Rebuild

Author: 9bow

docs/_downloads/03a48646520c277662581e858e680809/model_parallel_tutorial.ipynb

@@ -123,7 +123,7 @@
       },
       "outputs": [],
       "source": [
-        "# \uc2e4\ud5d8 \uacb0\uacfc, \ubaa8\ub378 \ubcd1\ub82c \ucca0\ub9ac\ud558\uc5ec \ud559\uc2b5\ud558\ub294 \uc2dc\uac04\uc774 \ub2e8\uc77c GPU\ub85c \ud559\uc2b5\ud558\ub294 \uc2dc\uac04\ubcf4\ub2e4 \uc57d 7% ``4.02/3.75-1=7%``\uc815\ub3c4\n# \uc624\ub798 \uac78\ub9ac\ub294 \uac83\uc744 \ud655\uc778\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uadf8\ub7ec\ubbc0\ub85c, \uc21c\uc804\ud30c\uc640 \uc5ed\uc804\ud30c\ub97c \uc9c4\ud589\ud558\uba74\uc11c GPU \uac04 \ud150\uc11c\uac12\ub4e4\uc774\n# \ubcf5\uc81c\ub418\uc5b4 \uc774\uc6a9\ud558\ub294 \uc2dc\uac04\uc774 \uc57d 7%\uc815\ub3c4 \uc18c\uc694\ub418\ub294 \uac83\uc73c\ub85c \uacb0\ub860\uc9c0\uc744 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \ud559\uc2b5\ud558\ub294 \uacfc\uc815 \uc18d\uc5d0\uc11c\n# 2\uac1c\uc758 GPU \uc911 1\uac1c\uc758 GPU\uac00 \uacc4\uc0b0\ud558\uc9c0 \uc54a\uace0 \ub300\uae30\ud558\uace0 \uc788\uae30 \ub54c\ubb38\uc5d0, \uc774\ub97c \ud574\uacb0\ud558\uc5ec\n# \ud559\uc2b5 \uc2dc\uac04\uc744 \ube60\ub974\uac8c \uac1c\uc120\uc2dc\ud0ac \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uadf8 \uc911 \ud55c \uac00\uc9c0 \ubc29\ubc95\uc740, \ud559\uc2b5 \ub2e8\uc704\uc778 \ubbf8\ub2c8 \ubc30\uce58 1\uac1c\uc758 \ub370\uc774\ud130\ub97c\n# 2\uac1c\ub85c \ubd84\ud560\ud558\ub294 \ud30c\uc774\ud504\ub77c\uc778\uc744 \uc0dd\uc131\ud558\uc5ec, \ubd84\ud560\ub41c \uccab \ubc88\uc9f8 \ub370\uc774\ud130\uac00 \uccab \ubc88\uc9f8 \uce35\uc744 \ud1b5\uacfc\ud558\uc5ec \ub450 \ubc88\uc9f8 \uce35\uc73c\ub85c\n# \ubcf5\uc81c\ub418\uace0, \ub450 \ubc88\uc9f8 \uce35\uc744 \ud1b5\uacfc\ud560 \ub54c, \ub450\ubc88\uc7ac\ub85c \ubd84\ud560\ub41c \ub370\uc774\ud130\uac00 \uccab \ubc88\uca30 \uce35\uc744 \ud1b5\ud574 \uacc4\uc0b0\ub418\ub294 \ubc29\uc2dd\uc73c\ub85c \uc124\uc815\ud558\ub294 \uac83\uc785\ub2c8\ub2e4.\n# \uc774\ub7ec\ud55c \ubc29\ubc95\uc744 \ud1b5\ud574\uc11c 2\uac1c\uc758 GPU\uac00 2\uac1c\ub85c \ubd84\ud560\ub41c \ub370\uc774\ud130\ub97c \ub3d9\uc2dc\uc5d0 \ucc98\ub9ac\ud560 \uc218 \uc788\uc73c\uba70 \ud559\uc2b5 \uc2dc\uac04\uc744 \ub2e8\ucd95\uc2dc\ud0ac \uc218 \uc788\uc2b5\ub2c8\ub2e4."
