{"id":2302,"date":"2026-06-21T19:08:20","date_gmt":"2026-06-21T19:08:20","guid":{"rendered":"https:\/\/berenkudaygorun.com\/blog\/?p=2302"},"modified":"2026-06-21T19:08:20","modified_gmt":"2026-06-21T19:08:20","slug":"pytorch-derin-ogrenme-gun-9-10","status":"publish","type":"post","link":"https:\/\/berenkudaygorun.com\/blog\/blog\/2026\/06\/21\/pytorch-derin-ogrenme-gun-9-10\/","title":{"rendered":"Pytorch \u2013 Derin \u00d6\u011frenme G\u00fcn 9-10"},"content":{"rendered":"<p>G\u00fcn 9'da asl\u0131nda ileri seviye python konusu ile ilgili s\u0131nav vard\u0131 ama bunu payla\u015fmak ger\u00e7ekten istemedim biraz zaman kayb\u0131 olabilir diye. Sadece bundan dolay\u0131 9. g\u00fcnde ne oldu\u011funun bilgisini vermek ama\u00e7l\u0131 bu bilgiyi payla\u015fmak istedim. \u015eimdi 10. g\u00fcn ile devam edelim.<\/p>\n<p>Pytorch k\u00fct\u00fcphanesine ba\u015fl\u0131yoruz ve a\u00e7\u0131k\u00e7as\u0131 e\u011fitmen bu k\u00fct\u00fcphaneyi tenserflow ile kar\u015f\u0131la\u015ft\u0131rd\u0131. Bende de a\u00e7\u0131k\u00e7as\u0131 tenserflow ile ilgili bir kitap vard\u0131 daha \u00f6ncesinde de belirtti\u011fim gibi Derin \u00d6\u011frenme kurslar\u0131n ba\u015flam\u0131\u015ft\u0131m ama tamamlamam\u0131\u015ft\u0131 o kurslarda da tenserflow'dan bahsediyordu. Ancak e\u011fitmenin burada pytorch'u se\u00e7mesi ilk ba\u015fta biraz beni rahats\u0131z etmi\u015fti. Daha sonras\u0131nda k\u00fc\u00e7\u00fck bir ara\u015ft\u0131rma yapt\u0131\u011f\u0131mda pytorch asl\u0131nda daha iyi bir se\u00e7enmi\u015f. K\u0131sacas\u0131 chatgpt'nin dedikleri a\u015fa\u011f\u0131daki gibi...<\/p>\n<hr \/>\n<h1>PyTorch vs TensorFlow: Hangisini Se\u00e7erdim?<\/h1>\n<p>Bug\u00fcn s\u0131f\u0131rdan bir yapay zek\u00e2 projesi geli\u015ftirecek olsayd\u0131m \u00e7o\u011fu senaryoda <strong>PyTorch'u se\u00e7erdim<\/strong>. Bunun nedeni TensorFlow'un k\u00f6t\u00fc olmas\u0131 de\u011fil; son y\u0131llarda yapay zek\u00e2 ekosisteminin b\u00fcy\u00fck \u00f6l\u00e7\u00fcde PyTorch taraf\u0131na kaym\u0131\u015f olmas\u0131d\u0131r.<\/p>\n<h2>Kar\u015f\u0131la\u015ft\u0131rma<\/h2>\n<p>| Konu | PyTorch | TensorFlow |<\/p>\n<p>|--------|----------|------------|<\/p>\n<p>| \u00d6\u011frenme E\u011frisi | Daha kolay | Daha karma\u015f\u0131k |<\/p>\n<p>| Python Hissi | Daha do\u011fal | Daha framework odakl\u0131 |<\/p>\n<p>| Ara\u015ft\u0131rma D\u00fcnyas\u0131 | Lider | Daha az tercih ediliyor |<\/p>\n<p>| Production Kullan\u0131m\u0131 | \u00c7ok iyi | \u00c7ok iyi |<\/p>\n<p>| Debug S\u00fcreci | Daha kolay | Daha zor |<\/p>\n<p>| Dok\u00fcmantasyon ve \u00d6rnekler | \u00c7ok fazla | Fazla ancak eski \u00f6rnekler de bulunuyor |<\/p>\n<p>| LLM Ekosistemi | A\u00e7\u0131k ara lider | Geride |<\/p>\n<p>| Hugging Face Deste\u011fi | Birinci s\u0131n\u0131f | \u0130kinci planda |<\/p>\n<h2>PyTorch'un G\u00fc\u00e7l\u00fc Oldu\u011fu Alanlar<\/h2>\n<p>A\u015fa\u011f\u0131daki alanlarda \u00e7al\u0131\u015fmay\u0131 planl\u0131yorsan\u0131z PyTorch g\u00fcn\u00fcm\u00fczde en yayg\u0131n tercih edilen \u00e7\u00f6z\u00fcmd\u00fcr:<\/p>\n<ul>\n<li>\n<p>B\u00fcy\u00fck Dil Modelleri (LLM)<\/p>\n<\/li>\n<li>\n<p>RAG (Retrieval-Augmented Generation) sistemleri<\/p>\n<\/li>\n<li>\n<p>Fine-tuning \u00e7al\u0131\u015fmalar\u0131<\/p>\n<\/li>\n<li>\n<p>Stable