Network　Function　Virtualization　(NFV)　decouples　network　functions　from　the　dedicated　hardware　and　produces　Virtual　Network　Functions　(VNFs)　in　software.　The　VNFs　are　placed　on　hardware　and　are　linked　together　to　build　a　service　chain.　The　design　of　an　efficient　VNF　placement　algorithm　is　crucial.　The　rapid　development　of　machine　learning,　especially　Deep　Reinforcement　Learning　(Deep　RL),　allows　us　to　address　this　problem.　In　this　paper,　we　present　an　attention　based　sequence　to　sequence　Deep　RL　method　for　VNF　placement.　Our　approach　is　a　policy　based　method　optimized　by　REINFORCE　with　baseline.　Our　model　receives　physical　hosts　and　service　chain　as　input　and　produces　the　output　sequence　step　by　step　with　attention　encoder　and　decoder.　We　demonstrate　that　our　method　outperforms　the　existing　learning　method　and　greedy　heuristic.　©　2020　IEEE.