+        "# \uc2e4\ud5d8 \uacb0\uacfc, \ubaa8\ub378 \ubcd1\ub82c \ucc98\ub9ac\ud558\uc5ec \ud559\uc2b5\ud558\ub294 \uc2dc\uac04\uc774 \ub2e8\uc77c GPU\ub85c \ud559\uc2b5\ud558\ub294 \uc2dc\uac04\ubcf4\ub2e4 \uc57d 7% ``4.02/3.75-1=7%``\uc815\ub3c4\n# \uc624\ub798 \uac78\ub9ac\ub294 \uac83\uc744 \ud655\uc778\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uadf8\ub7ec\ubbc0\ub85c, \uc21c\uc804\ud30c\uc640 \uc5ed\uc804\ud30c\ub97c \uc9c4\ud589\ud558\uba74\uc11c GPU \uac04 \ud150\uc11c\uac12\ub4e4\uc774\n# \ubcf5\uc81c\ub418\uc5b4 \uc774\uc6a9\ud558\ub294 \uc2dc\uac04\uc774 \uc57d 7%\uc815\ub3c4 \uc18c\uc694\ub418\ub294 \uac83\uc73c\ub85c \uacb0\ub860\uc9c0\uc744 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \ud559\uc2b5\ud558\ub294 \uacfc\uc815 \uc18d\uc5d0\uc11c\n# 2\uac1c\uc758 GPU \uc911 1\uac1c\uc758 GPU\uac00 \uacc4\uc0b0\ud558\uc9c0 \uc54a\uace0 \ub300\uae30\ud558\uace0 \uc788\uae30 \ub54c\ubb38\uc5d0, \uc774\ub97c \ud574\uacb0\ud558\uc5ec\n# \ud559\uc2b5 \uc2dc\uac04\uc744 \ube60\ub974\uac8c \uac1c\uc120\uc2dc\ud0ac \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uadf8 \uc911 \ud55c \uac00\uc9c0 \ubc29\ubc95\uc740, \ud559\uc2b5 \ub2e8\uc704\uc778 \ubbf8\ub2c8 \ubc30\uce58 1\uac1c\uc758 \ub370\uc774\ud130\ub97c\n# 2\uac1c\ub85c \ubd84\ud560\ud558\ub294 \ud30c\uc774\ud504\ub77c\uc778\uc744 \uc0dd\uc131\ud558\uc5ec, \ubd84\ud560\ub41c \uccab \ubc88\uc9f8 \ub370\uc774\ud130\uac00 \uccab \ubc88\uc9f8 \uce35\uc744 \ud1b5\uacfc\ud558\uc5ec \ub450 \ubc88\uc9f8 \uce35\uc73c\ub85c\n# \ubcf5\uc81c\ub418\uace0, \ub450 \ubc88\uc9f8 \uce35\uc744 \ud1b5\uacfc\ud560 \ub54c, \ub450\ubc88\uc7ac\ub85c \ubd84\ud560\ub41c \ub370\uc774\ud130\uac00 \uccab \ubc88\uca30 \uce35\uc744 \ud1b5\ud574 \uacc4\uc0b0\ub418\ub294 \ubc29\uc2dd\uc73c\ub85c \uc124\uc815\ud558\ub294 \uac83\uc785\ub2c8\ub2e4.\n# \uc774\ub7ec\ud55c \ubc29\ubc95\uc744 \ud1b5\ud574\uc11c 2\uac1c\uc758 GPU\uac00 2\uac1c\ub85c \ubd84\ud560\ub41c \ub370\uc774\ud130\ub97c \ub3d9\uc2dc\uc5d0 \ucc98\ub9ac\ud560 \uc218 \uc788\uc73c\uba70 \ud559\uc2b5 \uc2dc\uac04\uc744 \ub2e8\ucd95\uc2dc\ud0ac \uc218 \uc788\uc2b5\ub2c8\ub2e4."
