUu670M#lewn*SIE{zdTre|~u&*S|`xfm2(!`y5#vz0f?lgH=E7)E+*g
zU(PDP=(h6g7FS>hGmbv!%uv~Gu7NA$wri5=V96z@is>|THb*)yc^D>AHr7Temom4i
zA-8B1x`g<opm=!G2hmQ*;PpS6lK#K<0^SSs{ly9d1ObUOnF39rK{(&7Nw#w>h=aP$
zy^w3(<)!b*cWCYT=#)hFiN2vV`koxn%rQJ5h>|@oqDix!U6&KYJOtUqeqo_9Io(zA
zQ^Bw_*zNPHUJS;7i`)07ePNgv8##%1RG?p9em7|{7k1^rXL6gKo3p#>W0ZYudW2)1
z0}_pP!1`(1WwyUG+04R(Rv^Bl1iwQ(@700W6%)1*Q4fg#P9<(wHf`5HSKb=JhS$?{
zfBumLgQ}q!yB<$$NtP*s7eatUW=*J#ADyMg)A&1Rx~+!Ein@wJqJO4GXzO;&hJkk(
zw)=1A8ZWZ)xfOPk!ua2zM7eqku*eB!HwHTlI}7fszn5vxdCj@83Q)2_akiZv9QkPB
zd2J{{sLScFrE=T0`wY?G9G#M;H%HvhzwOF;Vx7ZIrQ2ULp6~+8VPjigQ<nvbFk}YU
zY<Rn~=xx=1SA+*^SsakgV&WfXOqBS4^Ao}y;(EK{xI{OVh;HOTn;P|`Ff4GfyJg}g
z!hj`1DX%q=k<0(gJ6(PCh>+8!un`;>h!_rfikDKTB&@JK)<uV>(;ZOgT4ghTxtR4+
zZl-GQauMHBQ2Qup?Pw`T)k>3V6}0nCcBAg;QUrdpELOshPmNnter<IwtinS!#Fcm(
zl#8=Wdog0Ftn45#0IW;Nd3l1hm!mtqLXhk5xxTAtc7w{^FyNkiOr)=h;CB4dG0Hk0
zh%fk|F9)}T|GBNtHS?0jR2#<S4&OY$we*Wf6uUDMtDNtnp7<wOEy=WnV!6iIhm?MJ
z)63au;g~|7{mY&H?px<{#1~VehHMj*<M*xUi@Jt_`fZZ>khag6Y=N0WX^UmAJt#&H
zpI%4SJ6P1Q4j7wpSvsT)je(*OA^r1UR#*gx?uvl>yS=yX0jj2s(?}F)t9YG$dc&Z@
zICDMG@fj}S$?X`Sug4yyKHIE|Y)0W0qlYrGFbd4l=|#_gsxoSX9ia;D+=#IL%IR87
zo0F8qItOc?nXQ)-bt+1YoNn-gK`=%#d%7gTCRK5qK6xRk&Me&9MXV&Bv;<Dx{&-j)
zk+^g|Mx+&WF~BjDZ_D#XL*)Xtnz=@Y`PEYTd5b2HJBt(rC?2{a<NNq*ML=j1wA|f*
z&dRO9y&V6jsHO;SUw6BXg^ZL=GmB{(|Lx8t^$p$g`<W4htN=Hp02LV2U{D}2BH~NJ
zYP(>t3{xFvk7}^r{$%P8HmJHnIMVvp(vD7fJ$f*|hx!ep2hTmzDhd&z-*^0<tP(W<
z|7O-9E~ZgNK0|)K*&J3`*a%c1*&o-;G5{9dC3#n_L%~GlY*&3AY$HV+W{T5p)V1`C
zM_0C>T~|FKUTuEh{K0YtF0(9)cw-6e!}-gPdG|jtS@A8%(8f&;acpFB?rVl8X-qR@
z`X3{-EEDdLi+Q<b;%>dYXyC59EO=H4C{w@o@`*#*MMr*4HKMV6edsGWa!t?3^a_*`
z0okaiCe6qV*Hk#FT6X#LjA-|;Xrh)D>;$>9QecC?H0BOfqnB$k81kkOQa0)W7OwDk
zQ??jgZM3}Q4wdap&D~i)UAF+{?54ek-*3aTu7+gllCKzfj(xxU?a3SQJJ?@R&V?Yo
zU8RuA=noZMHhq#dVr)yIC@t<Ok0NpifBt#0?@(^krJc3hIjj3Kc|^~u)6j=I48<KK
z8dp&yVU6NfTFnD@y^=cvtAezrP9+D|2p!zl^r4q@=;q8t3au3nAV#pCb(r1sk6OZ?
zG0-KN@a9LQ3g*swgRh^?@BQ^@SSVBknXogd3$ry`H)QaHM>|sc^Q8Q%TAQou3K-|)
zSj=xFhPs;|^=$koOh6Qg>~q>dVO|j#@O4NW72Z(KqIjPYYd_=G|B1DbRopgsFbLAd
zxe5apjrcy;tW|oesevZNKi+YIH9}-={I(p<-63lWq`UkL)ip|LdX*M_lAiePdZR8v
z%USb;HIaeXPrX-lVOrsTk+z_@U1_OOH%g$JJjML24B3&_&_?NEiZ+`dNd&k88R}u8
z31~Lls<ZHcVuCa3h{WDK+1fuZ)^qOKqh>J%E;wS=aL@Gd?8c(Swnmz$R@YG9gChnZ
zY8Pcf)+FMR7tC#f%%44rDQGt<6?t-v?x6k+M6Xaqan^r4WYroafEcBsB&(>3EeZHS
zQTB2i{<3PgkbHbJ7Ri5)z&4PWM3>MGXn*ITyVTgS)%8!6ZG5w}xB1gjuSrwhAytBi
zrpsh;xS{VtG(WImxmLvf9kSloGi?~23~Xf&l<At##dT(c+>t-lVNT<Xr=>yK4-929
z-~?WvpD)l(!H8eBiaqoMi^@Zir}!B5I8vH2&&>zFk|n|G#p;wIEk&55<W3SDdHDQ;
z>daD+4{nj^rst|?TJldb^=itVOAkQk=V0GYKYPs`&&^EZH&dm-7cYwDsu)m<xO}Zv
zQp#9zyO5Rh1PJQ1X4251y|}g?%%jkem|+=}Ct?pTAiubW_48wjfU1vj6dA;7@E+K2
zdq8T6rNsOgl+5`{a_pa99w*l^=kHBxW0j-`h3%CM79rw&%7>5G_D}gdx(Tu`tCFs%
zFDmPQt(Tcgx{P-0+OBd=)Y_xBgkczKGV4nP@+Km&Yxb)S|4C`&r%K*d8xd_=l)CU$
z8el8JxMFSLE_lRQ)DN!ZPCh2Lr<>=2I+Y=)Piv@gi?>4GbiFMsaxt97;Ct!(A;6eh
zbus~)9*bdmT;VvLqO@Z09l=C`ou1h9TcS*X<|`V@Ig1GTz9yu#(n*H9t~P;!Q2Psk
zrH2@)8A=MyKjZV#JB9AHAm@3NniLBVVkL8c8;S-T>Hge3XOn(`q+D+cL%KW*J*f#8
zIB}EoBvYig0K;b35;Ya}WR!4up)uFJKN1JyM8%2StjvdOKD>BwBX4cCRDI5O#ihgQ
z$l}X6#rpC(%PNY@+pwplsPE(3<@-u0yeu}xpuCxsUzuyP8D(#gsE;OE_k0a;injTH
zOPLfK3gqNE1LrI|^x}Rf&d2k}vnsS+SKU8^FrvQBsxWhh6@gWu5q~EAm8xDx1+6DV
z3auHh;z!dGWsQy;&dG@tT<UB<Vll$HFAfWtsx#=+2;xPWm^VJuO&HmM#Y5_06=W8x
z=<LJozg_4YDOnz2{>0~&zRNBMbv6&;$!|D3x5JB$&isxEs~VqB!h}a5oBvgJ+dt_z
ze|8%;mMFK+NWH>9L>(DRabdq1K+M-}%c&!VZLZU6nFk^;s^7~ZJKrzr;__52iT<tn
z@uVdO|H2L1-fHhGsnZ1NzzE9SEBMRRLXq{;fiey3YQxDRUiu=lVV<1AjIA3VtST6+
zuAT7dgg|n?Zl7MDJ-bDwgd~6kl;H3N+r6fi1|jBaH)WQ$XUSV+l8b}yyc?uEx|$7*
z`OE8|z#GXSBZ3bhSwGU&LnR9;X~C8>0>4{@$^E?sN8jJRO?|^oceFvn69gj;HapwF
zdNd+}wG#6(K+}zPsA2ql&q}7FTXlTHMkr0(JzCUk35?Gp@OSy8N5jlK3@1C5N+L{G
zlR%e;L`10Es@Tua*1i{2p<IVG%Z3?b8e<*fnMSs;g8+lY6tisCkV3l%rHM8R8qDcO
ze`GThlm4#j1Q2OqY=a|EQ8W3O3D6}tGSdA6C-n(-zS@2mOC+5B5G+m*H-gI_a54yS
z<3AO@A+M1!;he$58e=OUe*~`Ee(wIt!;CV;e^VQ2*XIGq;4%jyo*tT@BvmcJZd;LF
zO2IAV-n~xhXvQvxE!!@M=>Mh_cA*+4qc_MoL(1KxM<JI(Ve4NTE@~s{8B7lywgk3R
z5%)0{HB-=qq1}CKzuE0Sj~{%uIFw#`!-P{tuTf?hPMo10&)VElFQcybmCv>9op68E
z0T9{;G|}1c@0~Jaqo{pr8OqhDZNjhsy`DZi1d!2bT$Vl>mdzgI_nZh}TLyk^D$<%4
z?+Qf1BAJA-{@SfYKx&ZzM{2f~=K#<m9Q?Mj2nq#xd<5^3B1ZeopZ!^U*d4?k0dMIr
zOW|*tjCH2XMuec@vmlF&_w~7*9#n*x46pMjLO^vD3UZCeDBFU2R!W((R2cimh)*&D
z5{UXfR25oZ)@j?9yi@t8u)sdQt$fi6aTg_k0}J;hwv2<{Bw`gdp-XaaL89cKtI3*k
zDIn{6SI^?H1##FweN_eyVsoL~>&*o?ZFrKz+{(R3lqtzp@*euuRSD?jkOhlV_PL!y
z<`;MWV)*b7A}={X#djlRJw(P*SkjqeWfm3xVJKgYg%o8kJuHe(4|o!FH1p!~=ZJf`
znSSBA)-YAm+voD_yz0Znp5mXUl9P}h(z^TdYynhA(kx`>xZS$&`+zO9#f_xwe)xoP
zQ~sw22A#*YFFM)%9fKx)+!Bvi!_Dq8Afu)0P=BO?w%?UlWoR4^l`$9X%Y6J*c;4>n
zGG-zLj8Z}fWv8T~2~R)66mLFFXr0hKco^b_h39XiG~>q4aWIG=yGJ!)XDn6tCIHv^
z@Ztb_sh4GLTO>PbG?p#bov>PFd!|32#XjRRMezwm&=0-4A!g0Zn4)8~Efp1{$+&Vy
zvutk{+k2l#AB*1I=T8Or<L*6GEXiDKEaFzbZ`IDq2<rRriw-I$Hh8r=3%K^sTW;*D
