{"id":367,"date":"2022-03-04T14:28:10","date_gmt":"2022-03-04T06:28:10","guid":{"rendered":"https:\/\/www.wennroy.com\/?p=367"},"modified":"2022-03-04T14:28:11","modified_gmt":"2022-03-04T06:28:11","slug":"attention","status":"publish","type":"post","link":"https:\/\/wennroy.com\/index.php\/2022\/03\/04\/attention\/","title":{"rendered":"Attention"},"content":{"rendered":"\n<h2 class=\"wp-block-heading\">\u4eceLSTM\u5230Attention<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Sentence Representation<\/strong>: \u5bf9\u4e8eLSTM\u6765\u8bf4\uff0cLSTM\u5c06\u4e00\u53e5\u8bdd\u538b\u7f29\u6210\u4e00\u4e2a\u7279\u5f81\u5411\u91cf\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">It&#8217;s not ideal to compress the meaning of a sentence with variable length into a single vector.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u4e00\u4e2a\u8fc7\u957f\u7684\u53e5\u5b50\u53ea\u88ab\u4e00\u4e2aVector\u8868\u793a\u662f\u4e0d\u591f\u7684\u3002\u56e0\u6b64\uff0c\u5f88\u76f4\u63a5\u7684\u6765\u8bf4\uff0c\u6211\u4eec\u4f1a\u60f3\u8981\u7528\u591a\u4e2a\u5411\u91cf\u6765\u8868\u793a\u6211\u4eec\u7684\u53e5\u5b50\u3002\u80fd\u591f\u5b9e\u73b0\u4ed6\u7684\u65b9\u5f0f\u5c31\u662fAttention\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>\u7f16\u7801\u5c42\u9762\uff1a<\/strong>\u5c06\u53e5\u5b50\u4e2d\u7684\u6bcf\u4e2a\u5355\u8bcd\u90fd\u4f5c\u4e3a\u4e00\u4e2a\u5411\u91cf\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>\u89e3\u7801\u5c42\u9762\uff1a<\/strong>\u5229\u7528attention weights\uff08\u52a0\u6743\uff09\u8ba1\u7b97\u5404\u4e2a\u5411\u91cf\u7684\u7ebf\u6027\u7ec4\u5408\u3002\u6700\u540e\u5229\u7528\u8f93\u51fa\u6765\u51b3\u5b9a\u4e0b\u4e00\u4e2a\u8bcd\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5728seq2seq\u6a21\u578b\u4e2d\uff0c\u6211\u4eec\u4e5f\u7ecf\u5e38\u5c06 <em>target hidden vector (query)<\/em> to <em>all source vectors (keys)<\/em> \u79f0\u4f5c target-to-source cross attention.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Attention\u6a21\u578b<\/h2>\n\n\n\n<figure class=\"wp-block-image size-large is-resized\"><img loading=\"lazy\" decoding=\"async\" src=\"https:\/\/i.imgur.com\/sefA19N.png\" alt=\"\" width=\"670\" height=\"491\"\/><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">Attention score\u7531\u70b9\u79ef+Softmax\u5f97\u5230\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">$a_l$\u6211\u4eec\u8ba4\u4e3a\u662fcontext vector\uff0c\u662f\u5bf9\u6240\u6709\u539f\u59cb\u8f93\u5165\u7684\u52a0\u6743\uff08\u52a0\u6743\u548c\u7531\u6ce8\u610f\u529b\uff08$a_{t,l}$\uff09\u5f97\u5206\u6765\u5f97\u5230\uff09\u548c\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-large is-resized\"><img loading=\"lazy\" decoding=\"async\" src=\"https:\/\/i.imgur.com\/VSpHMon.png\" alt=\"\" width=\"633\" height=\"531\"\/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\">Different Attention Score functions<\/h2>\n\n\n\n<ul class=\"wp-block-list\"><li><strong>Multi-layer Perceptron<\/strong> (Bahdanau et al. 2015)<br>$$a(\\boldsymbol q, \\boldsymbol k ) = \\boldsymbol w_2^T\\tanh(W_1[\\boldsymbol{q};\\boldsymbol{k}])$$<br>$\\tanh$ is a non-linear function. \u603b\u4f53\u4e0a\u4f1a\u66f4\u52a0\u7075\u6d3b\uff08Flexible\uff09\uff0c\u5728\u5927\u6570\u636e\u96c6\u4e0a\u8868\u73b0\u4f1a\u597d\u4e00\u4e9b\u3002<\/li><li><strong>Bilinear<\/strong> (Luong et al. 