       ]
     },
     {

docs/_downloads/1b58d206e701317cf46c92dcf2a8978d/parallelism_tutorial.ipynb

@@ -33,7 +33,7 @@
       "cell_type": "markdown",
       "metadata": {},
       "source": [
-        "CPU \ubaa8\ub4dc\uc778 \ucf54\ub4dc\ub97c \ubc14\uafc0 \ud544\uc694\uac00 \uc5c6\uc2b5\ub2c8\ub2e4.\n\nDataParallel\uc5d0 \ub300\ud55c \ubb38\uc11c\ub294 [\uc5ec\uae30](https://pytorch.org/docs/stable/nn.html#dataparallel-layers-multi-gpu-distributed)\n\uc5d0\uc11c \ud655\uc778\ud558\uc2e4 \uc218 \uc788\uc2b5\ub2c8\ub2e4.\n\n**\ub798\ud551\ub41c \ubaa8\ub4c8\uc758 \uc18d\uc131**\n\n\ubaa8\ub4c8\uc744 ``DataParallel`` \ub85c \uac10\uc2fc \ud6c4\uc5d0\ub294 \ubaa8\ub4c8\uc758 \uc18d\uc131(\uc608. \uc0ac\uc6a9\uc790 \uc815\uc758 \uba54\uc18c\ub4dc)\uc5d0\n\uc811\uadfc\ud560 \uc218 \uc5c6\uac8c \ub429\ub2c8\ub2e4. \uc774\ub294 ``DataParallel`` \uc774 \uba87\uba87 \uc0c8\ub85c\uc6b4 \uba64\ubc84\ub97c \uc815\uc758\ud558\uae30 \ub584\ubb38\uc5d0\n\ub2e4\ub978 \uc18d\uc131\uc5d0 \uc811\uadfc\uc744 \ud5c8\uc6a9\ud558\ub294 \uac83\uc774 \ucda9\ub3cc\uc744 \uc77c\uc73c\ud0ac \uc218\ub3c4 \uc788\uae30 \ub54c\ubb38\uc785\ub2c8\ub2e4.\n\uadf8\ub798\ub3c4 \uc18d\uc131\uc5d0 \uc811\uadfc\ud558\uace0\uc790 \ud55c\ub2e4\uba74 \uc544\ub798\uc640 \uac19\uc774 ``DataParallel`` \uc758 \uc11c\ube0c\ud074\ub798\uc2a4\ub97c\n\uc0ac\uc6a9\ud558\ub294 \uac83\uc774 \uc88b\uc2b5\ub2c8\ub2e4.\n\n"
+        "CPU \ubaa8\ub4dc\uc778 \ucf54\ub4dc\ub97c \ubc14\uafc0 \ud544\uc694\uac00 \uc5c6\uc2b5\ub2c8\ub2e4.\n\nDataParallel\uc5d0 \ub300\ud55c \ubb38\uc11c\ub294 [\uc5ec\uae30](https://pytorch.org/docs/stable/nn.html#dataparallel-layers-multi-gpu-distributed)\n\uc5d0\uc11c \ud655\uc778\ud558\uc2e4 \uc218 \uc788\uc2b5\ub2c8\ub2e4.\n\n**\ub798\ud551\ub41c \ubaa8\ub4c8\uc758 \uc18d\uc131**\n\n\ubaa8\ub4c8\uc744 ``DataParallel`` \ub85c \uac10\uc2fc \ud6c4\uc5d0\ub294 \ubaa8\ub4c8\uc758 \uc18d\uc131(\uc608. \uc0ac\uc6a9\uc790 \uc815\uc758 \uba54\uc18c\ub4dc)\uc5d0\n\uc811\uadfc\ud560 \uc218 \uc5c6\uac8c \ub429\ub2c8\ub2e4. \uc774\ub294 ``DataParallel`` \uc774 \uba87\uba87 \uc0c8\ub85c\uc6b4 \uba64\ubc84\ub97c \uc815\uc758\ud558\uae30 \ub54c\ubb38\uc5d0\n\ub2e4\ub978 \uc18d\uc131\uc5d0 \uc811\uadfc\uc744 \ud5c8\uc6a9\ud558\ub294 \uac83\uc774 \ucda9\ub3cc\uc744 \uc77c\uc73c\ud0ac \uc218\ub3c4 \uc788\uae30 \ub54c\ubb38\uc785\ub2c8\ub2e4.\n\uadf8\ub798\ub3c4 \uc18d\uc131\uc5d0 \uc811\uadfc\ud558\uace0\uc790 \ud55c\ub2e4\uba74 \uc544\ub798\uc640 \uac19\uc774 ``DataParallel`` \uc758 \uc11c\ube0c\ud074\ub798\uc2a4\ub97c\n\uc0ac\uc6a9\ud558\ub294 \uac83\uc774 \uc88b\uc2b5\ub2c8\ub2e4.\n\n"
