{"id":4769,"date":"2023-09-11T13:28:18","date_gmt":"2023-09-11T10:28:18","guid":{"rendered":"https:\/\/datakapital.com\/blog\/?p=4769"},"modified":"2025-07-20T01:26:53","modified_gmt":"2025-07-19T22:26:53","slug":"buyuk-dil-modeli-nedir-ve-hangi-alanlarda-kullanilabilir","status":"publish","type":"post","link":"https:\/\/datakapital.com\/blog\/buyuk-dil-modeli-nedir-ve-hangi-alanlarda-kullanilabilir\/","title":{"rendered":"B\u00fcy\u00fck Dil Modeli Nedir ve Hangi Alanlarda Kullan\u0131labilir?"},"content":{"rendered":"<p>Yapay zek\u00e2 alan\u0131nda ya\u015fanan h\u0131zl\u0131 geli\u015fmeler, <strong>b\u00fcy\u00fck dil modeli<\/strong> gibi \u00e7e\u015fitli alanlarda yeni i\u015f modelleri geli\u015ftirilmesinin \u00f6n\u00fcn\u00fc a\u00e7m\u0131\u015ft\u0131r. <a href=\"https:\/\/datakapital.com\/blog\/yapay-zekanin-kronolojisi\/\">Yapay Zeka kronolojisine<\/a> bakt\u0131\u011f\u0131m\u0131zda asl\u0131nda \u201cTransformer\u201d teknolojisi ile b\u00fcy\u00fck bir ba\u015flang\u0131ca ilk ad\u0131m\u0131 insanl\u0131k olarak att\u0131\u011f\u0131m\u0131z\u0131 s\u00f6yleyebiliriz. Peki nedir Transformerlar k\u0131saca hat\u0131rlayal\u0131m:<\/p>\n<p>Transformer, \u00f6zellikle do\u011fal dil i\u015fleme (NLP) ve yapay zek\u00e2 (AI) alanlar\u0131nda b\u00fcy\u00fck bir etki yaratm\u0131\u015f olan bir derin \u00f6\u011frenme modelidir. Bu model, Google taraf\u0131ndan 2017 y\u0131l\u0131nda &#8220;Attention Is All You Need&#8221; \u00a0[1] ba\u015fl\u0131kl\u0131 bir makalede tan\u0131t\u0131lm\u0131\u015ft\u0131r. Transformer modeli, \u00f6zellikle b\u00fcy\u00fck veri k\u00fcmesi \u00fczerinde e\u011fitilmi\u015f ve sonu\u00e7 olarak \u00e7eviri, dil modellenmesi, metin s\u0131n\u0131fland\u0131rma ve daha bir\u00e7ok g\u00f6revde b\u00fcy\u00fck ba\u015far\u0131 elde etmi\u015ftir. Bu mimari Kodlay\u0131c\u0131 ve \u00c7\u00f6z\u00fcc\u00fc olmak \u00fczere iki bile\u015fenden olu\u015fan Seq2Seq bir mimaridir. Modelin mimari \u00e7izimini \u015eekil 1\u2019de g\u00f6rebilirsiniz:<\/p>\n<figure style=\"width: 608px\" class=\"wp-caption alignnone\"><img fetchpriority=\"high\" decoding=\"async\" class=\"wp-image-4770 size-full\" src=\"https:\/\/datakapital.com\/blog\/wp-content\/uploads\/2023\/09\/Transformer-Derin-Ogrenme-Mimarisi.jpg\" alt=\"Transformer Derin \u00d6\u011frenme\" width=\"608\" height=\"521\" title=\"\" srcset=\"https:\/\/datakapital.com\/blog\/wp-content\/uploads\/2023\/09\/Transformer-Derin-Ogrenme-Mimarisi.jpg 608w, https:\/\/datakapital.com\/blog\/wp-content\/uploads\/2023\/09\/Transformer-Derin-Ogrenme-Mimarisi-300x257.jpg 300w, https:\/\/datakapital.com\/blog\/wp-content\/uploads\/2023\/09\/Transformer-Derin-Ogrenme-Mimarisi-150x129.jpg 150w, https:\/\/datakapital.com\/blog\/wp-content\/uploads\/2023\/09\/Transformer-Derin-Ogrenme-Mimarisi-450x386.jpg 450w\" sizes=\"(max-width: 608px) 100vw, 608px\" \/><figcaption class=\"wp-caption-text\">\u015eekil 1. Transformer Derin \u00d6\u011frenme