Diffusion ve g\u00f6r\u00fcnt\u00fc \u00fcretimi<\/p>\n<\/li>\n<li>\n<p>Yapay zek\u00e2 ajanlar\u0131<\/p>\n<\/li>\n<li>\n<p>Computer Vision projeleri<\/p>\n<\/li>\n<li>\n<p>Derin \u00f6\u011frenme ara\u015ft\u0131rmalar\u0131<\/p>\n<\/li>\n<\/ul>\n<p>\u00d6rne\u011fin a\u015fa\u011f\u0131daki pop\u00fcler ekosistemlerin tamam\u0131 PyTorch merkezli \u00e7al\u0131\u015fmaktad\u0131r:<\/p>\n<ul>\n<li>\n<p>PyTorch<\/p>\n<\/li>\n<li>\n<p>Hugging Face Transformers<\/p>\n<\/li>\n<li>\n<p>Meta Llama<\/p>\n<\/li>\n<\/ul>\n<hr \/>\n<p>Tamam bu bilgiler olduk\u00e7a yeterli ve ikna ediciydi. \u015eimdi PyTorch ile e\u011fitimin 10. g\u00fcnde bahsedilen scaler, vektor, matris ve tensor kavramlar\u0131na de\u011finelim.<\/p>\n<p>E\u011fitmenin dedi\u011fine g\u00f6re asl\u0131nda pythorch hepsini birer tensor olarka kabul ediyormu\u015f. Numpy'da yapt\u0131\u011f\u0131m\u0131z konulara olduk\u00e7a benzer olacak \u015fekilde burada \u00e7e\u015fitli i\u015flemler yapt\u0131k.<\/p>\n<p>\u0130lk \u00f6rned\u011fimizde bir scaler olu\u015ftural\u0131m yani asl\u0131nda bir sabit yani 0 boyutlu:<\/p>\n<pre><code>import torch\n\nscaler = torch.tensor(5)\nprint(scaler)\nprint(scaler.ndim)\nprint(scaler.shape)<\/code><\/pre>\n<p>\u00c7\u0131kt\u0131s\u0131 a\u015fa\u011f\u0131daki gibi olacakt\u0131r.<\/p>\n<pre><code>tensor(5)\n0\ntorch.Size([])<\/code><\/pre>\n<p>\u015eimdi ise tek boyutlu bir dizi yani vekt\u00f6r olu\u015ftural\u0131m.<\/p>\n<pre><code>vektor = torch.tensor([1,2,3])\nprint(vektor)\nprint(vektor.ndim)\nprint(vektor.shape)<\/code><\/pre>\n<p>\u0130\u015fte \u00e7\u0131kt\u0131s\u0131:<\/p>\n<pre><code>tensor([1, 2, 3])\n1\ntorch.Size([3])<\/code><\/pre>\n<p>E\u011fitmenin dedi\u011fine g\u00f6re asl\u0131nda bir tens\u00f6r\u00fcn ka\u00e7 boyutlu oldu\u011funu anlamak isterseniz d\u0131\u015fardan i\u00e7eriye k\u00f6\u015feli parantezleri sayabilirsiniz dedi. G\u00fczel bir taktik oalbillir kod okurken... \u015eimdi matris tan\u0131mlayal\u0131m.<\/p>\n<pre><code>matris = torch.tensor([[1,2,3],[2,3,4]])\nprint(matris)\nprint(matris.ndim)\nprint(matris.shape)<\/code><\/pre>\n<p>\u0130\u015fte \u00e7\u0131kt\u0131s\u0131:<\/p>\n<pre><code>tensor([[1, 2, 3],\n        [2, 3, 4]])\n2\ntorch.Size([2, 3])<\/code><\/pre>\n<p>\u015eimdi ise 3 boyutlu bir matris tan\u0131mlayal\u0131m. Mesela 10px'lik kare bir ekran\u0131m\u0131z olsun. buradaki pikseller RGB olacaklar do\u011fal olarak her pikselde asl\u0131nda yanen 3 tane renk var. Bu renkleri bir vekt\u00f6r gibi d\u00fc\u015f\u00fcnebiliriz ama bu vekt\u00f6rlerin birle\u015fiminden 3 boyutlu bir yap\u0131 ortaya \u00e7\u0131kacak. \u00d6rne\u011fin:<\/p>\n<pre><code>image = torch.Tensor(torch.randint(0,255,(3,10,10)))\nprint(image)\nprint(image.ndim)\nprint(image.shape)<\/code><\/pre>\n<p>\u00c7\u0131kt\u0131 bu \u015fekilde olacak.