       ]
     },
     {

docs/_downloads/26d49aeeb9c710e27197fda28b7c3516/yi_jing_01_chien.jpg

@@ -55,7 +55,7 @@
       },
       "outputs": [],
       "source": [
-        "CONTEXT_SIZE = 2\nEMBEDDING_DIM = 10\n# \uc170\uc775\uc2a4\ud53c\uc5b4 \uc18c\ub124\ud2b8(Sonnet) 2\ub97c \uc0ac\uc6a9\ud558\uaca0\uc2b5\ub2c8\ub2e4.\ntest_sentence = \"\"\"When forty winters shall besiege thy brow,\nAnd dig deep trenches in thy beauty's field,\nThy youth's proud livery so gazed on now,\nWill be a totter'd weed of small worth held:\nThen being asked, where all thy beauty lies,\nWhere all the treasure of thy lusty days;\nTo say, within thine own deep sunken eyes,\nWere an all-eating shame, and thriftless praise.\nHow much more praise deserv'd thy beauty's use,\nIf thou couldst answer 'This fair child of mine\nShall sum my count, and make my old excuse,'\nProving his beauty by succession thine!\nThis were to be new made when thou art old,\nAnd see thy blood warm when thou feel'st it cold.\"\"\".split()\n# \uc6d0\ub798\ub294 \uc785\ub825\uc744 \uc81c\ub300\ub85c \ud1a0\ud070\ud654(tokenize) \ud574\uc57c\ud558\uc9c0\ub9cc \uc774\ubc88\uc5d4 \uac04\uc18c\ud654\ud558\uc5ec \uc9c4\ud589\ud558\uaca0\uc2b5\ub2c8\ub2e4.\n# \ud29c\ud50c\ub85c \uc774\ub8e8\uc5b4\uc9c4 \ub9ac\uc2a4\ud2b8\ub97c \ub9cc\ub4e4\uaca0\uc2b5\ub2c8\ub2e4. \uac01 \ud29c\ud50c\uc740 ([ i-CONTEXT_SIZE \ubc88\uc9f8 \ub2e8\uc5b4, ..., i-1 \ubc88\uc9f8 \ub2e8\uc5b4 ], \ubaa9\ud45c \ub2e8\uc5b4)\uc785\ub2c8\ub2e4.\nngrams = [\n    (\n        [test_sentence[i - j - 1] for j in range(CONTEXT_SIZE)],\n        test_sentence[i]\n    )\n    for i in range(CONTEXT_SIZE, len(test_sentence))\n]\n# \uccab 3\uac1c\uc758 \ud29c\ud50c\uc744 \ucd9c\ub825\ud558\uc5ec \ub370\uc774\ud130\uac00 \uc5b4\ub5bb\uac8c \uc0dd\uacbc\ub294\uc9c0 \ubcf4\uaca0\uc2b5\ub2c8\ub2e4.\nprint(ngrams[:3])\n\nvocab = set(test_sentence)\nword_to_ix = {word: i for i, word in enumerate(vocab)}\n\n\nclass NGramLanguageModeler(nn.Module):\n\n    def __init__(self, vocab_size, embedding_dim, context_size):\n        super(NGramLanguageModeler, self).