Mimarisi<\/figcaption><\/figure>\n<p>&nbsp;<\/p>\n<p>Peki Nedir Transformar\u2019\u0131n teknik \u00f6zellikleri:<\/p>\n<ul>\n<li>\u00d6z-dikkat ve Dikkat Mekan\u0131zmalar\u0131: Modelin en temel yenili\u011fini olu\u015fturan dikkat mekanizmas\u0131 metin girdisi ile birlikte hangi eleman\u0131n daha \u00f6ncelikli oldu\u011funu belirten bir attention dizisine sahip olmas\u0131d\u0131r. B\u00f6ylece model girdi metninin \u00f6nemli k\u0131s\u0131mlar\u0131na odaklanabilrmektedir. Ba\u011flamsal anlamlar\u0131 g\u00fc\u00e7lendirmeyi sa\u011flar. \u00d6z-dikkat mekanizmas\u0131 ise bir eleman\u0131n di\u011fer elemanlarla olan ili\u015fkisini modellemeyi sa\u011flar.<\/li>\n<li>Pozisyonel Kodlama ve Paralel \u0130\u015flem: Transformer bir kelimenin g\u00f6nderilen girdinin ka\u00e7\u0131nc\u0131 eleman\u0131 oldu\u011funu da kodlar. Pozisyonel kodlama RNN, LSTM gibi geleneksel derin \u00f6\u011frenme y\u00f6ntemlerinden farkl\u0131 olarak paralel i\u015flem yapabilmeyi sa\u011flar. Bu, e\u011fitim ve \u00e7eviri gibi g\u00f6revlerin daha h\u0131zl\u0131 tamamlanmas\u0131n\u0131 sa\u011flamaktad\u0131r.<\/li>\n<li>\u00c7oklu Ba\u015fl\u0131k Dikkat (Multi-Head Attention): Transformer modeli, dikkat mekanizmas\u0131n\u0131 birden fazla ba\u015fl\u0131k (head) kullanarak uygular. Her ba\u015fl\u0131k, farkl\u0131 \u00f6zelliklere dikkat eder ve daha sonra sonu\u00e7lar\u0131 birle\u015ftirir. Bu, modelin farkl\u0131 dil \u00f6zelliklerini yakalamas\u0131na yard\u0131mc\u0131 olur.<\/li>\n<\/ul>\n<p>Transformer\u2019\u0131n \u00f6nerilmesinin hemen arkas\u0131ndan B\u00fcy\u00fck Dil Modeli kavram\u0131 teknik olarak hayat\u0131m\u0131za girdi. \u015eekil1\u2019 de bir transformera ait iki alt mod\u00fcl\u00fc g\u00f6r\u00fcyoruz. Sol taraf Kodlay\u0131c\u0131 ve sa\u011f taraf \u00c7\u00f6z\u00fcc\u00fc olarak nitelendirilmektedir.<\/p>\n<p><strong>BERT (Bidirectional Encoder Representations from Transformers)<\/strong><\/p>\n<p>\u00d6nceden e\u011fitilmi\u015f bir dil modeli olan BERT [2] Kodlay\u0131c\u0131lar\u0131n ard arda ba\u011flanmas\u0131yla elde edilen Google taraf\u0131ndan \u00f6nerilmi\u015f bir modeldir.\u00a0 Model, 3.3 milyar kelime i\u00e7eren Wikipedia ve 2.5 milyar kelime i\u00e7eren BookCorpus adl\u0131 iki b\u00fcy\u00fck veri k\u00fcmesinde e\u011fitilmi\u015ftir. \u00d6nceden e\u011fitilmi\u015f kavram\u0131 \u015fu anlama gelmektedir: Elimizde b\u00fcy\u00fck bir veri ile e\u011fitilmi\u015f bir dil modelleyicimiz var ve biz istedi\u011fimiz bir \u00f6zel g\u00f6rev i\u00e7in ona ince-ayar yapabiliriz. \u015eekil 2\u2019de BERT modelini ve ince-ayarlanmas\u0131n\u0131 g\u00f6rebilirsiniz. E<sub>1<\/sub>, E<sub>2<\/sub>, \u2026, E<sub>m<\/sub> art arda ba\u011flanm\u0131\u015f Kodlay\u0131c\u0131lar\u0131 temsil etmektedir. BERT, maskeli dil modelleme (MLM) ve sonraki c\u00fcmle tahmini (NSP) hedefleriyle e\u011fitilmi\u015f bu sebeple birazdan bahsedece\u011fimiz B\u00fcy\u00fck Dil Modelleri (Large Language Model \u2013 LLM) gibi \u00fcretici de\u011fildir. Bunun yerine s\u0131n\u0131fland\u0131rma, soru cevaplama ve varl\u0131k tan\u0131ma (Named Entity Recognition \u2013 NER) problemlerini \u00e7\u00f6zmek i\u00e7in s\u0131kl\u0131kla kullan\u0131lmaktad\u0131r.