zTLjEMI9rtxu92LMMeE!s%FBB`dk?e!pb=LOsFAd30t>sx@t5TZd?yVk?8Yq?aj9h4
ziw5_(*2s94vpa=fyCd$u%?BPT)-4tQr}R`6#hrpi>f#K!4a=?l_~JlHwz5E<r9QiB
zfsx(X6H?1SzLk_VX-zNr4$ZIW&Km*cYTQ_+%f8<K($-Habl^3M<wFvKs%)MX=vP8M
zNr1taKM}k>HEg18=j_w4{Jv~6Ks?9)4kulQqOsnoCR~FaI+P&mkN!bsK}7@2F)VO3
zZfDQ~kA1@L<_Nbgor193c77JdEnoea&uz8y3YITW*kMO*BZpEs_g!;Gj#|_7WU$=f
zEB|wNdC{wdH$;!s?y0CDz2xF-N?E2lcrVI3_ReWgonJ@`y_>jRu{!nid;6TrI|qwV
z?|rUMDWnOa;A}^|?{vZk#$_}#4RNNa;C1P|eACAn#ck@B3&8E{$qT!c93@5(BP13?
zkI$vdIdaj}ES)-Y>7G47*u~A1$7hV){E2t#dU!T>YA@i2&d?5NFs_zRL<IEbbRF<a
z3a+v{mV6iV)bSE7sMf{+gUdZNtTo)l@8SLI8fuRi!33ij+T;xEHI?6Xhg1`3+OQy+
zg&7^S-G9AbrGlxAWRKMaF8i8tOSNy~O_6Q5myj{ta^n|SBDOwS_LA#8*|@PGoR>;<
zQ{8d68}umN8${Dt#rQD^3I^;;&jY=l&CW_N8v9bcmXw0v_>f7OWBP(2u@zQKt^fno
zus)IG>N~Z}ZIDdGDfc{9$gc^N<i;7!W4fgHYt#S;`=K$*D2o~JCbP5kSixRP3(ub-
zq@3glKPeVD#^E+05QJqJ7kF@<B~u(1R3}&JDC;A>QZn6)u)?nKQ-rA(&c!Mf`AwD^
z-MOJ>bqX3W4iRr)MLUOYBJet61NcvhI4#Y3j^_J=ddb41Eh;y7;EAb5Imwf$REd`#
zaKT%O8QNpeDT7D++VoGCwdoj8<t?Z00o3m8;<PU%AD_4$CrsC>D_>U_2gu5G9}Q$w
zryWyWuLBsoID!l<woFFTGn6+0L|LTgCYebnUz(U^LvG4+Pjs|6$s%X0<Nf?ZtdjcL
z9y4M@l_FEiD?#*_0m6!X(i1^nMn=AFa}i5Wv~+vd+LWt+Z8aS-iZRpqpj*lH5tZdk
zs)IcPo#Y0i&9;JyAO)%ICvH`a2J9EEDA)yv#dc=iHOkD{T{rduQ{$g|xOYNM8=u`C
zxxmyL!V9T*+_+UYhY-=CDAk(W)^)fqC}g|Kk!kmqR31Sm_#ccNuo_el4H{U`#{iCa
z(%?_<#vRTR#g*QIv*N<3@dir%T_`gL$=FOx#b{}@Zmq3Pm|NIF<s(jEh-?&4#9W5e
z>BvE0G>a5JDkML8O*vjvqde%a#l^_HfG~uUYEJqp!!8Z*`!b;oy|CKuK5xiMKea=C
zE1VOlxw|Y!Xl36{e*9a<EFE{!&v;{?60aVQj<^O*<{fA;y``j#{jCHQWV5Q6Z{kZB
zq6L-@+|YEiR)Le*7d`ZWw=V8(#+XP_jjGjUdCRPPG31=-hjc~h!VHq+yp~BTYL??0
z1IJmw>q1b9U2T_-MuI;xCL&R?(in~e%YX?QQy5;FJ>sFA4)m1Y1jV|jacRbC<VB4O
zykpWli}b=@qjwLgd>d@XwOT9D5kdovQ2}M#IEOGH^#J~*X^t~p=#0QJPJ@<)+q#}X
zl%{JIf;;4(m%(bZ9^>z?V^)1+H%!kd2YmSBP_-(MvL+!S&$fnyyC>sz6oNEARoD00
zE@DI-<;oNuB=41TWj#XPA-{fD-!1<urojy$W4Uk653^yZFC)&Dh59+TOxfkalt{Zy
z%#t(@SgtG+<5|T*<<G*&U%N+M`Zbt-+#VrXJRqc@`LTQG9_k3m-Pc&ZT_hkBMY|K5
z)BsO|sF*5M$OJpH*_5_{8Zc^>co$%jos88e%&5j0^Kgto3A)QI!ZNDm@udCZhPFTW
zR_vTs%WvzENIL&iEsSnsCJ<V%jjfQ&hO0V;U96h<9Q9K&(+vy^@(@cc9gE#7z2{bR
zaf+ZlMgAIKer_2=SrQC&kL_{1+NJ4+??N?$4qP<%JvoU8s0nwP&O7a<#sIw@#;1Bb
ztb^+*Ypxa}`-pX2WD;#=eE}Q=p^1gCg!}GLQu?WXI`3sQ8?xGjukc{5-|B6^mGVO`
z>@|hw%bw`zH@z%y+%1S9+l-RHBz;e=#Gg|Re3ZuexuL^n4z0lK9BK9(hnOezt1hjc
zHv?Ohic+@RF&Yz55Th2IjQUrq@uP&5_u)rF{(-#OITXk6NcPtZ{m2%Qz_BQ(krkl-
zN^=IE60wo%VD|2k@O32<6eoV;B!YUGJpe<I-DuBcQL1v8ikFnauD}yDz^9u_y|tmY
zj0VOeemff^La_hicRXt#_Q$?Hv1`^o*7W5<$OVCe#Z>98)r>8bh)*&`_+@*Cfz=1N
zm*SxKJl;`RT_pFW@x!R}_GkYM>KG4TNWi3YJsK8Mgab^0<W{HUdyqt2W2N4dh=g9%
zK1*FB@eih)%OjQFg_nPudB=wQRopSsk0SXDyk&x@lwcNHc#91QxzTT5ob#>>$K`9R
zJEKynS&)0+8_Qv|POv)^zNfSFXVb<!jvxiX&bDoLteHBw3tiy&XMER%y*M(qX*IL~
zOse}sJ+sSSVH;S;gkE^gGZs{Y4j)4tRG#=evc5OKjG}dOeSICeJ2^@~z!BIP%jToJ
zS0M*L;7ve~9zf8sK(N+8Op2CVJz&}7VAg*0Z=Q#%<yF&R5FCh@OG%KI!A#7k&W$w!
zL8EOO#8C;T)Lf|Vu{(V8r<@gn?mxxtq(1_z`|;la2n&*(y2?2}c)9O3NLsbxeNo|A
z=MiopRzxK<kt5(pG}+z4m)MKk<V1E929{WC&)|T5;>nopR54=+1gv@t{F-z%GZcS7
zbfca3&PO2xR+H%)AfSmD!ZFIhiN-~A+xkVKc0co|%{cTpz%2-(gwKxkR%u)^+7SR+
zCkX5V1oUUkWxaf}sB|>rp5a}vKg`b{A}J3MEO}(^sCw#PM^}Nk7uS8<A5+48)S>jy
zw;y@fUxqW(eupqIs58f<kdyb-x$k3B%bzQ<9b%_Px_mN>TA|PoN#6dUvHI8KDa=SM
zc{(2Kx}b$f6gk;=7Eb_isd81@W@7-X#7t*9N7mP3tv&8tM(=Ethyzwm-sfeU;Q~#<
zbR()UFzWof*RJj?9(;e;YqgPA=Z%~HO^MN3gDHogO}_~={q|}st5kR-dT+Y-gPgh8
zUnfY*68$41PSwt76>itR&IK@WOy~_jC|mjS_*W3q8Q0@4<kl6ulz=KkL3-w3fO)Xu
z_m1W%ks>V4g}5RH-8zvNETE#EhvaA4e>zsQ12?|_Nv4I6eVm3>s++{Gq(9@ig&iB&
z#saWj2}fU2aT>$#2^TcV`9nr~g9jrXh5g}fx}(cdT?z;x*HEOjYcP>zq@1udEhu~T
zUXtWtD>--*!ke7S^ucGdx05z36ycx4fqs-{kbQ4J;0md;gn;wV3vve}eoe0`rhw=-
z1A3W_9kU{Qmt1No6W9fyDrNnVai$ZSD160#wfTe;+r8mEf_CNG7^J@fqY=?$tBoYv
zo$Rr5drvUQsPgGE3s<}^HkVLvEaS1|3m-))qT|QsknrD(teM+SA6pKX8XIHTz0j|G
zF(W633v5_l8&2WD8?1R3778f6_WB**<13b_(3o|W9TJ56D3JuuQHNam?T>B@eKMCT
z?u+%Pv;#zv3@b_%nrkUPxcdF}I}fCAA)6_1+fxOo)+P8|FOQPZ+%O0K#KcL8FkdDR
z@~hJM5#w@V?*vvLD?UQUcCuTgIO?RYI(o#wt%ozltP=`avvh{bVAI`OHX(=gK&!iO
z(A*g8l8f3Z>QK!M7FOGfoZLZ+v{;9TytHvjs~EqrQL&_KN<G%$8nW9dbhjZl;xd|g
zupb`5QH9<9<Q#jI<Y>>AwcA2xXi~`ezOCZVc}|VuLFy_)Q9A<<%|fu4c6LdA2T==|
z2GdW$Ej`mHy7p;a%wyk*3Nu~Gp<i{`BcZ=1_H(PrTsgpyi{Fjc3VcLzkj=eQ{58<J
zISXH8>sEHk>gFU1A+^nZV?l?8K>XS3IyMtgex6V~Mnfv>JT!Z6xrH%&Ou+KyEC`Sj
zMxzJlMVCp!Tx}iE4tRmW3Two1n**D8pf<<EyTRZl97)amHW&)CwHBgP{33QjK7YFm
zH?1~d5{bI`t?6L3xET7GvFLB(U|@e*`X?64&E%08Ex7>5kF-D{Yn2#x{CfpGlLF;b
zI^OoNl91*D-7@2vegZpxsKD=}<}7;9Hd~I=-wI<4Bny#yHxehe#~aYD-(7zIE<6dX
z{iUpH{K7m}eh0U+^V&UuW~X{E)w&j(8;26pPfM*C*vWGyf7BzxW9_S3N3}OL95(+3
zss+$!D-*_gBJOuNDWU^Wr@!&r8ZM%3AQ`UJ^hTl_eWa^c%;0FaB9Ms|pY_eh;L5sH
z4DnJY9j{3($oB%};5SOxI(D<8x<i(#mrG#3Bj-%gcrwd&8I!8s0?*<=hfg3)0sP7@
zqvPy3Wz>q0KN4i%TNUvuMq}fQs&;U?z?gDUFWvK-0-Gy_H}hy6douNBC{3WT$t#4$
zb_V+E15VzX<J1${Z9bY-hZs9aCT)#Voo?ZVcdkJm9=?lTuN6T%JaEb(e=gWrK49r2
