{"id":177969,"date":"2026-01-16T14:31:54","date_gmt":"2026-01-16T14:31:54","guid":{"rendered":"https:\/\/ktromedia.com\/?p=177969"},"modified":"2026-01-16T14:31:54","modified_gmt":"2026-01-16T14:31:54","slug":"%e4%bd%bf%e7%94%a8-torch-compile-%e5%92%8c%e6%a2%af%e5%ba%a6%e7%b4%af%e7%a7%af%e6%9b%b4%e5%bf%ab%e5%9c%b0%e8%ae%ad%e7%bb%83%e6%a8%a1%e5%9e%8b","status":"publish","type":"post","link":"http:\/\/ktromedia.com\/?p=177969","title":{"rendered":"\u4f7f\u7528 torch.compile \u548c\u68af\u5ea6\u7d2f\u79ef\u66f4\u5feb\u5730\u8bad\u7ec3\u6a21\u578b"},"content":{"rendered":"<div id=\"\">\n<p>\u4f7f\u7528\u6df1\u5c42\u8f6c\u6362\u5668\u67b6\u6784\u8bad\u7ec3\u8bed\u8a00\u6a21\u578b\u975e\u5e38\u8017\u65f6\u3002\u4f46\u662f\uff0c\u60a8\u53ef\u4ee5\u4f7f\u7528\u4e00\u4e9b\u6280\u672f\u6765\u52a0\u901f\u8bad\u7ec3\u3002\u5728\u672c\u6587\u4e2d\uff0c\u60a8\u5c06\u4e86\u89e3\uff1a<\/p>\n<ul>\n<li>\u4f7f\u7528 <code>torch.compile()<\/code> \u52a0\u901f\u6a21\u578b<\/li>\n<li>\u4f7f\u7528\u68af\u5ea6\u7d2f\u79ef\u8bad\u7ec3\u5177\u6709\u66f4\u5927\u6709\u6548\u6279\u91cf\u5927\u5c0f\u7684\u6a21\u578b<\/li>\n<\/ul>\n<p>\u8ba9\u6211\u4eec\u5f00\u59cb\u5427\uff01<\/p>\n<div id=\"attachment_22899\" style=\"width: 810px\" class=\"wp-caption aligncenter\"><\/p>\n<p id=\"caption-attachment-22899\" class=\"wp-caption-text\">\u4f7f\u7528 torch.compile \u548c\u68af\u5ea6\u7d2f\u79ef\u66f4\u5feb\u5730\u8bad\u7ec3\u6a21\u578b<br \/>\u6444\u5f71\uff1a <a href=\"https:\/\/unsplash.com\/photos\/people-at-the-park-near-gate-during-day-IvlV_Dlt9hg\">\u5f17\u6717\u7d22\u74e6\u00b7\u70ed\u519c<\/a>\u3002\u4fdd\u7559\u4e00\u4e9b\u6743\u5229\u3002<\/p>\n<\/div>\n<h2>\u6982\u8ff0<\/h2>\n<p>\u672c\u6587\u5206\u4e3a\u4e24\u90e8\u5206\uff1b\u4ed6\u4eec\u662f\uff1a<\/p>\n<ul>\n<li>\u4f7f\u7528 <code>torch.compile()<\/code><\/li>\n<li>\u68af\u5ea6\u7d2f\u79ef<\/li>\n<\/ul>\n<h2>\u4f7f\u7528 torch.compile<\/h2>\n<p>\u5f53\u60a8\u7f16\u5199\u6a21\u578b\u4ee3\u7801\u5e76\u5728 PyTorch \u4e2d\u8fd0\u884c\u5b83\u65f6\uff0c\u5b83\u4f1a\u4ee5 eager \u6a21\u5f0f\u6267\u884c\u3002\u8fd9\u610f\u5473\u7740\u4ee3\u7801\u662f\u9010\u884c\u6267\u884c\u7684\uff0c\u7ed3\u679c\u5b58\u50a8\u5728\u5185\u5b58\u4e2d\u3002\u8fd9\u662f Python \u7684\u672c\u673a\u7279\u6027\uff0c\u56e0\u4e3a\u5b83\u662f\u4e00\u79cd\u89e3\u91ca\u6027\u8bed\u8a00\u3002\u60a8\u77e5\u9053\u662f\u8fd9\u79cd\u60c5\u51b5\uff0c\u56e0\u4e3a\u5f53\u60a8\u5728\u4ee3\u7801\u4e2d\u72af\u9519\u65f6\uff0c\u53ea\u6709\u8fd0\u884c\u8be5\u884c\u4ee3\u7801\u540e\u60a8\u624d\u4f1a\u770b\u5230\u8be5\u9519\u8bef\u3002<\/p>\n<p>\u5728 