<\/p>\n<p><img decoding=\"async\" class=\"alignnone size-full wp-image-4771\" src=\"https:\/\/datakapital.com\/blog\/wp-content\/uploads\/2023\/09\/BERT-Mimarisi.jpg\" alt=\"BERT Mimarisi Nas\u0131l \u00c7al\u0131\u015f\u0131r\" width=\"605\" height=\"302\" title=\"\" srcset=\"https:\/\/datakapital.com\/blog\/wp-content\/uploads\/2023\/09\/BERT-Mimarisi.jpg 605w, https:\/\/datakapital.com\/blog\/wp-content\/uploads\/2023\/09\/BERT-Mimarisi-300x150.jpg 300w, https:\/\/datakapital.com\/blog\/wp-content\/uploads\/2023\/09\/BERT-Mimarisi-150x75.jpg 150w, https:\/\/datakapital.com\/blog\/wp-content\/uploads\/2023\/09\/BERT-Mimarisi-450x225.jpg 450w\" sizes=\"(max-width: 605px) 100vw, 605px\" \/><\/p>\n<p><em>GPT (Generative Pre-trained Transformer)<\/em><\/p>\n<p>OpenAI taraf\u0131ndan \u00f6nerilen GPT modeli transformer modelinin \u00c7\u00f6z\u00fcc\u00fc bloklar\u0131n\u0131n art arda ba\u011flanmas\u0131 ile elde edilmi\u015ftir. Bu model daha \u00e7ok \u00e7eviri, \u00fcretme gibi g\u00f6revlerde kullan\u0131lmaktad\u0131r. \u00dcretici modeller metin i\u015fleme konusunda \u00e7ok say\u0131da probleme \u00e7\u00f6z\u00fcm \u00fcretmektedir. \u00d6rne\u011fin elinizdeki \u00fcretici modelinize bir \u015fiir veya bir hik\u00e2ye yazd\u0131rabilirsiniz. Veya kendi verileriniz i\u00e7in <a href=\"https:\/\/www.chatbase.co\/#demo\" target=\"_blank\" rel=\"noopener\">https:\/\/www.chatbase.co\/#demo<\/a> \u00fczerinden ChatGPT chatbot olu\u015fturabilirsiniz. GPT\u2019 ye birlikte bir \u015fiir yazd\u0131ral\u0131m:<\/p>\n<p><img decoding=\"async\" class=\"alignnone size-full wp-image-4772\" src=\"https:\/\/datakapital.com\/blog\/wp-content\/uploads\/2023\/09\/Chat-GPT-Siir.jpg\" alt=\"Chat GPT Metin \u00dcretme\" width=\"608\" height=\"479\" title=\"\" srcset=\"https:\/\/datakapital.com\/blog\/wp-content\/uploads\/2023\/09\/Chat-GPT-Siir.jpg 608w, https:\/\/datakapital.com\/blog\/wp-content\/uploads\/2023\/09\/Chat-GPT-Siir-300x236.jpg 300w, https:\/\/datakapital.com\/blog\/wp-content\/uploads\/2023\/09\/Chat-GPT-Siir-150x118.jpg 150w, https:\/\/datakapital.com\/blog\/wp-content\/uploads\/2023\/09\/Chat-GPT-Siir-450x355.jpg 450w\" sizes=\"(max-width: 608px) 100vw, 608px\" \/><\/p>\n<p>OpenAI GPT-2 [3] gibi baz\u0131 modelleri \u00fccretsiz eri\u015fime a\u00e7maktad\u0131r. GPT-2, b\u00fcy\u00fck bir dikkat mekanizmas\u0131 kullanarak \u00f6\u011frenir. Bu dikkat mekanizmas\u0131, modelin \u00f6nceki kelime ve c\u00fcmleleri anlamas\u0131na ve ard\u0131ndan bir sonraki kelimeyi \u00f6ng\u00f6rmesine olanak tan\u0131r. Model, milyonlarca parametre i\u00e7eren bir derin \u00f6\u011frenme a\u011f\u0131d\u0131r ve do\u011fal dil i\u015fleme alan\u0131nda \u00f6nemli bir ilerlemedir.