zRhY6+;MR#qg5!A_*Z_9-aoQ6IYDmREYT9T7*KL4$D#V<4Jv;)`vc?nWKSv~`Z7`e$
zVJS0?{?Y5fHYgiplD5@M&|EuKJep;1RM>ve>Jb8#&wCvsJr7`?it8*$uQz=wi&_d|
zqmpGj)uss4BPVj~&&fLZ@bu29lKZkLJ!@x8>9|{pdcP)K7F#;%^A6Am3L4yV7t7Vv
zpKekjXB)f8|E_#}<SYC{vv1FA$d=X7g!jw{dB>OVAY6G2$zSKIsN33@mPX=_W2xMr
zTn_ei$}|iSMA(rxCDBaZ2**3fc-f^+XU}vzLx!^%t!35KqDESH$Q(K&xSQC94Cb)8
z&#*7sVx<1#U${CQ@hamslytJ;4uT7|BOIrns^k_-sDf^T<|~DoQHmkynx?06M|RRT
zoZ{Oty^QT#*}8VgppFue3R;*3np$esKaw1xUH_-iHh+kfVEyf<r8EpdI*7>>FL#JS
z=_VnEdmBqWvAqt%k6y>A5qZ*z50liTW+D-)JFOoz49Z#3tkQ=9$s9p*LoE{%fWc>x
zYed^^l?{s!hUD9Y0qduQ`$>VL4u9y*TNe$`p(X7<b!GEY`%CJ!`0u1H@r_AfiT7AD
z=&U4C)%Zyzdg-B_RJ38OLE>6O>j8k>Hb)$XZphuHal<<I%Zlp{0Zka*yrlpjgg0jB
zj`(sO+P6N>yx)V)zAl5%QeN>x9MB9QSm6}UOo6Gmhk9k!6voY#(?1qERg<mQEq#j>
zpV=$4su$4iX_yH4`B-+`o~8q{xykOJ!pK7yAV^dWYgUjrzby5kRY3&zox>1#3WaP4
z-d0o;E}i1}-0#gG>929AhJw_zF*9$w90Ux?%Ew8((TdgKGP2vPfjKVQc`4kDyDGaW
zpJI|)U3XGe>EctV0{oTL)y2y<4=Vh5{ECKWAJ<OM9uGaT(YSd)x%}{DK*mf;)LeVc
zQOT<n<~pol)hP@4fNtnb8QTGjI)`h!Xrt{IUi}iLY0ReDE0|tbsC>&VJ2;fn-^+$K
zZ>JFikIHmc#$VElXx9=RMMKsmE#AO8RvLx1w;n7<79Z!+!=T_M8w8_T9y8HK%V`tp
zXK^V9+xlv9m(O*#9O!wgr1cUh!7HJT+!9(29s{f$H8>g3+=(&Eo;@XCkdDJu-%Tqc
zMMX_ntp4cEUW{n&STM{P*!mg~gZ6lshooq`mbQq@ja4>o@pWv>D<ktfgBL!0%uDCN
zckj8V@TFjWAr`?riB=XsuBveJx30G3_q^Y8=<c<emsaZX>EnF5(=kd=f3Ugu;fio(
z1G_w^X-Q)r`@Wnf>z@s>&P0$psIA~3ptqrpE@-D}a$^w@9Rh$Jgm#Z;j{;d)+$;cZ
zuHtN4?{LhlQ*ik~pC@3>bf23o<q#z6SI;|Bhf{R50>(!L_<|)>B1709GWQazw|Y_v
zny}4x?mq|Vg~uqjsDuc1)CNh18PrM1^on{k(yU_(Y7oDH+Hzx%kBRF1;XB;eu=Wls
zyO2na!JTbdWn|T2gHMxWj+Jtg<HaEidC{vX+C!C3z)!?xFCt*)e_80jOy_<_wT>=h
zS8_yvNIJ%umjdiCIyOqcV44Z)gzZqLuWP03CDt4yjBPymaDcYiZzLELi#J|L@PBWO
zC~r3@X;#^b(AI-q0%6=KyX8s$Gzda~4n!*wEW}-Xirnymu{$X-<F$XV=U`GNWPw+d
zqlAUcf+2iG$9DQDg=#tSSo1}YjP)Y`Zs4OLk=0>;EvpNM6SUrz$cNeficw68ep1*i
z6XLTN6}5np-bhTRe!tel;F45ak&ZP>)TdYH(SxxKg-2)=xN1vPoUQIgFdFchLNaSq
z)g_2xs?JQBXP9n_@z-mgzS;%!w^d{M<9^O2hDl%T7#0md+KN=~_7e4ntX<(DJ+11F
z;M5XB#=CtBP!wCx6?4+-P9G$DJ^AiHitH^qUz{t@z8(F30jTS&EsJEkWU5-8t!L_I
z2B1@P5W;Wti*fzId%hx(3zEH^cA4u8mUJcGaS8r>v;38lMPy_7epW(*>bkGa@tE$P
z^S6?$)iM2qr-5~e55XuoIg@Eh;$ai%)wFJXYH8|d*cP59EPfM7Kd~a7D|O18TA@d~
zot%}DTzoG%0;cuM0sNrqKl?1F@}Nx6xtm7>jG?JbIGAAvZxHurhfVq73tgd@34E!{
zhY_R&Xs$1qb<cJu9b)pWAArd>&rS33bv>?g!Mwr9E}gu|W&Iv)#$*v!M+_cUg8gGk
zy+gwYN1`LEcWOpNXCaSR)VqZ6a~B{sQ-vBfjPWw(TY*3?iCuWIM@3?(o?>E6k9xsv
zV5qesQq6EhCjGa&F^1@Q)NrC{nF_L25uV)nT<V)D*+|@bbOM=vd2hb!x{rbHbVte+
z?8y=bL3Tm8m|-89uo(%VcTa#sGls{_auT9%3N+Z0a?#L<;WO^cV?rfs+f~uT+gD+l
zd!N1tL4?I{rWZ;&(JPCGH5+{Y`IZ3Pi7z6$m`h?$9_Qb_2@#3N?`TQgY?{oeH9n#g
zcWaXK+x}V1$b=f^8PzDClT`ELZ10;qj+D9zaRUJ_fb>VU;M&EMiL#%ai4%}U#lvSl
zSj{2MbO9TK8{9cU(sZQQgG<ZROQA_<>n(&=Re6?Y3l5|6-M(Y1Sf~sl$1iWQJWBZ=
zX&mur<!2_0cT#MYVPfyH){A8k!ebX_?+Un>f`&81(HL16Eo9Gd^ajF};D=IAbTxkp
zT+{-&(FlK>cZ*FYo*BS;muE~7^)bUf5fNp@|BgvsQxS~~smynMR5gtL5bSsP@mM*A
z5jOsV9dM{9DzbFgh^LuO@HuBy(i8kg^1{PP6l(FWv+=#_f@=O@m6)aWK8wIxYE|k7
z_y%XTU1M`=ex?9ysz{F4Q;mmT+Ya{o4*<hlRA~nT+n2^1r?XAGroH2h@9&0rv`uXf
zGh3b=H^7$?`1kG2^k)3TbHD^n;T`@N>lyF$b&tE{vh8WXTHd0iJTSEaiB{q#bJ6)(
z6m&F88KQ$KhL-pO`mNIPH#It8Ynx|gkx?LsgN~$$ltF`@WCXCyh&6AX)1b7+AN~1v
zKXtY#EKl|mhfA9_IhvsB{LpblDSLk@J35eFd=%P925$JAUV=n~;~h?hins4D$a1*e
zCQ=g&@MomUk%=>=aM%mZ;YCq9PN%PITF2W~<V(h$G(W;mKaq|f)skoiLOus1mdxK&
zv4>g@9HRTu*8_;GRl8i(_0Y)pJ>&4Rv1?siEx^1Mb?%fz<Z?c$I8Rk&$`fHXG<p-i
zR=QY0Fp;x~O~@Rmk5jU|C{QK{EdStd4??1nu`X1L7f+lEg9pTHdS+5fU3(0PkX<8x
z6(rK-?$Q>~uyj|bIHce#+FUI-MSo5dP0|`B^O;JnHl3T^c2?3Dym2lTo`ZW)sExOU
z0E+{4Z^F$>k{oj2OBlXBC2$QL$o<M9ul1Ggja595C7ufJTm?4fKGPBo-u7C%)+k|q
z+k*7NM-VpX;tQC8nRjH)H|ZAL>nqx9U3MqZQW9hLjH>?OV-Vsb!I|~K9}YX{P+dbj
zROdri-HMZ9#x@BAD9Jl?`)PLfx$|>@P1i-2nqq`ki>Ouy2!rgzvw2Bjn{c10*y!4$
zs0yZvDF~TbG2J$n!#71j4cJJ#Di4_Z9Wqy>y=7cHD)N?In)<USS5NS&R@!SY;jg2h
z?4*;pW~$k2t((p4F3`Zl4$z@O^9(3M0;BksooNZ^x4FPi_0>(FIrz2RDUQVzHg$yY
zNt>c2PMyRK33)!V+m`*l+z(H|#X2tL463L-5=7b)MREMup0y&7>>j9NaIn-5?$2{R
zr}src&3)Le$ed;fO&*~ci_S2_*;ox^z@vaER!53%mnv6-ng-&um)UNXe!e&}66V}j
zWS~ha7jnLibRNoEEw40fXIFcdhl*;$2ad%xi(q0TNC`f01E<hPOrR;!tYN}uF+RZf
zTbD?+XpP(h>?`s4Ee_gA!H+A*0o723y2A?rO1`}fLKjdmnn}d_x4#EK;Mj2aAQv~<
zCYJUG5*ZQH0&JgmZG<2nU5Dd}9g4)rtvbs4nlp1f%ND*$Z_j*!-c7J(3hks79=dK~
zfNzLhJVfE(cb1J_Zi~e^r?{exs<GeM9)%c&t$V-T%o2n}e#xT6{WMyNGBwv~5#jhz
ziIkbm42inTFlnMYW1@#90p@f1Zg8X{&61&tJdc&|b5J=*Q6g<E${bZ9Ug6FXHdLXf
zpJO)IG%@hBK?-rbdd8e4CYJ!B^GmRox#9qklr+)I)5n0Gp<Wj=Z=jWWKFUm#q5*)U
zc$PPf@+W&a+vX4#(O9X1t+-OV1@p&A_NG~+-0xkMJR#Y5B=Ev<t#6eeipFmxt|AB>
ziSZ!66OB3KGz!&-nT{9-1Wbt48l<R+H-Y$d%<0CooLSv@TG%XSMElLJ0?hmSFo&7s
z%VR={)Y~61bIKvzWx@UKu82s?PgNr!4#aJrw@GU8_2OkWG+k!YaDvXCdu?TfJe`$+
z@+)ghJtC*50laqX*LBTco)VtM-N)%JFdh-2&QVuyGCZ$4Q%vH2-EHuRC`GPW^1~TV
z#Z-#xIN}%DRfeFMLNNn{l=K<%3lQJZZ^vhm4JLxPl~sZbXKHIv$5|h>a2bku$D19j
zJP2opQ>XMI?pWUm&{y$rB9z(^y7gG3@yl|C4NzQLl)Kt)mKhjw8HB+XgX>v9y*DGd
zv?(K};I}3S!(Kvy+ZszW2F<)Cd2mdvBTvsScg?$>)Anh`_jNEE3rf{{rt%H+@v>ct