2015)<br>$$a(\\boldsymbol q, \\boldsymbol k) = \\boldsymbol q^T W \\boldsymbol k$$<br>\u5229\u7528$W$\u77e9\u9635\u5c06$\\boldsymbol q$\u5411\u91cf\u6295\u5f71\u5230$\\boldsymbol k$\u6240\u5904\u7684\u7a7a\u95f4\u4e0a\uff0c\u518d\u8fdb\u884c\u70b9\u79ef\u64cd\u4f5c\u3002<\/li><li><strong>Dot Product<\/strong> (Luong et al. 2015)<br>$$a(\\boldsymbol q, \\boldsymbol k) = \\boldsymbol q^T \\boldsymbol k$$<br>\u8981\u6c42\u4e24\u4e2a\u5411\u91cf\u7684sizes\u5fc5\u987b\u4e00\u81f4\u3002\u5f53\u7ef4\u5ea6\u589e\u52a0\u7684\u65f6\u5019\uff0c\u8f93\u51fa\u7684value\u5c06\u4f1a\u589e\u52a0\u3002\u5982\u679cvalue\u4e0d\u591f\u7a33\u5b9a\uff0c\u90a3\u4e48\u8bad\u7ec3\u53ef\u80fd\u4e0d\u4f1a\u5f88\u7a33\u5b9a\u3002\u56e0\u6b64<strong>Scaled Dot Product<\/strong>\u662f\u4e2a\u89e3\u51b3\u529e\u6cd5\u3002<\/li><li><strong>Scaled Dot Product<\/strong><br>$$a(\\boldsymbol q, \\boldsymbol k)  = \\frac{\\boldsymbol q^T\\boldsymbol k}{\\sqrt{|\\boldsymbol k|}}$$<br><\/li><\/ul>\n\n\n\n<h2 class=\"wp-block-heading\">Attention is all you need<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\">Transformer<\/h3>\n\n\n\n<figure class=\"wp-block-image size-large is-resized\"><img loading=\"lazy\" decoding=\"async\" src=\"https:\/\/i.imgur.com\/mX8ouQi.png\" alt=\"\" width=\"304\" height=\"370\"\/><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">\u4e00\u4e9b\u7279\u70b9\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\"><li>\u76f8\u6bd4\u4e8eLSTM\u4ece\u5de6\u5230\u53f3\u7684\u8ba1\u7b97\u65b9\u5f0f\uff0cTransformer\u53ef\u4ee5\u5e76\u884c\u8ba1\u7b97\uff0c\u6548\u7387\u66f4\u9ad8\u3002<\/li><li>seq2seq\u6a21\u578b\uff0c\u4f46\u5b8c\u5168\u57fa\u4e8e\u6ce8\u610f\u529b\u673a\u5236(Attention)\u3002<\/li><li>\u53ea\u6709\u77e9\u9635\u8ba1\u7b97\uff0c\u610f\u5473\u7740\u8bad\u7ec3\u8f83\u5feb\u3002<\/li><\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">\u4e00\u4e9b\u91cd\u8981\u7684\u7ec4\u6210\u90e8\u5206\uff1a<\/p>\n\n\n\n<ol class=\"wp-block-list\"><li><strong>Self-attention <\/strong>&#8211; allows parallel computing of all tokens<\/li><li><strong>Multi-headed attention<\/strong> \u2014 allows querying multiple positions at each layer<\/li><li><strong>Position encoding <\/strong>&#8211; adds position information to each token<\/li><li><strong>Adding nonlinearities <\/strong>\u2014 combines features from a self-attention layer<\/li><li><strong>Masked decoding<\/strong> &#8211; prevents attention lookups in the future tokens<\/li><\/ol>\n\n\n\n<h4 class=\"wp-block-heading\">Self-Attention<\/h4>\n\n\n\n<p class=\"wp-block-paragraph\">Querys\u4e0ekeys\u90fd\u662f\u81ea\u5df1\u7684attention\u53eb\u4f5cSelf-Attention\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u4e00\u822c\u6765\u8bf4\uff0c\u6211\u4eec\u5c06\u8f93\u5165$x_t$\u5148\u901a\u8fc7feedforward layer\u6216\u8005\u975e\u7ebf\u6027\u7684\u51fd\u6570\u6620\u5c04\u5230$h_t$\uff0c\u7136\u540e\u590d\u5236\u4e09\u4efd\uff0c\u6620\u5c04\u5230\u4e09\u4e2a\u5411\u91cf$k_t,q_t,v_t$\uff0c\u5206\u522b\u88ab\u79f0\u4f5ckeys, querys, values. <\/p>\n\n\n\n<figure class=\"wp-block-image size-large is-resized\"><img loading=\"lazy\" decoding=\"async\" src=\"https:\/\/i.imgur.com\/fJXnnwW.png\" alt=\"\" width=\"585\" height=\"394\"\/><\/figure>\n\n\n\n<figure class=\"wp-block-image size-large is-resized\"><img loading=\"lazy\" decoding=\"async\" src=\"https:\/\/i.imgur.com\/n3wrJIF.png\" alt=\"\" width=\"597\" height=\"406\"\/><\/figure>\n\n\n\n<h4 class=\"wp-block-heading\">Multi-headed Attention<\/h4>\n\n\n\n<p class=\"wp-block-paragraph\">Repeat Attention many times.