<\/p>\n<p>GTP3.5 ve GPT4 gibi daha b\u00fcy\u00fck mimariye sahip geli\u015fmi\u015f modeller ise \u00fccret kar\u015f\u0131l\u0131\u011f\u0131nda SaaS olarak OpenAI taraf\u0131ndan sunmaktad\u0131r. GPT t\u00fcr\u00fc modellerde b\u00fcy\u00fckl\u00fck mimaride kullan\u0131lan \u00c7\u00f6z\u00fcc\u00fc say\u0131s\u0131 ile do\u011fru orant\u0131l\u0131d\u0131r.<\/p>\n<p><strong>Llma<\/strong><\/p>\n<p>Meta taraf\u0131ndan geli\u015ftirilmi\u015f bu b\u00fcy\u00fck dil modelinin en \u00f6nemli \u00f6zelli\u011fi MIT lisans\u0131 ile a\u00e7\u0131k kaynak kodlu olarak bireysel ve ticari kullan\u0131ma a\u00e7\u0131lmas\u0131d\u0131r. Llama2 Llama1&#8217;e g\u00f6re %40 daha fazla kaynakla beslenmi\u015ftir, 2 kat\u0131 daha fazla ba\u011flam uzunlu\u011funa sahiptir. Llma 2 modeli 18 Temmuz 2023\u2019 te kullan\u0131ma sunulmu\u015ftur. Bu modeli farkl\u0131 k\u0131lan bir di\u011fer \u00f6zellik g\u00fcvenlilik odakl\u0131 verilerle e\u011fitilmesidir. Bu model e\u011fitilirken 1 milyon\u2019a yak\u0131n insan geribildirimi verisi peki\u015ftirmeli \u00d6\u011frenme s\u00fcrecine dahil edilmi\u015ftir. Kendine \u00f6zg\u00fc hayalet (ghost) dikkat mekanizmas\u0131 kullanm\u0131\u015ft\u0131r. \u0130nsan de\u011ferlendirmesine g\u00f6re GPT 3\u2019ten daha iyi sonu\u00e7lar elde etmi\u015ftir fakat en b\u00fcy\u00fck k\u0131s\u0131t\u0131 neredeyse tamamen \u0130ngilizce dili i\u00e7in e\u011fitilmi\u015f olmas\u0131d\u0131r. Bu modelin T\u00fcrk\u00e7e deste\u011fi bulunmamaktad\u0131r. Modelden teknik bir konuya a\u00e7\u0131klama olu\u015fturmas\u0131n\u0131 istedim:<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" class=\"alignnone size-full wp-image-4773\" src=\"https:\/\/datakapital.com\/blog\/wp-content\/uploads\/2023\/09\/Lima-Dogal-Dil-Isleme.jpg\" alt=\"Lima Dil \u0130\u015fleme \" width=\"599\" height=\"259\" title=\"\" srcset=\"https:\/\/datakapital.com\/blog\/wp-content\/uploads\/2023\/09\/Lima-Dogal-Dil-Isleme.jpg 599w, https:\/\/datakapital.com\/blog\/wp-content\/uploads\/2023\/09\/Lima-Dogal-Dil-Isleme-300x130.jpg 300w, https:\/\/datakapital.com\/blog\/wp-content\/uploads\/2023\/09\/Lima-Dogal-Dil-Isleme-150x65.jpg 150w, https:\/\/datakapital.com\/blog\/wp-content\/uploads\/2023\/09\/Lima-Dogal-Dil-Isleme-450x195.jpg 450w\" sizes=\"(max-width: 599px) 100vw, 599px\" \/><\/p>\n<p><strong>Falcon<\/strong><\/p>\n<p>Abu Dabi&#8217;deki Teknoloji \u0130novasyon Enstit\u00fcs\u00fc (TII) taraf\u0131ndan geli\u015ftirilen yeni bir A\u00e7\u0131k Kaynak B\u00fcy\u00fck Dil Modeli ise Falcon\u2019dur. Falcon [4], Apache 2.0 lisans\u0131 alt\u0131nda piyasaya s\u00fcr\u00fclen ilk \u201cger\u00e7ekten a\u00e7\u0131k\u201d modeldir. Falcon, GPT-3&#8217;ten (Brown ve di\u011ferleri, 2020) uyarlanm\u0131\u015f, yaln\u0131zca kod \u00e7\u00f6z\u00fcc\u00fc i\u00e7eren bir modeldir ancak konumsal yerle\u015ftirmeler, dikkat (multiquery ve FlashAttention) ve kod \u00e7\u00f6z\u00fcc\u00fc blo\u011fu konusunda baz\u0131 mimari farkl\u0131l\u0131klara sahiptir. Falcon ailesinde Falcon-40B ve daha k\u00fc\u00e7\u00fck olan Falcon-7B olmak \u00fczere iki temel model bulunur. Modelin performans sonu\u00e7lar\u0131 LLM Leaderboard da di\u011fer a\u00e7\u0131k kaynak kodlu modeller aras\u0131nda en y\u00fcksek olarak payla\u015f\u0131lm\u0131\u015ft\u0131r.