zA+}Q8l<0{-5mW+HDTl}!zul#5A?DuAG$UkYw7#PgpZ0{iA&8_2z-Q9f1&%UU3!E89
zmYjp341QK@(h1Yi!?hus-H$k0S)ww3iSj_%vT;Em%_#2{!O}><#;OCUH7)2=Q{_U6
z1nxy-LDf8v6UcLxaR)Pnv47O!B24sv>3kPE40kdsjJ6xxY$CiXhlbZ)Aj!v(h%EOg
z1imR}TcF^$nf2HoI(nrD&y(mc3G3-biGXe!y8nW=6?*3fvlM9TZRC#ljr5zg5uTp?
z;!fbNE2}`@cP^(*ToGg-C1UCA{y+nuCVnkZfWl(!k-}^gIZ7&|NKZChiOs6CWHG5=
zl_sR|4^UaCzjYLB5P1{uN>OhnhMfhz{F(>CRv@kq22++gyJp)ku;?N-SpMrtEVorE
zeUh<!CxR>UzAn{+!s^mEzwjBg6J$~P#|&?%|59Rzw!(uX7V(wejgJ2-ltodzVje&%
zeAokz$K}&%XZ54cl(^XqgeFNcZh41cf6Y>F`c*TC1+_HpCy4%Vi{tg2U8Q-CuEP<_
zwvD@Cd$;D8(#!4bQwK{_fpA7?Dkn%wMZJM?`}cV~M<86rdlylKH1tO|sTm7ZRJ-&^
zbra#0wo4r2SiD|!@Ua>K!igM6K&*PMCwy|Y(3QC(O{qMA;99;YoDo{-$$Sf+Ch1ZB
zqdhTPXro`GMRUB9Zp7Sa1GNyM5Ic0oGAI&+AqoZXP2s3>@lFDa@aTK^M-x%Z5CwPI
zVRlS|F_meX5fSrS1e2g?^t=Tfu)=0m>cX9iCqHziuoAwv?`6emv0Yi2boQv;?YRJV
z{DqQ)H>e*Y1%-;|CVRT$s=dCFWad=Eif45g&XlMpsspEAa$3r|8Q~|C`g|A=R6uq7
z-p{<7#?a-TKM{X;*<8UMF30=fUTcjkFu{=Sebg@pkYGbF2yl)n>4FFNOEdYNC?~*k
z`jtp?svtDe6w%7|at7ka?(NuBlnS;{lthJN<=wY){E{siJi_>*z27!uj&wz3?lzdY
zN;yVems$B$p0=5BAIKOMX?K>)pJ;!v>rK}bMhKr;TQJdQ;@lx=5D`cc5g`QKLFtCf
z**LR}7Qrit=N2{ilBJ-y*Z{vlHqsZJ?JTc1jWbE$TORXCd>Yh8n@PJL5o@JP?Yb$a
zu%sxBwCf}Ns<$;e!HuFvrt?$R4v>WdcyrXExHTNs7BlwD!Mi5j10}ywhf4wtrJOzv
zThYsCGIl9-^+_N3;qvn-J**Z82qYx3eO1?k+-V$v6@q%cqkufV?kh<T#wA-Vew+1l
zLSECj_bE7mubu)kcJAmV!I-Bp;E$=nn184tZI}Eki}oHLV)C+euf})3KQv%H=ybL-
zOq!mm*whVN;fhDUrnl3w(oEN?iZb;TCv;M-edhamT?1;1Obe~EP$gfE4q6HnFPT?f
z)^Xz%?qzxKv~u^3;3w1W@*60&RWT~gNKX_lBVH`aE=JwktQ*D<jV)0e5AordM(@q@
zZpKQRt@^g^B~J}Rjuk5X0m#7oxg+;}^2WuJsj{z}IJuV63FW0E!zfHJDZw=_c3KEL
zyit3Dwqel5OrRP*8ly=EyH?eT?AKlYCr~ZyOj`X?Q{KQ^mKk|eT%{YY-_q2gBEjy9
zV|yUY%NmEu(EHrF-Y-t~JDqqRY}sW(uR)Qo+440mLdVRX6P)^*D(>@{0>Q%>)krq-
z$j=x$e$|Bqn&oCuUXC)zy@U|2F#12t@T?lx@beHM6%yja7Ubg-(0-$gx)I>d@#47~
zT23z=P<)C@5f|JC08x&!)?0Ai1hS?q6eEh{woBZ4N7uK*PaA6Iv!5CGE}Lm?iB=Dw
zZQbjB;LQCD(>6^Jqc1<hEjO;q!Inm4dS|Y<3~vGXJqdDf(U!q=6>q0z!^j~lK*>P*
z?ZM0=M(JJIA5p7eV3lb6^G4zB>7zB<?T+pGiSqX+k9RM7G~=B&|3(lIVVBG<H|CV!
z5f<ofLW#ROyfX#P5jgqNvK~<Hyms(w-GmV`rS{Nk>BJDJrjfh{$GMSrxIV)=<piT-
z5?WNT2?K-fR|?w0R@Bj-CNZz+R8vV@oI68>a#p&)DsoZMici+UEBf#YtJ|se3j)Q?
zG4$X5pbaAdP7ZutR~d9SAENRPh`Bz&%9AiZY02P>5+kLrr08yi3Y6ZSfag-4kKe2p
zFn*WPXEDaC@McPqN%4fnmB;)6T!svm%*3u?)8#;dtCFxv<{z3I^b~X)iq-ST;RFrY
zGT@KaTEM@YJPR?H6e+OTk3$OAKw*4W7T>`V*m<J|3OWw&s_1@Ngy<$DETDpjq#?YT
zA}ihy@L=Mv(x&(H6zN0^$65l@sN6K`QHj+5shEysB3HbK04nP|A*CoU45se|_JQNW
z-x;War1Qp2mb+V};o#44o?*;@b4~ua#gwz;s=nMcHtlYiFvczlgYwo9>Y!;aA%Kk0
zB6JgL0j1t(T@$cVCfm$%*52G}@m&`B#k5G`IOucko$v6-7aP8Jr8->GbS{_PDr}3&
zUfyfcO@6{iDYetl9|R2%ll<PFPEOWs-md~IWf6dL!b>9nB{<#{o7HwT!J?ktwQ++!
z2Xs;-qJ$cPWl6ou?tU#ueE>vj&@jX}F&oj7T6iETg+z@7TjeY$?KMyhwod7sitOn~
z$#v7^8xM}d-1JMD`erea?yb{GznXkzp>Oz>)GKYxY8W}pBpZd2Z{7=4g=q&E<)Bl7
z8n`2#U-4)QKaa&*^>RYS`yvf=xV#}cE&8G~lgTU4eU%*_gkFg}<uGiSEp<Rv0>Jtw
z_0x8zjRzU>1b9C^0|^+UIvSmk^ij_gz!zYYL=&yMY?qvRulxLIUtk&WG=1BZv#sl7
z`++Vn4+O(s#oOv;JyAZ^&&EE86BPr}XA~M_YI$sJ!o5s%2gJ>EK2lyBC)p~rNWEGY
zB~ojzcFP<tlkrl)vl}n$UG4gjuX(Xg5+#g(b;?V`eIfdz2w2qKUy}ba=PzZ~AXdne
zrKN~b9lu_u)bJt{1{Tc!<7$IAXN9|8Q?~YXvg|Wapy|y#C^iNrN8~>wYBd@CCoCLk
zDF_zHWl+ZmRudcY77Kw}&Q16T^Z|a19h06cYE4-qe7iMd2q$u|pgh~2^9$TYRhA1T
zW#u;r9g??TIE9;K7<5Lc;}my<b2c;qsRleNW~c}y6p-2^^rR0eRwS$(W{5s{=$3iL
zhfkg?2SD0YZvaDWzYI9dw%b4ZnAO<|)py&$Nh|{rF3z$8Gl#E%tK<>ON79S2bISz}
z1CRb?%-vFwAmKBmK@?l4F;5E!J*Em@u)Wy|U*k87h?^RJ;3NIzV5h#uDU=hXtiz!Q
zII;@}gS&bS`~~+zdZCqecl4K^EQcKl=P;&&lM++w&|hf~K=eVE&&MYGaHw8Rjy(`-
zk<d4dF$dywzikKbm{n&>G+9>Kdu25K`VJ&IoeOq7_(cd_c^uSD{B{9mZ=6`3&dW%s
z$LyuAj;*2HRSrj$L=>jt6lt_9wSH`qc!!(a)S=>^YJmGY-%v~xS6#@iV#`+KGWm>~
z)DUzWf;KwFcmR)K&1>whyK+kF>RE+`Qg~ae7x<@$d3A-~oR=UyKn}*x=1!?W54_fc
ztOmb(Z2&KC+1#1|6<QgIW;pbNAc$1;Y*3l9%&gF6K3Wv|>GSZV_?JKeA<Kt2%Jl}Q
zUT&Gl@|-`clfo~~b)lcHr|B^wrEKWAAcm(AZjx22{ol6Vx@gP<-ks_RnoUDBzJqxI
zk!1H7dqA2Y(s&0aCZHSONf<AW6^=fs&wh#0zwn@k;GuQaUQAn87#>=S(Km67kF6)k
zw`#g>Jlh7ef{@|KT@CD)vS|ctQa|W%fPj21;pH%zReP_4xcZNFR@)UlBY+Rma4HPQ
z9=4{aMSbuEM?^-Z2oc2+K{P$p=;Ndv<*2IAfRPe>jD1fv`;DQ9Zp%cvB=?<B7h6=5
zY!IO`K#-bn!ezB-fh~=Vb7Yx<Bzm(}>_v6{R{%l~z3-1<#Aq<=bO88nF<#pbumEYV
zmKRMnc@VTI75slu4mvR8IXOUI3H7aJ)*m_*hwuZ_=(7Jv=_ZK1j<f6Q7(?yG2~<uc
zKD=x)^KS}7$qO^&Llq!~_s?zj8DeZjyC(Rp*0XIIFcJdN#=i>vXlgKGL$Qa!W}iS`
zo;&LJf1appKZJ*$fARkj|5X0JuNmOV$NM|_k$7+?_NUZ%b0uh&2>Ls8gpY$}^3rX~
z3bQK?&wj4we*gA_d-RS3Qb%AynWImSA&DPa81VP~(Dw;$!)~X2-ZE%<^KTcWksVPd
z>LJm<D?SpHYx$51(<!fRjCczT*IQ!gJfQsVmF(2uvJGyb<wR9C?*Ar;vusD+eYNDd
ztjr*G<9gXPR~XMI{+||meVfLh^lPveIqio*gtk=KF;xMk?Bp&b1l3o0K<8{<={_Z<
zD-b`Ei$U_sk%W1oz)%2${VXE;?founKsX_!J%%k#Gi5`YU`O6y%5E9BknC`C64aE1
ziZGwZpT~<FOVz6`Pk6qklNCduLP5Z7MvOT+AWW$Ec1NpC01F)KaML*<d+*#{o)>pb
z)UlNU1ju;tdH8lU_*d&g0w@bAA(!4@JWPJ2+M{V@ZUn9KR-9J=%5XO5117zMu4&+!
zt76GoPKHA_;()9O0{1GSdE`FW_+V^v9$<vg<N#i@v4#%|Nr7|_wvO>$pUqK!21WM+