<\/p>\n\n\n\n<figure class=\"wp-block-image size-large is-resized\"><img loading=\"lazy\" decoding=\"async\" src=\"https:\/\/i.imgur.com\/714QUbt.png\" alt=\"\" width=\"519\" height=\"379\"\/><figcaption>Multi-head Self-attention<\/figcaption><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">\u91cd\u590dattention\u53ef\u4ee5\u5b66\u5230\u4e00\u4e9b\u5176\u4ed6\u5173\u7cfb\uff0c\u4f8b\u5982\u4e00\u4e2a\u8bcd\u524d\u540e\u7684\u8bcd\uff0c\u6700\u7ec8\u5f97\u5230$d$\u7ef4\u5ea6\u7684context vector\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">$$\\boldsymbol a_l = [a_{l,I},\\cdots,a_{l,1}]\\in\\mathbb{R}^d,\\quad a_{l,i}\\in\\mathbb{R}^{\\frac{d}{I}}$$<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Where $I$ is the umber of heads. 8\u4e2a\u5de6\u53f3\u7684heads\u5728\u5927\u578b\u6a21\u578b\u4e2d\u8868\u73b0\u8f83\u597d\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-large is-resized\"><img loading=\"lazy\" decoding=\"async\" src=\"https:\/\/i.imgur.com\/aiWaxgK.png\" alt=\"\" width=\"790\" height=\"355\"\/><figcaption>Multi-head Self-attention layer<\/figcaption><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">\u4f46\u662f\u8fd9\u79cd\u591a\u5934\u81ea\u6ce8\u610f\u529b\u673a\u5236\u4e0b\uff0c\u81ea\u6ce8\u610f\u529b\u4ecd\u7136\u662f\u524d\u4e00\u4e2a\u5c42\u7684\u7ebf\u6027\u53d8\u5316\uff0c\u8fd9\u4f1a\u5bfc\u81f4\u5f88\u96be\u5b66\u4e60complex data\u3002\u56e0\u6b64\u4e00\u822c\u4f1a\u9009\u62e9\u5728multi-head self-attention\u4e4b\u540e\u5916\u52a0\u4e00\u5c42feedforward layer\uff0c\u589e\u52a0\u4e00\u4e2a\u975e\u7ebf\u6027\u51fd\u6570\uff0c\u6700\u7ec8\u5f97\u5230\u6211\u4eec\u8fd9\u4e00\u5c42\u7684\u8f93\u51fa\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-large is-resized\"><img loading=\"lazy\" decoding=\"async\" src=\"https:\/\/i.imgur.com\/U4mAL0F.png\" alt=\"\" width=\"379\" height=\"388\"\/><figcaption>With Feedforward (Non-linear function appliled) Layer<\/figcaption><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">\u6700\u7ec8\u6784\u6210\u4e86\u5b8c\u6574\u7684Transformer\u5757\u3002<\/p>\n\n\n\n<h4 class=\"wp-block-heading\">Positional Encoding<\/h4>\n\n\n\n<p class=\"wp-block-paragraph\">\u7531\u4e8e\u6ce8\u610f\u529b\u673a\u5236\u5b8c\u5168\u6ca1\u6709\u5728\u6a21\u578b\u91cc\u5f15\u5165\u4f4d\u7f6e\u4fe1\u606f\uff0c\u6211\u4eec\u989d\u5916\u5f15\u5165\u4f4d\u7f6e\u7f16\u7801\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5728\u4e0d\u662fTransformer\u7684\u4e00\u4e9b\u4f4d\u7f6e\u4fe1\u606fencoding\u4e2d\uff0c\u6211\u4eec\u4f1a\u9009\u7528Naive Positional Encoding\uff1a<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">$$\\bar{x}_t = [x_t, t]^T$$<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u8fd9\u79cd\u65b9\u5f0f\u5e76\u4e0d\u6709\u6548\uff0c\u56e0\u4e3a\u6211\u4eec\u66f4\u9700\u8981\u7684\u662f\u76f8\u5bf9\u4f4d\u7f6e\u4fe1\u606f\u800c\u4e0d\u662f\u7edd\u5bf9\u4f4d\u7f6e\u4fe1\u606f\u3002\u4f8b\u5982\u4e00\u4e2a\u540d\u8bcd\u5728\u4e0d\u540c\u7684\u53e5\u5b50\u4e2d\u7684\u4f4d\u7f6e\u53ef\u80fd\u5b8c\u5168\u4e0d\u4e00\u6837\u3002\u56e0\u6b64\u6211\u4eec\u8003\u8651frequency-based representations.