<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" class=\"alignnone size-full wp-image-4774\" src=\"https:\/\/datakapital.com\/blog\/wp-content\/uploads\/2023\/09\/Falcon-Buyuk-Dil-Modeli.png\" alt=\"Falcon Dil \u0130\u015fleme\" width=\"454\" height=\"190\" title=\"\" srcset=\"https:\/\/datakapital.com\/blog\/wp-content\/uploads\/2023\/09\/Falcon-Buyuk-Dil-Modeli.png 454w, https:\/\/datakapital.com\/blog\/wp-content\/uploads\/2023\/09\/Falcon-Buyuk-Dil-Modeli-300x126.png 300w, https:\/\/datakapital.com\/blog\/wp-content\/uploads\/2023\/09\/Falcon-Buyuk-Dil-Modeli-150x63.png 150w, https:\/\/datakapital.com\/blog\/wp-content\/uploads\/2023\/09\/Falcon-Buyuk-Dil-Modeli-450x188.png 450w\" sizes=\"(max-width: 454px) 100vw, 454px\" \/><\/p>\n<p>Bu yaz\u0131m\u0131zda b\u00fcy\u00fck dil modeli kavram\u0131na bir giri\u015f yapt\u0131k. Ayr\u0131ca yaz\u0131 i\u00e7erisinde en b\u00fcy\u00fck d\u00f6rt oyuncunun modelleri ile ilgili bilgiler verdik. Bir di\u011fer yaz\u0131m\u0131zda b\u00fcy\u00fck dil modellerinin alana \u00f6zel problemlerde nas\u0131l kullan\u0131labilece\u011finden \u00f6rneklerle bahsedece\u011fiz.<\/p>\n<p><strong>Referanslar<\/strong><\/p>\n<p>[1] Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., &#8230; &amp; Polosukhin, I. (2017). Attention is all you need. Advances in neural information processing systems, 30.<\/p>\n<p>[2] Devlin, J., Chang, M. W., Lee, K., &amp; Toutanova, K. (2018). Bert: Pre-training of deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805.<\/p>\n<p>[3] Radford, A., Wu, J., Child, R., Luan, D., Amodei, D., &amp; Sutskever, I. (2019). Language models are unsupervised multitask learners. OpenAI blog, 1(8), 9.<\/p>\n<p>[4] Almazrouei, E., Alobeidli, H., Alshamsi, A., Cappelli, A., Cojocaru, R., Debbah, M., &#8230; &amp; Penedo, G. (2023). Falcon-40B: an open large language model with state-of-the-art performance.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>Yapay zek\u00e2 alan\u0131nda ya\u015fanan h\u0131zl\u0131 geli\u015fmeler, b\u00fcy\u00fck dil modeli gibi \u00e7e\u015fitli alanlarda yeni i\u015f modelleri geli\u015ftirilmesinin \u00f6n\u00fcn\u00fc a\u00e7m\u0131\u015ft\u0131r. Yapay Zeka kronolojisine bakt\u0131\u011f\u0131m\u0131zda asl\u0131nda \u201cTransformer\u201d teknolojisi ile b\u00fcy\u00fck bir ba\u015flang\u0131ca ilk ad\u0131m\u0131 insanl\u0131k olarak att\u0131\u011f\u0131m\u0131z\u0131 s\u00f6yleyebiliriz. Peki nedir Transformerlar k\u0131saca hat\u0131rlayal\u0131m: Transformer, \u00f6zellikle do\u011fal dil i\u015fleme (NLP) ve yapay zek\u00e2 (AI) alanlar\u0131nda b\u00fcy\u00fck bir etki