z^4jP~i*`-%Tdii=G*4wQXPET1<#;ewlBp`^R$c-VoY0RnKg5j`{=s4k_dMTOug5t-
zgKb?5m95_c;@1b-+e~;3q6l`Wba3A=gBmm`bqgKKN29-1Y<&us(Dndthov#C@{Wz~
z;!32eomqGYPIE#$(EkztRQ|ub=bS!5wWCWNdZTx92|2p0e1nVL4sMmFf+bz15bp1}
zEZO|8>ug5l>C-or=v*fjeal4{zN2C-**C>*wVP<sfRGlJHTYNSLj$P*3lw2Lkw1?e
zrI8&`C+Z>5z-rC65mmbz<&E{XNJ`@i*wRsy-j!tcUXSAMQl#$^Td$EM_h^uR-MLGb
z!5B;YfV8o%!oOM<pIh=6u{iu2r?*9BhQde^CcuA&p2X0HrZX24K;Gg`(KT1e{r(@@
z%*y}Ve4cz`aSPLlGzQnX`2M0CbYaMHa)6xiF!BH%N(n~QU*1?<m@kyi`H8ap>K)ZD
zkqi8Z;v<ggvbncfxWy$LY(={!_^sBnZ5l8V0@B963jJtboeK!{Nz4_!<V0uiBw5Ix
z$Bxp-j;NFM5a??u*|rCIj)wH!8dq4*z&KDMr2q7>j%Ig+Gh5D-L9Yx7kR$cB7FOC^
zr7&C43NCRTqA9_LcULrAnE5=N=xUpKL&N9p4qu0bceN3>&JN`x*OEL@3k#2qfo05s
zES4Ak_!zx5P`q1dq5s$*?#7CjKdsFzf|ljnC{>M+Lk_ZX^dI$X0FoS~7M5ivEB?$`
zcL?o(v&a_aBW_tzjYGZH7g<Zfam0B%_Spms!6*)P)ag`VyD*I40(x!%ZJk2#1#JOS
z?9rL&&^XO~Jh?N<QZUnpt^`^@3NfX@0D#VrMEVO)IaaRv1n~y2I&R*QVR@{B3vh)?
z1CkaJQ`&zF^dQl`siHqnfrA87C_M@(GHcTJ9D!$-V~1WLU9;D7AwtMYkay6}d|{sm
z$YzTn?-&l<XIbKTE!|dE*<C!D)6_51aEA(#_>9YG&Zmm?qHxFktAEBA3b^-)2{a1+
zs|(_Q;w(X0oM*0^l>$D_KeSkT&1Ow4dD+|hu$cUn4(jKkB)%gHvd=d6hoM}#Huu%(
z%y^Z9Rv%sxQQ`Wgi2Xta3-Ni@rs?e8Hw|Y$2FTrVwrG3>yx+X4a3NSKmn_DV_v?Q{
zYg6tw+F#vs59-W*T1*mfcj@75?Xp6YRy@5&OnJ3OGV(l($?c+Zs*fz(Qtsk^1hh$7
zopHk2$`3-89g`(-NKjrwan4)O099$BqB9INmfnw(s<KkDCzGEXcxC9;@bc4WoaP=f
zuN&U1>qK;PN+#AD2E_MsQrV~?LW@yuMx><j*spjkcs!Y(XB?k?jLfU?g0(sj3JV@f
z))uCbwnYBb=UTG_JCXtxxIudWuYKnZc)jGl0_mE|c;O9AdO)Rd^KbLvMqD}bKN!w!
zIB+i%HSPq5we+`>D?|k!#+~g!UpO+@!vq>xraRr;*ky4vI-~r-{`#DQb0p}aG84sm
z(Dv%Ad>l9O>yh8MCrs2KhKSI=!Hz=w8+8SP#EHIbWiH98@CIrUE)DgBbh>TW&<$5;
z6#28YOK3`Lf;;Ds*kPw~+(D%Dk1;dXUfUOyAX^Reu)-?zh8XA#_gixU9zPrs`!eCx
zAt};CuUi|_hn-6nay>xW-!RSPw)S52XIdkpqEWDVXtT)4fSVVizzCpO940?|IFROq
z_i8A0u(Aj74POQg7HUa8NMWx!(VioCpX{Yb$*w9*46lrQ32n_z41I(A&u;yr_?|+^
zMSbKZGiHmq^i1EZ3eItoeCx{YIjdbhJ?01Hrprh&n2w`b72r1>bDK>*Y5Vr}>oJ7k
z!*C7;xvjf(v%w!7&g#>K3E*fRCw5ERM#iWBEy>9kg1P!Hh1rvKmj@X3x3m=nAZT-v
z6z|Ql`asGrnI83eDm!|H3rnhdlZYZjQ0qRiSe($p0tuMc4$VYZCul5w0Ww`Lk{$5E
z{k&9z8f}?Kf5-8yzEop4b^fqb9)XeuSS{Sa@_3xlx{;aWE&)N*5iZmoiM%HOzEAyq
zcDl@Q#ZOrnSrcJ@<@nDVuVCp&cO|PfUZS_2%1~MLpf*vdi27xGeN!v7;y$V>L^N3B
z{-KKtWQ9eWIjJ>9U;*z}Y%0^=!nU`Vv2-qE`fWN}tbYU7s#oCu$n!Dk^dge=+n1a)
zLLN4)=`Y57)@KHMu#{;Ofh&_kxpf^_A4r${VZ$K{9b7p5`0a?H?{<bG38}rq<Bxb7
zjVps$p#Sj{w(y&&OIiTN9%mrx^p5UDQ%<fr)YKfz?eS%i5e{q$I+xB$!3oBi{6QR%
zE^Q`_3D8I_0;v<NUzKk47u$0BoWLAP84%A|rdOzKrzeY;n>&lul_8^+Op%X^rFne~
zo^N0$A)e=Z_CfiHc4lq;68;%2puc{Q3N8;^dYCPXp!_28%{`PIus$MUYzYI21?B~O
zV_ibAb`5aZJX-*~72Oa8$Vnsd*&$2_!=d6|+(EyU+0$y2$26zTY&`fr?f!$~EDlY0
zh<U_ok+HPk<jH*wd*-F7K%^tBVoU77X9Gd>4F~MEpJY5P02AyCPSWfJ!2NtqC~f6B
zcNbz|a7=neBRga&n1Y`vfOg*WF4SiII!#^kwT{)q(-f~}U)D9z%D_D>Ay#ZmjvK?3
z6PxOj<eRIClairiIT-e*<Gt%+$3BD<)XfuZj48#l&Lx-oUW=(KrHP8yw%Hu3k2_J6
zSfc2e6A7yhsqhUS7STBh6Dq*%IT5mu2o?T7b?&RygT`Qzz42#E_orM9j@!&Iptu=+
zTK<}F1mE#Ibs9-wD3o`pfQP|XGH^#>ed1Rddsvo}a1(WI{%3YZEYPNpis5B!x3$qV
z^3<Mf(qJ=3><N0up1qP9&%YR1Y_K9~1hYWfrpLy9sL2UZSk#Gw1HMf1B$9xL7?5I{
zFv5nG^Nl*?$wM~ZK+Ei?;5~>Z!e~M=Z+Ke5>uK02Gf+eX6r!DgGR=iBf=*=n><Vou
zc59O)L9OKw+WE2YNIW}%U%e^}D_y>7y<%mwY!Qhv4XJoj&q2+pVYvhqCXLD<;7>^4
zEi5t}@=jJrQnS5Wa5uHtEE#|F;b)HenVA<AFbM0meV`DyH9#nGwA415MX-t^G`tfK
z3DgY!j@dk+IXQWNL!C9Xc7)^!ffJ%hV)i>Vwl`FKf(e$eDZ8sjgiG;Ss+&bcSioJ_
zh{qDKw9~AUPA#_?41GEV1(03)D}Vq7<?UGr0000Vz|H$qLP=5&@Fy8d$F1|Ev@qo9
zRja(Uno>eJuY9Q-AV05czb@zk_EQ7piOm2sSb?1Gw@@oe#A!6p%Q$g_pmARGbtpCv
z7YAWnQXpLV9bwf&A|KrlQk-+FVRsdcjXt=<)+k$;xzDxM1>o*5Vg-c6LWSqcZ>%C>
zBD(>z_mE3MNdFi*Zr9i9nArscS&pV%#uf$fL--AMy;IX1w!)KvIOkt%x)LB&D6GHb
zhd39$l7XPPsb5~hKQLl;Km^ehjXye7(i2~f<UNi)SLMU{*ocIUh-D=spJ)ZI8U1BM
z#buTk!aEFw;kI`ir;9z7uc4W)g}QK2Q0d-R2nUo&6;FAA1cR~jMb2dACf#(KLGk$M
zbGqNgmY@It#>0<kW5ECkbw+YEXzO4v_hjfQ;TSOo6G3kSbe{*512vD5VLo=3nkN%0
z$qC+b0#c21oUKsTbB1V$N>4<m<ab<R%F=c*?ygldI9O(Eer(0K9mPIA9XaFAQTxNT
zE<j}u*EggLA`M@S$vS$xZXnp-Ro>Brrf9Z&n}pxU);4-TiF_>#xzZlHQjE-V$R#0Z
zlk?Y(s!YAfvf3W)KN=LnuA5NME-<w!ovKbY=%&id{yNUmTwU$sT6kYh9n_Vr2OC7Z
zD-bhrEQ;a0iS1u}(Fmue5=U1%HVd|rSpG2PiJjBZhH|~=th32Q`2-@ut|)MCLF4Q}
zIVKR4xJh?$kUacso=_{j_+H#<QAZL;#T%^_W8<uO^3#HyqlzAwnd&J7sGDe)dF!)q
zFL@>|&T@W~E5&&f6-^4}@L;vc&d^qVmR)zz15Ker)m2WS4pT5=o62q##MP?Qo)$L-
zD4?COH<47iHPgrD$RvU-#ubxHN>-qYTFJ_zdhot&kQ<fkpeVSTfnC}T(z=yj)|%P@
z&9hE1_9iI|@&3&Qa(UG*{G}kdmD&a<iY1AnP$%JNSoJNyw|u#|oF-s&x#>7~P{<Lg
zLTv)14z)6P_<9X?D25u6hq0(Bo@n%2$|h%_+ss=GOyfLK>ebGR&gxMv6bu+KT-Ogn
z9uk)Z*moaIlKUpkdx)vT`Us$V^{7WF^}hA%=chR>t!KL=t=+D6?EuUaJTxd`GKu!3
z@PNzop$BsQx~&a=A6vg!PX`^pU@s6h;=-Cn)An9Fv)aXwAfbwb_u-(o;Nttv@O43N
z5w(4DBSmw&xlLs3JH=4Q_EEMpCL4TbR?h7(QZ2S(&Yn|lS4DBVx4Bb>M8z);g!km<
zHd_nvOYQaKr`*h>JmcyF^i<-XdA|i-ymbyneykqYPiJQzq%>tA;CAgzPZiY+P&2<b
zA(J6+KaB<TCF9^Oic^ID#p!H``0AJq*vr=at|mQkMjdla!_(imd3=2$uo7@^PD5Nr
z4%s~p6fQf3;NWF@XE8OwtfLPKz|2{S_=v+8N~<_Do*f&s+Tner6c@Bx8<0OzPhc65
n?D?bqB+A!~p9R~)lsZ&BDRD7XUDQYb04t$<T>Iz%000003V-=_