__init__()\n        self.embeddings = nn.Embedding(vocab_size, embedding_dim)\n        self.linear1 = nn.Linear(context_size * embedding_dim, 128)\n        self.linear2 = nn.Linear(128, vocab_size)\n\n    def forward(self, inputs):\n        embeds = self.embeddings(inputs).view((1, -1))\n        out = F.relu(self.linear1(embeds))\n        out = self.linear2(out)\n        log_probs = F.log_softmax(out, dim=1)\n        return log_probs\n\n\nlosses = []\nloss_function = nn.NLLLoss()\nmodel = NGramLanguageModeler(len(vocab), EMBEDDING_DIM, CONTEXT_SIZE)\noptimizer = optim.SGD(model.parameters(), lr=0.001)\n\nfor epoch in range(10):\n    total_loss = 0\n    for context, target in ngrams:\n\n        # \uccab\ubc88\uc9f8. \ubaa8\ub378\uc5d0 \ub123\uc5b4\uc904 \uc785\ub825\uac12\uc744 \uc900\ube44\ud569\ub2c8\ub2e4. (i.e, \ub2e8\uc5b4\ub97c \uc815\uc218 \uc778\ub371\uc2a4\ub85c\n        # \ubc14\uafb8\uace0 \ud30c\uc774\ud1a0\uce58 \ud150\uc11c\ub85c \uac10\uc2f8\uc90d\uc2dc\ub2e4.)\n        context_idxs = torch.tensor([word_to_ix[w] for w in context], dtype=torch.long)\n\n        # \ub450\ubc88\uc9f8. \ud1a0\uce58\ub294 \uae30\uc6b8\uae30\uac00 *\ub204\uc801* \ub429\ub2c8\ub2e4. \uc0c8 \uc778\uc2a4\ud134\uc2a4\ub97c \ub123\uc5b4\uc8fc\uae30 \uc804\uc5d0\n        # \uae30\uc6b8\uae30\ub97c \ucd08\uae30\ud654\ud569\ub2c8\ub2e4.\n        model.zero_grad()\n\n        # \uc138\ubc88\uc9f8. \uc21c\uc804\ud30c\ub97c \ud1b5\ud574 \ub2e4\uc74c\uc5d0 \uc62c \ub2e8\uc5b4\uc5d0 \ub300\ud55c \ub85c\uadf8 \ud655\ub960\uc744 \uad6c\ud569\ub2c8\ub2e4.\n        log_probs = model(context_idxs)\n\n        # \ub124\ubc88\uc9f8. \uc190\uc2e4\ud568\uc218\ub97c \uacc4\uc0b0\ud569\ub2c8\ub2e4. (\ud30c\uc774\ud1a0\uce58\uc5d0\uc11c\ub294 \ubaa9\ud45c \ub2e8\uc5b4\ub97c \ud150\uc11c\ub85c \uac10\uc2f8\uc918\uc57c \ud569\ub2c8\ub2e4.)\n        loss = loss_function(log_probs, torch.tensor([word_to_ix[target]], dtype=torch.long))\n\n        # \ub2e4\uc12f\ubc88\uc9f8. \uc5ed\uc804\ud30c\ub97c \ud1b5\ud574 \uae30\uc6b8\uae30\ub97c \uc5c5\ub370\uc774\ud2b8 \ud574\uc90d\ub2c8\ub2e4.\n        loss.backward()\n        optimizer.step()\n\n        # tensor.item()\uc744 \ud638\ucd9c\ud558\uc5ec \ub2e8\uc77c\uc6d0\uc18c \ud150\uc11c\uc5d0\uc11c \uc22b\uc790\ub97c \ubc18\ud658\ubc1b\uc2b5\ub2c8\ub2e4.\n        total_loss += loss.item()\n    losses.append(total_loss)\nprint(losses)  # \ubc18\ubcf5\ud560 \ub584\ub9c8\ub2e4 \uc190\uc2e4\uc774 \uc904\uc5b4\ub4dc\ub294 \uac83\uc744 \ubd05\uc2dc\ub2e4!\n\n# \"beauty\"\uc640 \uac19\uc774 \ud2b9\uc815 \ub2e8\uc5b4\uc5d0 \ub300\ud55c \uc784\ubca0\ub529\uc744 \ud655\uc778\ud558\ub824\uba74,\nprint(model.embeddings.weight[word_to_ix[\"beauty\"]])"