yaratm\u0131\u015f<\/p>\n","protected":false},"author":13,"featured_media":4771,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[6,38,41],"tags":[],"class_list":["post-4769","post","type-post","status-publish","format-standard","has-post-thumbnail","category-finansal-veri-okuryazarligi","category-python-ile-veri-isleme","category-veri-turleri-ve-kavramlar"],"better_featured_image":{"id":4771,"alt_text":"BERT Mimarisi Nas\u0131l \u00c7al\u0131\u015f\u0131r","caption":"","description":"","media_type":"image","media_details":{"width":605,"height":302,"file":"2023\/09\/BERT-Mimarisi.jpg","filesize":55849,"sizes":{"medium":{"file":"BERT-Mimarisi-300x150.jpg","width":300,"height":150,"mime-type":"image\/jpeg","filesize":10245,"source_url":"https:\/\/datakapital.com\/blog\/wp-content\/uploads\/2023\/09\/BERT-Mimarisi-300x150.jpg"},"thumbnail":{"file":"BERT-Mimarisi-150x150.jpg","width":150,"height":150,"mime-type":"image\/jpeg","filesize":5374,"source_url":"https:\/\/datakapital.com\/blog\/wp-content\/uploads\/2023\/09\/BERT-Mimarisi-150x150.jpg"},"bunyad-small":{"file":"BERT-Mimarisi-150x75.jpg","width":150,"height":75,"mime-type":"image\/jpeg","filesize":3363,"source_url":"https:\/\/datakapital.com\/blog\/wp-content\/uploads\/2023\/09\/BERT-Mimarisi-150x75.jpg"},"bunyad-medium":{"file":"BERT-Mimarisi-450x225.jpg","width":450,"height":225,"mime-type":"image\/jpeg","filesize":19601,"source_url":"https:\/\/datakapital.com\/blog\/wp-content\/uploads\/2023\/09\/BERT-Mimarisi-450x225.jpg"}},"image_meta":{"aperture":"0","credit":"","camera":"","caption":"","created_timestamp":"0","copyright":"","focal_length":"0","iso":"0","shutter_speed":"0","title":"","orientation":"0","keywords":[]}},"post":4769,"source_url":"https:\/\/datakapital.com\/blog\/wp-content\/uploads\/2023\/09\/BERT-Mimarisi.jpg"},"amp_enabled":true,"_links":{"self":[{"href":"https:\/\/datakapital.com\/blog\/wp-json\/wp\/v2\/posts\/4769","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/datakapital.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/datakapital.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/datakapital.com\/blog\/wp-json\/wp\/v2\/users\/13"}],"replies":[{"embeddable":true,"href":"https:\/\/datakapital.com\/blog\/wp-json\/wp\/v2\/comments?post=4769"}],"version-history":[{"count":4,"href":"https:\/\/datakapital.com\/blog\/wp-json\/wp\/v2\/posts\/4769\/revisions"}],"predecessor-version":[{"id":5025,"href":"https:\/\/datakapital.com\/blog\/wp-json\/wp\/v2\/posts\/4769\/revisions\/5025"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/datakapital.com\/blog\/wp-json\/wp\/v2\/media\/4771"}],"wp:attachment":[{"href":"https:\/\/datakapital.com\/blog\/wp-json\/wp\/v2\/media?parent=4769"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/datakapital.com\/blog\/wp-json\/wp\/v2\/categories?post=4769"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/datakapital.com\/blog\/wp-json\/wp\/v2\/tags?post=4769"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}