literal 0
HcmV?d00001


From 8d2482125859e4cff8aca5df9ba0b64225863429 Mon Sep 17 00:00:00 2001
From: Chris Alexiuk <calexiuk@nvidia.com>
Date: Tue, 28 Oct 2025 18:31:49 -0400
Subject: [PATCH 02/12] Nano V2 VLM Blog

Signed-off-by: Chris Alexiuk <calexiuk@nvidia.com>
---
 ...0-28-run-multimodal-reasoning-agents-nvidia-nemotron.md | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/_posts/2025-10-28-run-multimodal-reasoning-agents-nvidia-nemotron.md b/_posts/2025-10-28-run-multimodal-reasoning-agents-nvidia-nemotron.md
index 3718b34..3ae049b 100644
--- a/_posts/2025-10-28-run-multimodal-reasoning-agents-nvidia-nemotron.md
+++ b/_posts/2025-10-28-run-multimodal-reasoning-agents-nvidia-nemotron.md
@@ -128,8 +128,11 @@ To summarize, Nemotron Nano 2 VL helps build scalable, cost-efficient agentic AI
 
 Ready to build enterprise-ready agents? 
 
-* Download Nemotron Nano 2 VL model weights from Hugging Face \- [BF16](https://huggingface.co/nvidia/Nemotron-Nano-12B-v2-VL-BF16), [FP8](https://huggingface.co/nvidia/Nemotron-Nano-12B-v2-VL-FP8), [FP4-QAD](https://huggingface.co/nvidia/Nemotron-Nano-12B-v2-VL-FP4-QAD)  
-* Run with vLLM for inference using this Jupyter Notebook
+-   Download model weights from Hugging Face - [BF16](https://huggingface.co/nvidia/NVIDIA-Nemotron-Nano-12B-v2-VL-BF16), [FP8](https://huggingface.co/nvidia/NVIDIA-Nemotron-Nano-12B-v2-VL-FP8), [FP4-QAD](https://huggingface.co/nvidia/NVIDIA-Nemotron-Nano-12B-v2-VL-NVFP4-QAD)
+-   Run with vLLM for inference with [this notebook](https://github.com/NVIDIA-NeMo/Nemotron/blob/main/usage-cookbook/Nemotron-Nano2-VL/vllm_cookbook.ipynb)
+-   [Technical report](https://research.nvidia.com/labs/adlr/files/NVIDIA-Nemotron-Nano-V2-VL-report.pdf) to build custom, optimized models with Nemotron techniques.
+-   [Training dataset](https://huggingface.co/datasets/nvidia/Nemotron-VLM-Dataset-v2) is hosted on Hugging Face. Learn more [here](https://huggingface.co/blog/nvidia/nemotron-vlm-dataset-v2).
+
 
 [*Share your ideas*](http://nemotron.ideas.nvidia.com/?ncid=so-othe-692335) *and vote on what matters to help shape the future of Nemotron.* 
 

From e0ef49106f2fc677aa547817b770aa97818f66fe Mon Sep 17 00:00:00 2001
From: Roger Wang <hey@rogerw.io>
Date: Wed, 29 Oct 2025 19:05:52 -0700
Subject: [PATCH 03/12] update

Signed-off-by: Roger Wang <hey@rogerw.io>
---
 ...modal-reasoning-agents-nvidia-nemotron.md} | 104 +++++++-----------
 1 file changed, 38 insertions(+), 66 deletions(-)
 rename _posts/{2025-10-28-run-multimodal-reasoning-agents-nvidia-nemotron.md => 2025-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md} (74%)

diff --git a/_posts/2025-10-28-run-multimodal-reasoning-agents-nvidia-nemotron.md b/_posts/2025-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md
similarity index 74%
rename from _posts/2025-10-28-run-multimodal-reasoning-agents-nvidia-nemotron.md
rename to _posts/2025-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md
index 3ae049b..db9fe60 100644
--- a/_posts/2025-10-28-run-multimodal-reasoning-agents-nvidia-nemotron.md
+++ b/_posts/2025-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md
@@ -4,7 +4,6 @@ title: "Run Multimodal Reasoning Agents with NVIDIA Nemotron on vLLM"
 author: "NVIDIA Nemotron Team"   
 ---
 
-
 # Run Multimodal Reasoning Agents with NVIDIA Nemotron on vLLM
 
 We are excited to release [NVIDIA Nemotron Nano 2 VL](https://huggingface.co/nvidia/Nemotron-Nano-12B-v2-VL-BF16), supported by vLLM. This open vision language model ([VLM](https://www.nvidia.com/en-us/glossary/vision-language-models/)) is built for video understanding and document intelligence.
@@ -13,6 +12,33 @@ Nemotron Nano 2 VL uses a hybrid Transformer–Mamba design and delivers higher
 
 In this blog post, we’ll explore how Nemotron Nano 2 VL advances video understanding and document intelligence, showcase real-world use cases and benchmark results, and guide you through getting started with vLLM for inference to unlock high-efficiency multimodal AI at scale.
 
+## Leading multimodal model for efficient video understanding and document intelligence 
+
+NVIDIA Nemotron Nano 2 VL brings both video understanding and document intelligence capabilities together in a single, highly efficient model. Built on the hybrid Transformer–Mamba architecture, it combines the reasoning strength of Transformer models with the compute efficiency of Mamba, achieving high throughput and low latency, allowing it to process multi-image inputs faster. 
+
+Trained on NVIDIA-curated, high-quality multimodal data, [Nemotron Nano 2 VL](https://huggingface.co/blog/nvidia/nemotron-vlm-dataset-v2) leads in video understanding and document intelligence benchmarks such as MMMU, MathVista, AI2D, OCRBench, OCRBench-v2, OCR-Reasoning, ChartQA, DocVQA, and Video-MME, delivering top-tier accuracy in multimodal [reasoning](https://www.nvidia.com/en-us/glossary/ai-reasoning/), character recognition, chart reasoning, and visual question answering. This makes it ideal for building multimodal applications that automate data extraction and comprehension across videos, documents, forms, and charts with enterprise-grade precision.
+
+
+<p align="center">
+<picture>
+<img src="/assets/figures/2025-multimodal-nvidia-nemotron/figure1.png" width="100%">
+</picture>
+<br>
+Figure 1: Nemotron Nano 2 VL provides leading accuracy on various video understanding and document intelligence benchmarks
+</p>
+
+Improving Efficiency with EVS  
+With EVS, the model achieves higher throughput and faster response times without sacrificing accuracy. EVS technique prunes redundant frames, preserving semantic richness while enabling longer video processing efficiently. As a result, enterprises can analyze hours of footage, from meetings and training sessions to customer calls, in minutes, gaining actionable insights faster and at lower cost.
+
+
+<p align="center">
+<picture>
+<img src="/assets/figures/2025-multimodal-nvidia-nemotron/figure2.png" width="100%">
+</picture>
+<br>
+Figure 2: Accuracy trend of the Nemotron Nano 2 VL model across various token-drop thresholds using efficient video sampling on Video-MME and LongVideo benchmarks  
+</p>
+
 ## About Nemotron Nano 2 VL
 
 * Architecture:  
@@ -33,49 +59,35 @@ In this blog post, we’ll explore how Nemotron Nano 2 VL advances video underst
 
 ## Run optimized inference with vLLM
 
-Nemotron Nano 2 VL, achieves accelerated [inference](https://www.nvidia.com/en-us/glossary/ai-inference/) and serves more requests on the same GPU with BF16, FP8 and FP4 precision support. Follow these instructions to get started: 
+This guide demonstrates how to run Nemotron Nano 2 VL on vLLM, achieving accelerated [inference](https://www.nvidia.com/en-us/glossary/ai-inference/) and serving concurrent requests efficiently with BF16, FP8 and FP4 precision support.
 
-`Create a fresh conda environment`
+### Install vLLM
 
-````shell
+The support for Nemotron Nano 2 VL is available in the nightly version of vLLM. Run the command below to install vLLM:
 ```bash
-conda create -n nemotron-vllm-env python=3.10 -y
-conda activate nemotron-vllm-env
+uv venv
+uv pip install vllm --extra-index-url https://wheels.vllm.ai/nightly --prerelease=allow
 ```
-````
-
-`Make sure to use the main branch of the vLLM. Run the command below to install vLLM`
-
-````shell
-```bash
-!VLLM_USE_PRECOMPILED=1 pip install git+https://github.com/vllm-project/vllm.git@main 
-````
 
-`We can then serve this model via an OpenAI-compatible API`
 
-````shell
+### Deploy and query the inference server
+Deploy an OpenAI-compatible inference server with vLLM by running the following commands for BF16, FP8 and FP4 precision:
 
 ```bash
 vllm serve nvidia/Nemotron-Nano-12B-v2-VL-BF16 --trust-remote-code --dtype bfloat16 --video-pruning-rate 0
-```
 
 # FP8
-```bash
 vllm serve nvidia/Nemotron-Nano-VL-12B-V2-FP8 --trust-remote-code --quantization modelopt --video-pruning-rate 0
-```
 
 # FP4
-```bash
 vllm serve nvidia/Nemotron-Nano-VL-12B-V2-FP4-QAD --trust-remote-code --quantization modelopt_fp4 --video-pruning-rate 0
 ```
 