<\/p>\n\n\n\n<figure class=\"wp-block-image size-large is-resized\"><img loading=\"lazy\" decoding=\"async\" src=\"https:\/\/i.imgur.com\/TWY5A0f.png\" alt=\"\" width=\"858\" height=\"234\"\/><figcaption>Positional Encoding 1<\/figcaption><\/figure>\n\n\n\n<figure class=\"wp-block-image size-large is-resized\"><img loading=\"lazy\" decoding=\"async\" src=\"https:\/\/i.imgur.com\/ahH5aYc.png\" alt=\"\" width=\"496\" height=\"376\"\/><figcaption>Positional Encoding 2<\/figcaption><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">\u4e00\u4e2a\u5f88\u5de7\u5999\u7684\u8bbe\u5b9a\u662f\u7c7b\u4f3c\u4e8e\u4e0a\u56fe\u7684\u8bbe\u5b9a\u3002\u4e00\u5f00\u59cb\u7684\u524d\u51e0\u4e2a\u4f4d\u7f6e\u4fe1\u606f$\\sin(t\/10000^{2*1\/d})$\u9891\u7387\u8f83\u9ad8\uff0c\u8fd9\u4f1a\u5bfc\u81f4\u503e\u5411\u4e8e\u533a\u5206\u76f8\u90bb\u7684\u4e24\u4e2a\u8bcd\uff0c\u4f8b\u5982\u533a\u5206\u662f\u7b2c\u5947\u6570\u4e2a\u8bcd\u6216\u8005\u7b2c\u5076\u6570\u4e2a\u8bcd \uff08\u5bf9\u5e94\u56fe\u4e0a\u7684&#8221;even-odd&#8221; indicator\uff09\uff0c\u800c\u8f83\u9ad8\u7ef4\u5ea6\u7684\u4f4d\u7f6e\u4fe1\u606f$\\sin(t\/10000^{2*\\frac{d}{2}\/d})$\u6709\u8f83\u4f4e\u7684\u9891\u7387\uff0c\u9707\u8361\u5468\u671f\u957f\uff0c\u6b63\u503c\u6216\u8005\u8d1f\u503c\u5c31\u80fd\u533a\u5206\u4ed6\u5230\u5e95\u662f\u4f4d\u4e8e\u53e5\u5b50\u7684\u524d\u534a\u6bb5\u8fd8\u662f\u540e\u534a\u6bb5\u3002\u4ece\u7b2c\u4e00\u5f20\u56fe\u6765\u770b\uff0c\u80fd\u591f\u533a\u5206\u53e5\u5b50\u524d\u534a\u6bb5\u540e\u534a\u6bb5\u7684\u7ef4\u6570\u5728\u4e2d\u95f4\u7684\u4f4d\u7f6e\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u6700\u540e\uff0c\u6211\u4eec\u5c06concat\u539f\u59cb\u8f93\u5165\u5411\u91cf\u5f97\u5230\u65b0\u7684\u8f93\u5165\u5411\u91cf\uff0c<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">$$\\tilde{x}_t = [x_t,p_t]^T$$<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Downside: \u901a\u5e38\u60c5\u51b5\u4e0b\u6211\u4eec\u8981\u786e\u8ba4\u6700\u957f\u7684\u5e8f\u5217\uff0c\u6211\u4eec\u5e76\u4e0d\u80fd\u751f\u4ea7\u8d85\u8fc7\u6700\u957f\u5e8f\u5217\u7684\u4f4d\u7f6e\u4fe1\u606f\u3002<\/p>\n\n\n\n<h4 class=\"wp-block-heading\">Masked attention for target sentence<\/h4>\n\n\n\n<p class=\"wp-block-paragraph\">\u4e0d\u540c\u4e8eBERT (Bidirectional)\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u6211\u4eec\u53ea\u5141\u8bb8\u81ea\u6ce8\u610f\u529b\u5173\u6ce8\u5230\u4e4b\u524d\u7684tokens\uff0c\u800c\u4e0d\u8003\u8651\u672a\u6765\u7684tokens\uff0c\u4e00\u4e2a\u7b80\u5355\u7684\u529e\u6cd5\u5c31\u662f\u8ba9<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">$$e_{l,t} =\\left\\{ \\begin{array}{ll}q_l\\cdot k_t &amp;\\text{if }l\\geq t\\\\-\\infty &amp;\\text{Otherwise}\\end{array}\\right.$$<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"> \u5b9e\u9645\u64cd\u4f5c\u4e2d\uff0c\u4e00\u822c\u76f4\u63a5\u5728softmax\u4e2d\uff0c\u5c06$\\exp(e_{l,t})$\u66ff\u6362\u4e3a0 if $l&lt;t$. <strong>Multiply the attention matrix by 0-1 masking matrix.