+        "CONTEXT_SIZE = 2\nEMBEDDING_DIM = 10\n# \uc170\uc775\uc2a4\ud53c\uc5b4 \uc18c\ub124\ud2b8(Sonnet) 2\ub97c \uc0ac\uc6a9\ud558\uaca0\uc2b5\ub2c8\ub2e4.\ntest_sentence = \"\"\"When forty winters shall besiege thy brow,\nAnd dig deep trenches in thy beauty's field,\nThy youth's proud livery so gazed on now,\nWill be a totter'd weed of small worth held:\nThen being asked, where all thy beauty lies,\nWhere all the treasure of thy lusty days;\nTo say, within thine own deep sunken eyes,\nWere an all-eating shame, and thriftless praise.\nHow much more praise deserv'd thy beauty's use,\nIf thou couldst answer 'This fair child of mine\nShall sum my count, and make my old excuse,'\nProving his beauty by succession thine!\nThis were to be new made when thou art old,\nAnd see thy blood warm when thou feel'st it cold.\"\"\".split()\n# \uc6d0\ub798\ub294 \uc785\ub825\uc744 \uc81c\ub300\ub85c \ud1a0\ud070\ud654(tokenize) \ud574\uc57c\ud558\uc9c0\ub9cc \uc774\ubc88\uc5d4 \uac04\uc18c\ud654\ud558\uc5ec \uc9c4\ud589\ud558\uaca0\uc2b5\ub2c8\ub2e4.\n# \ud29c\ud50c\ub85c \uc774\ub8e8\uc5b4\uc9c4 \ub9ac\uc2a4\ud2b8\ub97c \ub9cc\ub4e4\uaca0\uc2b5\ub2c8\ub2e4. \uac01 \ud29c\ud50c\uc740 ([ i-CONTEXT_SIZE \ubc88\uc9f8 \ub2e8\uc5b4, ..., i-1 \ubc88\uc9f8 \ub2e8\uc5b4 ], \ubaa9\ud45c \ub2e8\uc5b4)\uc785\ub2c8\ub2e4.\nngrams = [\n    (\n        [test_sentence[i - j - 1] for j in range(CONTEXT_SIZE)],\n        test_sentence[i]\n    )\n    for i in range(CONTEXT_SIZE, len(test_sentence))\n]\n# \uccab 3\uac1c\uc758 \ud29c\ud50c\uc744 \ucd9c\ub825\ud558\uc5ec \ub370\uc774\ud130\uac00 \uc5b4\ub5bb\uac8c \uc0dd\uacbc\ub294\uc9c0 \ubcf4\uaca0\uc2b5\ub2c8\ub2e4.\nprint(ngrams[:3])\n\nvocab = set(test_sentence)\nword_to_ix = {word: i for i, word in enumerate(vocab)}\n\n\nclass NGramLanguageModeler(nn.Module):\n\n    def __init__(self, vocab_size, embedding_dim, context_size):\n        super(NGramLanguageModeler, self).__init__()\n        self.embeddings = nn.Embedding(vocab_size, embedding_dim)\n        self.linear1 = nn.Linear(context_size * embedding_dim, 128)\n        self.linear2 = nn.Linear(128, vocab_size)\n\n    def forward(self, inputs):\n        embeds = self.embeddings(inputs).view((1, -1))\n        out = F.relu(self.linear1(embeds))\n        out = self.linear2(out)\n        log_probs = F.log_softmax(out, dim=1)\n        return log_probs\n\n\nlosses = []\nloss_function = nn.NLLLoss()\nmodel = NGramLanguageModeler(len(vocab), EMBEDDING_DIM, CONTEXT_SIZE)\noptimizer = optim.SGD(model.parameters(), lr=0.001)\n\nfor epoch in range(10):\n    total_loss = 0\n    for context, target in ngrams:\n\n        # \uccab\ubc88\uc9f8. \ubaa8\ub378\uc5d0 \ub123\uc5b4\uc904 \uc785\ub825\uac12\uc744 \uc900\ube44\ud569\ub2c8\ub2e4. (i.e, \ub2e8\uc5b4\ub97c \uc815\uc218 \uc778\ub371\uc2a4\ub85c\n        # \ubc14\uafb8\uace0 \ud30c\uc774\ud1a0\uce58 \ud150\uc11c\ub85c \uac10\uc2f8\uc90d\uc2dc\ub2e4.)