-````
-
-`Once the server is up and running, you can prompt the model using the below code snippets`
+Once the server is up and running, you can prompt the model using the below code snippet:
 
 ```python
 from openai import OpenAI
-client = OpenAI(base_url="http://127.0.0.1:8033/v1", api_key="null")
+client = OpenAI(base_url="http://localhost:8000/v1", api_key="null")
 # Simple chat completion
 resp = client.chat.completions.create(
     model="nvidia/Nemotron-Nano-12B-v2-VL-BF16",
@@ -92,48 +104,8 @@ resp = client.chat.completions.create(
 )
 print(resp.choices[0].message.content)
 ```
-
-For an easier setup with vLLM, refer to our getting started cookbook, available here.
-
-## Leading multimodal model for efficient video understanding and document intelligence 
-
-NVIDIA Nemotron Nano 2 VL brings both video understanding and document intelligence capabilities together in a single, highly efficient model. Built on the hybrid Transformer–Mamba architecture, it combines the reasoning strength of Transformer models with the compute efficiency of Mamba, achieving high throughput and low latency, allowing it to process multi-image inputs faster. 
-
-Trained on NVIDIA-curated, high-quality multimodal data, [Nemotron Nano 2 VL](https://huggingface.co/blog/nvidia/nemotron-vlm-dataset-v2) leads in video understanding and document intelligence benchmarks such as MMMU, MathVista, AI2D, OCRBench, OCRBench-v2, OCR-Reasoning, ChartQA, DocVQA, and Video-MME, delivering top-tier accuracy in multimodal [reasoning](https://www.nvidia.com/en-us/glossary/ai-reasoning/), character recognition, chart reasoning, and visual question answering. This makes it ideal for building multimodal applications that automate data extraction and comprehension across videos, documents, forms, and charts with enterprise-grade precision.
-
-
-<p align="center">
-<picture>
-<img src="/assets/figures/2025-multimodal-nvidia-nemotron/figure1.png" width="100%">
-</picture>
-<br>
-Figure 1: Nemotron Nano 2 VL provides leading accuracy on various video understanding and document intelligence benchmarks
-</p>
-
-Improving Efficiency with EVS  
-With EVS, the model achieves higher throughput and faster response times without sacrificing accuracy. EVS technique prunes redundant frames, preserving semantic richness while enabling longer video processing efficiently. As a result, enterprises can analyze hours of footage, from meetings and training sessions to customer calls, in minutes, gaining actionable insights faster and at lower cost.
-
-
-<p align="center">
-<picture>
-<img src="/assets/figures/2025-multimodal-nvidia-nemotron/figure2.png" width="100%">
-</picture>
-<br>
-Figure 2: Accuracy trend of the Nemotron Nano 2 VL model across various token-drop thresholds using efficient video sampling on Video-MME and LongVideo benchmarks  
-</p>
-
-## Get Started
-
-To summarize, Nemotron Nano 2 VL helps build scalable, cost-efficient agentic AI systems that truly understand documents and video. With open weights, training datasets, and recipes, developers gain full transparency and flexibility to fine-tune and deploy the model across any environment, from on-premise to cloud, for maximum security and privacy.  
-
-Ready to build enterprise-ready agents? 
-
--   Download model weights from Hugging Face - [BF16](https://huggingface.co/nvidia/NVIDIA-Nemotron-Nano-12B-v2-VL-BF16), [FP8](https://huggingface.co/nvidia/NVIDIA-Nemotron-Nano-12B-v2-VL-FP8), [FP4-QAD](https://huggingface.co/nvidia/NVIDIA-Nemotron-Nano-12B-v2-VL-NVFP4-QAD)
--   Run with vLLM for inference with [this notebook](https://github.com/NVIDIA-NeMo/Nemotron/blob/main/usage-cookbook/Nemotron-Nano2-VL/vllm_cookbook.ipynb)
--   [Technical report](https://research.nvidia.com/labs/adlr/files/NVIDIA-Nemotron-Nano-V2-VL-report.pdf) to build custom, optimized models with Nemotron techniques.
--   [Training dataset](https://huggingface.co/datasets/nvidia/Nemotron-VLM-Dataset-v2) is hosted on Hugging Face. Learn more [here](https://huggingface.co/blog/nvidia/nemotron-vlm-dataset-v2).
-
+For more examples, check out our vLLM cookbook: [Nemotron-Nano2-VL/vllm_cookbook.ipynb](https://github.com/NVIDIA-NeMo/Nemotron/blob/main/usage-cookbook/Nemotron-Nano2-VL/vllm_cookbook.ipynb)
 
 [*Share your ideas*](http://nemotron.ideas.nvidia.com/?ncid=so-othe-692335) *and vote on what matters to help shape the future of Nemotron.* 
 
-*Stay up to date on [NVIDIA Nemotron](https://developer.nvidia.com/nemotron) by subscribing to NVIDIA news and following NVIDIA AI on [LinkedIn](https://www.linkedin.com/showcase/nvidia-ai/posts/?feedView=all), [X](https://x.com/NVIDIAAIDev), [YouTube](https://www.youtube.com/@NVIDIADeveloper)*, *and the [Nemotron channel](https://discord.com/channels/1019361803752456192/1407781691698708682) on [Discord](https://discord.com/invite/nvidiadeveloper).*
\ No newline at end of file
+*Stay up to date on [NVIDIA Nemotron](https://developer.nvidia.com/nemotron) by subscribing to NVIDIA news and following NVIDIA AI on [LinkedIn](https://www.linkedin.com/showcase/nvidia-ai/posts/?feedView=all), [X](https://x.com/NVIDIAAIDev), [YouTube](https://www.youtube.com/@NVIDIADeveloper)*, *and the [Nemotron channel](https://discord.com/channels/1019361803752456192/1407781691698708682) on [Discord](https://discord.com/invite/nvidiadeveloper).*

From e7626f9770a300ab0485555e15be87fd4ee24e49 Mon Sep 17 00:00:00 2001
From: Roger Wang <hey@rogerw.io>
Date: Wed, 29 Oct 2025 19:16:34 -0700
Subject: [PATCH 04/12] update

Signed-off-by: Roger Wang <hey@rogerw.io>
---
 ...25-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/_posts/2025-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md b/_posts/2025-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md
index db9fe60..4c09ddf 100644
--- a/_posts/2025-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md
+++ b/_posts/2025-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md
@@ -4,8 +4,6 @@ title: "Run Multimodal Reasoning Agents with NVIDIA Nemotron on vLLM"
 author: "NVIDIA Nemotron Team"   
 ---
 
-# Run Multimodal Reasoning Agents with NVIDIA Nemotron on vLLM
-
 We are excited to release [NVIDIA Nemotron Nano 2 VL](https://huggingface.co/nvidia/Nemotron-Nano-12B-v2-VL-BF16), supported by vLLM. This open vision language model ([VLM](https://www.nvidia.com/en-us/glossary/vision-language-models/)) is built for video understanding and document intelligence.
 
 Nemotron Nano 2 VL uses a hybrid Transformer–Mamba design and delivers higher throughput while maintaining state-of-the-art multimodal reasoning accuracy. The model also features Efficient Video Sampling (EVS), a new technique that reduces redundant [tokens](https://blogs.nvidia.com/blog/ai-tokens-explained/) generation for video workloads , allowing processing of more videos with higher efficiency. 
@@ -106,6 +104,7 @@ print(resp.choices[0].message.content)
 ```
 For more examples, check out our vLLM cookbook: [Nemotron-Nano2-VL/vllm_cookbook.ipynb](https://github.com/NVIDIA-NeMo/Nemotron/blob/main/usage-cookbook/Nemotron-Nano2-VL/vllm_cookbook.ipynb)
 
+
 [*Share your ideas*](http://nemotron.ideas.nvidia.com/?ncid=so-othe-692335) *and vote on what matters to help shape the future of Nemotron.* 
 
 *Stay up to date on [NVIDIA Nemotron](https://developer.nvidia.com/nemotron) by subscribing to NVIDIA news and following NVIDIA AI on [LinkedIn](https://www.linkedin.com/showcase/nvidia-ai/posts/?feedView=all), [X](https://x.com/NVIDIAAIDev), [YouTube](https://www.youtube.com/@NVIDIADeveloper)*, *and the [Nemotron channel](https://discord.com/channels/1019361803752456192/1407781691698708682) on [Discord](https://discord.com/invite/nvidiadeveloper).*

From 194b929f30ef8d0ad7d2da4ce968dd260dd2f2b8 Mon Sep 17 00:00:00 2001
From: Roger Wang <hey@rogerw.io>
Date: Wed, 29 Oct 2025 19:41:38 -0700
Subject: [PATCH 05/12] add activation

Signed-off-by: Roger Wang <hey@rogerw.io>
---
 ...2025-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md | 1 +
 1 file changed, 1 insertion(+)

diff --git a/_posts/2025-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md b/_posts/2025-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md
index 4c09ddf..6b98f75 100644
--- a/_posts/2025-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md
+++ b/_posts/2025-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md
@@ -64,6 +64,7 @@ This guide demonstrates how to run Nemotron Nano 2 VL on vLLM, achieving acceler
 The support for Nemotron Nano 2 VL is available in the nightly version of vLLM. Run the command below to install vLLM:
 ```bash
 uv venv
+source .venv/bin/activate
 uv pip install vllm --extra-index-url https://wheels.vllm.ai/nightly --prerelease=allow
 ```
 

From 8230019eb88de37752421f2ba340e9685572c689 Mon Sep 17 00:00:00 2001
From: Roger Wang <hey@rogerw.io>
Date: Wed, 29 Oct 2025 19:44:15 -0700
Subject: [PATCH 06/12] clean up

Signed-off-by: Roger Wang <hey@rogerw.io>
---
 ...5-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/_posts/2025-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md b/_posts/2025-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md
index 6b98f75..8412fd4 100644
--- a/_posts/2025-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md
+++ b/_posts/2025-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md
@@ -6,7 +6,7 @@ author: "NVIDIA Nemotron Team"
 
 We are excited to release [NVIDIA Nemotron Nano 2 VL](https://huggingface.co/nvidia/Nemotron-Nano-12B-v2-VL-BF16), supported by vLLM. This open vision language model ([VLM](https://www.nvidia.com/en-us/glossary/vision-language-models/)) is built for video understanding and document intelligence.
 
-Nemotron Nano 2 VL uses a hybrid Transformer–Mamba design and delivers higher throughput while maintaining state-of-the-art multimodal reasoning accuracy. The model also features Efficient Video Sampling (EVS), a new technique that reduces redundant [tokens](https://blogs.nvidia.com/blog/ai-tokens-explained/) generation for video workloads , allowing processing of more videos with higher efficiency. 
+Nemotron Nano 2 VL uses a hybrid Transformer–Mamba design and delivers higher throughput while maintaining state-of-the-art multimodal reasoning accuracy. The model also features **Efficient Video Sampling (EVS)**, a new technique that reduces redundant [tokens](https://blogs.nvidia.com/blog/ai-tokens-explained/) generation for video workloads, allowing processing of more videos with higher efficiency. 
 
 In this blog post, we’ll explore how Nemotron Nano 2 VL advances video understanding and document intelligence, showcase real-world use cases and benchmark results, and guide you through getting started with vLLM for inference to unlock high-efficiency multimodal AI at scale.
 
@@ -25,7 +25,7 @@ Trained on NVIDIA-curated, high-quality multimodal data, [Nemotron Nano 2 VL](ht
 Figure 1: Nemotron Nano 2 VL provides leading accuracy on various video understanding and document intelligence benchmarks
 </p>
 
-Improving Efficiency with EVS  
+### Improving Efficiency with EVS  
 With EVS, the model achieves higher throughput and faster response times without sacrificing accuracy. EVS technique prunes redundant frames, preserving semantic richness while enabling longer video processing efficiently. As a result, enterprises can analyze hours of footage, from meetings and training sessions to customer calls, in minutes, gaining actionable insights faster and at lower cost.
 
 

From a1edfd473b580ff09a84b9630cd2cd191e7218c9 Mon Sep 17 00:00:00 2001
From: Roger Wang <hey@rogerw.io>
Date: Thu, 30 Oct 2025 09:09:53 -0700
Subject: [PATCH 07/12] Update
 _posts/2025-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md

Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Signed-off-by: Roger Wang <hey@rogerw.io>
---
 ...25-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/_posts/2025-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md b/_posts/2025-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md
index 8412fd4..f9d3eb9 100644
--- a/_posts/2025-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md
+++ b/_posts/2025-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md
@@ -103,7 +103,8 @@ resp = client.chat.completions.create(
 )
 print(resp.choices[0].message.content)
 ```
-For more examples, check out our vLLM cookbook: [Nemotron-Nano2-VL/vllm_cookbook.ipynb](https://github.com/NVIDIA-NeMo/Nemotron/blob/main/usage-cookbook/Nemotron-Nano2-VL/vllm_cookbook.ipynb)
+
+For more examples, check out our vLLM cookbook: [Nemotron-Nano2-VL/vllm_cookbook.ipynb](https://github.com/NVIDIA-NeMo/Nemotron/blob/main/usage-cookbook/Nemotron-Nano2-VL/vllm_cookbook.ipynb).
 