<\/strong><\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Attention Tricks<\/h3>\n\n\n\n<ul class=\"wp-block-list\"><li><strong>Self Attention:<\/strong> Each layer combines words with others<\/li><li><strong>Multi-headed Attention:<\/strong> 8 attention heads learned independently<\/li><li><strong>Normalized Dot-product Attention:<\/strong> Remove bias in dot product when using large networks<\/li><li><strong>Positional Encodings:<\/strong> Make sure that even if we don&#8217;t have RNN, can still distinguish positions<\/li><\/ul>\n\n\n\n<h3 class=\"wp-block-heading\">Training Tricks<\/h3>\n\n\n\n<ul class=\"wp-block-list\"><li><strong>Layer Normalization:<\/strong> Help ensure that layers remain in reaonable range<\/li><li><strong>Specialized Training Schedule:<\/strong> Adjust default learning rate of the Adam optimizer<\/li><li><strong>Label Smoothing:<\/strong> Insert some uncertainty in the training process. Add some values on ground true values. Better for generalization.<\/li><li><strong>Masking for Efficient Training<\/strong><\/li><\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">Code Walk: <a href=\"https:\/\/nlp.seas.harvard.edu\/2018\/04\/03\/attention.html\" data-type=\"URL\" data-id=\"https:\/\/nlp.seas.harvard.edu\/2018\/04\/03\/attention.html\">https:\/\/nlp.seas.harvard.edu\/2018\/04\/03\/attention.html<\/a><\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Some Drawbacks<\/h3>\n\n\n\n<ul class=\"wp-block-list\"><li>Slow to decode<\/li><li>don&#8217;t necessarily outperform RNNs<\/li><li>hard to train on small data<\/li><\/ul>\n\n\n\n<h2 class=\"wp-block-heading\">Some models better than Attention<\/h2>\n\n\n\n<ul class=\"wp-block-list\"><li>Hard Attention. (Xu et al. 2015)  (Lei et al. 2016)<br>Instead of a soft interpolation, Make a <strong>zero-one decision<\/strong> about where to attend. Requires methods such as reinforcement.<\/li><li>Monotonic Attention.<\/li><li>Bidirectional Training.<\/li><\/ul>\n\n\n\n<h2 class=\"wp-block-heading\">References<\/h2>\n\n\n\n<ol class=\"wp-block-list\"><li>Vaswani, Ashish, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, \u0141ukasz Kaiser, and Illia Polosukhin. &#8220;Attention is all you need.&#8221;\u00a0<em>Advances in neural information processing systems<\/em>\u00a030 (2017).<\/li><\/ol>\n","protected":false},"excerpt":{"rendered":"<p>\u4eceLSTM\u5230Attention Sentence  &hellip;<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_import_markdown_pro_load_document_selector":0,"_import_markdown_pro_submit_text_textarea":"","footnotes":""},"categories":[42,43],"tags":[],"class_list":["post-367","post","type-post","status-publish","format-standard","hentry","category-deeplearning","category-nlp"],"_links":{"self":[{"href":"https:\/\/wennroy.com\/index.php\/wp-json\/wp\/v2\/posts\/367","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/wennroy.com\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/wennroy.com\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/wennroy.com\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/wennroy.com\/index.php\/wp-json\/wp\/v2\/comments?post=367"}],"version-history":[{"count":52,"href":"https:\/\/wennroy.com\/index.php\/wp-json\/wp\/v2\/posts\/367\/revisions"}],"predecessor-version":[{"id":419,"href":"https:\/\/wennroy.com\/index.php\/wp-json\/wp\/v2\/posts\/367\/revisions\/419"}],"wp:attachment":[{"href":"https:\/\/wennroy.com\/index.php\/wp-json\/wp\/v2\/media?parent=367"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/wennroy.com\/index.php\/wp-json\/wp\/v2\/categories?post=367"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/wennroy.com\/index.php\/wp-json\/wp\/v2\/tags?post=367"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}