\n        context_idxs = torch.tensor([word_to_ix[w] for w in context], dtype=torch.long)\n\n        # \ub450\ubc88\uc9f8. \ud1a0\uce58\ub294 \uae30\uc6b8\uae30\uac00 *\ub204\uc801* \ub429\ub2c8\ub2e4. \uc0c8 \uc778\uc2a4\ud134\uc2a4\ub97c \ub123\uc5b4\uc8fc\uae30 \uc804\uc5d0\n        # \uae30\uc6b8\uae30\ub97c \ucd08\uae30\ud654\ud569\ub2c8\ub2e4.\n        model.zero_grad()\n\n        # \uc138\ubc88\uc9f8. \uc21c\uc804\ud30c\ub97c \ud1b5\ud574 \ub2e4\uc74c\uc5d0 \uc62c \ub2e8\uc5b4\uc5d0 \ub300\ud55c \ub85c\uadf8 \ud655\ub960\uc744 \uad6c\ud569\ub2c8\ub2e4.\n        log_probs = model(context_idxs)\n\n        # \ub124\ubc88\uc9f8. \uc190\uc2e4\ud568\uc218\ub97c \uacc4\uc0b0\ud569\ub2c8\ub2e4. (\ud30c\uc774\ud1a0\uce58\uc5d0\uc11c\ub294 \ubaa9\ud45c \ub2e8\uc5b4\ub97c \ud150\uc11c\ub85c \uac10\uc2f8\uc918\uc57c \ud569\ub2c8\ub2e4.)\n        loss = loss_function(log_probs, torch.tensor([word_to_ix[target]], dtype=torch.long))\n\n        # \ub2e4\uc12f\ubc88\uc9f8. \uc5ed\uc804\ud30c\ub97c \ud1b5\ud574 \uae30\uc6b8\uae30\ub97c \uc5c5\ub370\uc774\ud2b8 \ud574\uc90d\ub2c8\ub2e4.\n        loss.backward()\n        optimizer.step()\n\n        # tensor.item()\uc744 \ud638\ucd9c\ud558\uc5ec \ub2e8\uc77c\uc6d0\uc18c \ud150\uc11c\uc5d0\uc11c \uc22b\uc790\ub97c \ubc18\ud658\ubc1b\uc2b5\ub2c8\ub2e4.\n        total_loss += loss.item()\n    losses.append(total_loss)\nprint(losses)  # \ubc18\ubcf5\ud560 \ub54c\ub9c8\ub2e4 \uc190\uc2e4\uc774 \uc904\uc5b4\ub4dc\ub294 \uac83\uc744 \ubd05\uc2dc\ub2e4!\n\n# \"beauty\"\uc640 \uac19\uc774 \ud2b9\uc815 \ub2e8\uc5b4\uc5d0 \ub300\ud55c \uc784\ubca0\ub529\uc744 \ud655\uc778\ud558\ub824\uba74,\nprint(model.embeddings.weight[word_to_ix[\"beauty\"]])"
       ]
     },
     {

docs/_downloads/363afc3b7c522e4e56981679c22f1ad6/word_embeddings_tutorial.ipynb

@@ -345,8 +345,8 @@ def plot_specgram(waveform, sample_rate, title="Spectrogram", xlim=None):
 # 전화 녹음 모의 실험하기
 # ---------------------------
 #
-# 이전 기술들을 혼합하여, 반향있는 방의 사람들이 이야기하는 배경에서 전화 통화하는
-# 것 처럼 들리는 오디오를 모의 실험할 수 있습니다.
+# 이전 기술들을 혼합하여, 반향 있는 방의 사람들이 이야기하는 배경에서 전화 통화하는
+# 것처럼 들리는 오디오를 모의 실험할 수 있습니다.
 #
 
 sample_rate = 16000