 
 [*Share your ideas*](http://nemotron.ideas.nvidia.com/?ncid=so-othe-692335) *and vote on what matters to help shape the future of Nemotron.* 

From 05d18cc5dab52003967a69f167c494d97ce3f488 Mon Sep 17 00:00:00 2001
From: Roger Wang <hey@rogerw.io>
Date: Thu, 30 Oct 2025 09:10:04 -0700
Subject: [PATCH 08/12] Update
 _posts/2025-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md

Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Signed-off-by: Roger Wang <hey@rogerw.io>
---
 ...025-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/_posts/2025-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md b/_posts/2025-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md
index f9d3eb9..32734e0 100644
--- a/_posts/2025-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md
+++ b/_posts/2025-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md
@@ -43,7 +43,7 @@ Figure 2: Accuracy trend of the Nemotron Nano 2 VL model across various token-dr
   * [CRADIOH-V2](https://huggingface.co/nvidia/C-RADIOv2-H) based Vision Encoder  
   * Efficient video sampling as token compression module  
   * Hybrid Transformer-Mamba Architecture- [Nemotron Nano 2 LLM](https://huggingface.co/nvidia/NVIDIA-Nemotron-Nano-9B-v2) backbone with reasoning.   
-* Accuracy  
+* Accuracy:
   * Leading accuracy on OCRBench v2  
   * 74 on average score (compared to 64.2 with current top VL model) on the following benchmarks: MMMU, MathVista, AI2D, OCRBench, OCRBench-v2, OCR-Reasoning, ChartQA, DocVQA, and Video-MME  
 * Model size: 12B  

From 2f090f179ae3fab4077713074483bd3474ef1374 Mon Sep 17 00:00:00 2001
From: Roger Wang <hey@rogerw.io>
Date: Thu, 30 Oct 2025 09:10:14 -0700
Subject: [PATCH 09/12] Update
 _posts/2025-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md

Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Signed-off-by: Roger Wang <hey@rogerw.io>
---
 ...025-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/_posts/2025-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md b/_posts/2025-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md
index 32734e0..efaeb5a 100644
--- a/_posts/2025-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md
+++ b/_posts/2025-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md
@@ -42,7 +42,7 @@ Figure 2: Accuracy trend of the Nemotron Nano 2 VL model across various token-dr
 * Architecture:  
   * [CRADIOH-V2](https://huggingface.co/nvidia/C-RADIOv2-H) based Vision Encoder  
   * Efficient video sampling as token compression module  
-  * Hybrid Transformer-Mamba Architecture- [Nemotron Nano 2 LLM](https://huggingface.co/nvidia/NVIDIA-Nemotron-Nano-9B-v2) backbone with reasoning.   
+  * Hybrid Transformer-Mamba Architecture - [Nemotron Nano 2 LLM](https://huggingface.co/nvidia/NVIDIA-Nemotron-Nano-9B-v2) backbone with reasoning.   
 * Accuracy:
   * Leading accuracy on OCRBench v2  
   * 74 on average score (compared to 64.2 with current top VL model) on the following benchmarks: MMMU, MathVista, AI2D, OCRBench, OCRBench-v2, OCR-Reasoning, ChartQA, DocVQA, and Video-MME  

From 1662ee5d8786964ae51229b1ef116654ea4b0ff8 Mon Sep 17 00:00:00 2001
From: Roger Wang <hey@rogerw.io>
Date: Fri, 31 Oct 2025 00:13:46 -0700
Subject: [PATCH 10/12] add

Signed-off-by: Roger Wang <hey@rogerw.io>
---
 ...5-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/_posts/2025-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md b/_posts/2025-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md
index efaeb5a..6fb862a 100644
--- a/_posts/2025-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md
+++ b/_posts/2025-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md
@@ -6,7 +6,7 @@ author: "NVIDIA Nemotron Team"
 
 We are excited to release [NVIDIA Nemotron Nano 2 VL](https://huggingface.co/nvidia/Nemotron-Nano-12B-v2-VL-BF16), supported by vLLM. This open vision language model ([VLM](https://www.nvidia.com/en-us/glossary/vision-language-models/)) is built for video understanding and document intelligence.
 
-Nemotron Nano 2 VL uses a hybrid Transformer–Mamba design and delivers higher throughput while maintaining state-of-the-art multimodal reasoning accuracy. The model also features **Efficient Video Sampling (EVS)**, a new technique that reduces redundant [tokens](https://blogs.nvidia.com/blog/ai-tokens-explained/) generation for video workloads, allowing processing of more videos with higher efficiency. 
+Nemotron Nano 2 VL uses a hybrid Transformer–Mamba design and delivers higher throughput while maintaining state-of-the-art multimodal reasoning accuracy. The model also features [**Efficient Video Sampling (EVS)**](https://arxiv.org/abs/2510.14624), a new technique that reduces redundant [tokens](https://blogs.nvidia.com/blog/ai-tokens-explained/) generation for video workloads, allowing processing of more videos with higher efficiency. 
 
 In this blog post, we’ll explore how Nemotron Nano 2 VL advances video understanding and document intelligence, showcase real-world use cases and benchmark results, and guide you through getting started with vLLM for inference to unlock high-efficiency multimodal AI at scale.
 
@@ -53,7 +53,7 @@ Figure 2: Accuracy trend of the Nemotron Nano 2 VL model across various token-dr
 * Get started:   
   * Download model weights from Hugging Face \- [BF16](https://huggingface.co/nvidia/Nemotron-Nano-12B-v2-VL-BF16), [FP8](https://huggingface.co/nvidia/Nemotron-Nano-12B-v2-VL-FP8), [FP4-QAD](https://huggingface.co/nvidia/Nemotron-Nano-12B-v2-VL-FP4-QAD)  
   * Run with vLLM for inference  
-  * [Technical report](https://www.overleaf.com/project/68d1d48c83696e11ba669f70) to build custom, optimized models with Nemotron techniques..
+  * [Technical report](https://research.nvidia.com/labs/adlr/files/NVIDIA-Nemotron-Nano-V2-VL-report.pdf) to build custom, optimized models with Nemotron techniques..
 
 ## Run optimized inference with vLLM
 

From 56723df18e355a75e42ca3b2533e36b97addf903 Mon Sep 17 00:00:00 2001
From: Roger Wang <hey@rogerw.io>
Date: Fri, 31 Oct 2025 00:17:47 -0700
Subject: [PATCH 11/12] update date and add recipe link

Signed-off-by: Roger Wang <hey@rogerw.io>
---
 ...25-10-31-run-multimodal-reasoning-agents-nvidia-nemotron.md} | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
 rename _posts/{2025-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md => 2025-10-31-run-multimodal-reasoning-agents-nvidia-nemotron.md} (97%)

diff --git a/_posts/2025-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md b/_posts/2025-10-31-run-multimodal-reasoning-agents-nvidia-nemotron.md
similarity index 97%
rename from _posts/2025-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md
rename to _posts/2025-10-31-run-multimodal-reasoning-agents-nvidia-nemotron.md
index 6fb862a..edfbfd2 100644
--- a/_posts/2025-10-29-run-multimodal-reasoning-agents-nvidia-nemotron.md
+++ b/_posts/2025-10-31-run-multimodal-reasoning-agents-nvidia-nemotron.md
@@ -104,7 +104,7 @@ resp = client.chat.completions.create(
 print(resp.choices[0].message.content)
 ```
 
-For more examples, check out our vLLM cookbook: [Nemotron-Nano2-VL/vllm_cookbook.ipynb](https://github.com/NVIDIA-NeMo/Nemotron/blob/main/usage-cookbook/Nemotron-Nano2-VL/vllm_cookbook.ipynb).
+For more examples, check out our vLLM cookbook: [Nemotron-Nano2-VL/vllm_cookbook.ipynb](https://github.com/NVIDIA-NeMo/Nemotron/blob/main/usage-cookbook/Nemotron-Nano2-VL/vllm_cookbook.ipynb) and vLLM recipe for [Nemotron Nano 2 VL](https://docs.vllm.ai/projects/recipes/en/latest/NVIDIA/Nemotron-Nano-12B-v2-VL.html).
 
 
 [*Share your ideas*](http://nemotron.ideas.nvidia.com/?ncid=so-othe-692335) *and vote on what matters to help shape the future of Nemotron.* 

From c9ac55f8b6017e91ecc3c9b55e29662a06e26236 Mon Sep 17 00:00:00 2001
From: Roger Wang <hey@rogerw.io>
Date: Fri, 31 Oct 2025 00:20:38 -0700
Subject: [PATCH 12/12] update link

Signed-off-by: Roger Wang <hey@rogerw.io>
---
 ...025-10-31-run-multimodal-reasoning-agents-nvidia-nemotron.md | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/_posts/2025-10-31-run-multimodal-reasoning-agents-nvidia-nemotron.md b/_posts/2025-10-31-run-multimodal-reasoning-agents-nvidia-nemotron.md
index edfbfd2..3d2f63c 100644
--- a/_posts/2025-10-31-run-multimodal-reasoning-agents-nvidia-nemotron.md
+++ b/_posts/2025-10-31-run-multimodal-reasoning-agents-nvidia-nemotron.md
@@ -104,7 +104,7 @@ resp = client.chat.completions.create(
 print(resp.choices[0].message.content)
 ```
 
-For more examples, check out our vLLM cookbook: [Nemotron-Nano2-VL/vllm_cookbook.ipynb](https://github.com/NVIDIA-NeMo/Nemotron/blob/main/usage-cookbook/Nemotron-Nano2-VL/vllm_cookbook.ipynb) and vLLM recipe for [Nemotron Nano 2 VL](https://docs.vllm.ai/projects/recipes/en/latest/NVIDIA/Nemotron-Nano-12B-v2-VL.html).
+For more examples, check out our [vLLM cookbook](https://github.com/NVIDIA-NeMo/Nemotron/blob/main/usage-cookbook/Nemotron-Nano2-VL/vllm_cookbook.ipynb) and [vLLM recipe for Nemotron Nano 2 VL](https://docs.vllm.ai/projects/recipes/en/latest/NVIDIA/Nemotron-Nano-12B-v2-VL.html).
 
 
 [*Share your ideas*](http://nemotron.ideas.nvidia.com/?ncid=so-othe-692335) *and vote on what matters to help shape the future of Nemotron.*