<\/p>\n<pre><code>tensor([[[249, 215, 195, 166, 116, 215, 113, 173, 135,  38],\n         [238, 205, 193, 172,  86,  88,  78, 167, 109, 125],\n         [ 17,  36,  26, 251, 154, 220, 100,  82,  59, 172],\n         [ 87, 147,  36, 189,  16, 115, 123, 143,  14, 162],\n         [234, 182, 127, 206,  36, 184, 125, 113, 248, 111],\n         [ 81, 148, 148,   4,  13, 170, 246, 192,  54, 125],\n         [ 74, 110, 187, 133, 251, 240,  96, 173, 184, 174],\n         [104,  19, 222,  18,  18, 148, 104, 205,  11, 182],\n         [ 14,  46,  66,  60, 198,  17, 121, 114, 197, 178],\n         [ 33, 184, 127, 121, 196,  37, 105, 105, 139, 236]],\n\n        [[ 18, 226, 230,  13, 240, 201, 234,  16, 232, 101],\n         [215, 226,  78, 113,  67, 174,  54,  16, 230, 249],\n         [ 90, 254, 137,  40, 204, 221, 252, 175,  74, 135],\n         [217,  18, 194, 243, 157, 241, 195, 138, 244, 212],\n         [228, 192,  10,  65,  14,  34,  90, 236,  98, 143],\n         [152, 222, 180,  61, 236,  49,  18, 116,  25,  29],\n         [ 23,  85,  33, 243, 227, 249, 152, 190, 189, 156],\n         [157,  98,  31, 120,  10, 189,  61,  85, 180,   4],\n         [234,  72,  83, 134, 217, 146,   3,  62, 166,  68],\n         [130,  44, 220,  74, 153, 232, 191, 251, 196,  48]],\n\n        [[ 87,  69, 112, 189, 145, 169, 212,  72, 103, 240],\n         [133, 111,  22, 241, 114,  63, 156, 151,  56,   4],\n         [122,   3,  17, 209, 200, 244, 145, 114,  20,  47],\n         [  5,   7, 131, 208,  29, 250, 123,  77, 221, 184],\n         [  2,  45,  11, 140, 236,  89, 165,  42, 220, 173],\n         [104, 141,  17, 143,  15, 248, 221,  45, 248,  79],\n         [156,  10, 222, 136, 192,  94, 183, 154, 154,  82],\n         [ 89, 237, 216,  94,  88,  60, 126,  54,   5, 201],\n         [146,  50,  75, 242, 227, 135,  96, 253,  64, 113],\n         [ 96, 137, 241, 184, 164, 148, 153, 186, 150,  40]]])\n3\ntorch.Size([3, 10, 10])<\/code><\/pre>\n","protected":false},"excerpt":{"rendered":"<p>G\u00fcn 9&#8217;da asl\u0131nda ileri seviye python konusu ile ilgili s\u0131nav vard\u0131 ama bunu payla\u015fmak ger\u00e7ekten istemedim biraz zaman kayb\u0131 olabilir diye. Sadece bundan dolay\u0131 9.&#8230;<\/p>\n<div class=\"more-link-wrapper\"><a class=\"more-link\" href=\"https:\/\/berenkudaygorun.com\/blog\/blog\/2026\/06\/21\/pytorch-derin-ogrenme-gun-9-10\/\">Devam\u0131n\u0131 oku<span class=\"screen-reader-text\">Pytorch \u2013 Derin \u00d6\u011frenme G\u00fcn 9-10<\/span><\/a><\/div>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[708],"tags":[739],"class_list":["post-2302","post","type-post","status-publish","format-standard","hentry","category-derin-ogrenme","tag-pytorch","entry"],"_links":{"self":[{"href":"https:\/\/berenkudaygorun.com\/blog\/wp-json\/wp\/v2\/posts\/2302","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/berenkudaygorun.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/berenkudaygorun.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/berenkudaygorun.com\/blog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/berenkudaygorun.com\/blog\/wp-json\/wp\/v2\/comments?post=2302"}],"version-history":[{"count":1,"href":"https:\/\/berenkudaygorun.com\/blog\/wp-json\/wp\/v2\/posts\/2302\/revisions"}],"predecessor-version":[{"id":2303,"href":"https:\/\/berenkudaygorun.com\/blog\/wp-json\/wp\/v2\/posts\/2302\/revisions\/2303"}],"wp:attachment":[{"href":"https:\/\/berenkudaygorun.com\/blog\/wp-json\/wp\/v2\/media?parent=2302"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/berenkudaygorun.com\/blog\/wp-json\/wp\/v2\/categories?post=2302"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/berenkudaygorun.com\/blog\/wp-json\/wp\/v2\/tags?post=2302"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}