Eager \u6a21\u5f0f\u4e0b\u8fd0\u884c\u6a21\u578b\u901f\u5ea6\u5f88\u6162\u3002\u4ece PyTorch 2.0 \u5f00\u59cb\uff0c\u60a8\u53ef\u4ee5\u4f7f\u7528 <code>torch.compile()<\/code> \u7f16\u8bd1\u6a21\u578b\u4ee5\u63d0\u9ad8\u6027\u80fd\u3002\u8fd9\u4f1a\u751f\u6210\u4e00\u4e2a\u7ecf\u8fc7\u4f18\u5316\u7684\u65b0\u6a21\u578b\u5bf9\u8c61\u3002\u5b83\u4e0e\u60a8\u4f7f\u7528\u521b\u5efa\u7684\u6a21\u578b\u5bf9\u8c61\u4e0d\u540c <code>nn.Module<\/code>\uff0c\u4f46\u5b83\u4e0e\u539f\u59cb\u6a21\u578b\u5171\u4eab\u76f8\u540c\u7684\u5f20\u91cf\u3002\u60a8\u53ef\u4ee5\u50cf\u5f80\u5e38\u4e00\u6837\u4f7f\u7528\u6b64\u7f16\u8bd1\u6a21\u578b\u8fdb\u884c\u524d\u5411\u4f20\u9012\u3001\u540e\u5411\u4f20\u9012\u548c\u4f18\u5316\u5668\u66f4\u65b0\u3002<\/p>\n<p>\u6784\u5efa\u6a21\u578b\u5e76\u5c06\u5176\u7f16\u8bd1\u4e3a\u8ba1\u7b97\u56fe\u5c31\u662f TensorFlow 1.0 \u7684\u5de5\u4f5c\u65b9\u5f0f\u3002\u8fd9\u4f7f\u5f97\u8c03\u8bd5\u53d8\u5f97\u66f4\u52a0\u56f0\u96be\uff0c\u56e0\u4e3a\u60a8\u6267\u884c\u7684\u6a21\u578b\u65e0\u6cd5\u4e0e\u60a8\u7f16\u5199\u7684\u4ee3\u7801\u9010\u884c\u5339\u914d\u3002\u56e0\u6b64\uff0c\u5728\u8fd0\u884c\u8bd5\u9a8c\u5e76\u786e\u8ba4\u6a21\u578b\u6ca1\u6709\u9519\u8bef\u4e4b\u524d\uff0c\u4e0d\u5e94\u7f16\u8bd1\u6a21\u578b\u3002<\/p>\n<p>\u5e76\u975e\u6240\u6709\u6a21\u578b\u90fd\u53ef\u4ee5\u7f16\u8bd1\u3002\u4f46\u662f\uff0c\u5982\u679c\u60a8\u7684\u6a21\u578b\u652f\u6301\u7f16\u8bd1\uff0c\u60a8\u5c06\u7acb\u5373\u53d7\u76ca\u4e8e\u52a0\u901f\u3002\u8981\u7f16\u8bd1\u6a21\u578b\uff0c\u60a8\u9700\u8981\u505a\u7684\u5c31\u662f\u5728\u51c6\u5907\u4f7f\u7528\u6a21\u578b\u5bf9\u8c61\u4e4b\u524d\u66ff\u6362\u5b83\uff1a<\/p>\n<div id=\"urvanov-syntax-highlighter-696489205a082242090026\" class=\"urvanov-syntax-highlighter-syntax crayon-theme-classic urvanov-syntax-highlighter-font-monaco urvanov-syntax-highlighter-os-pc print-yes notranslate\" data-settings=\" minimize scroll-mouseover disable-anim\" style=\" margin-top: 12px; margin-bottom: 12px; font-size: 12px !important; line-height: 15px !important;\">\n<p><textarea wrap=\"soft\" class=\"urvanov-syntax-highlighter-plain print-no\" data-settings=\"dblclick\" readonly=\"readonly\" style=\"-moz-tab-size:4; -o-tab-size:4; -webkit-tab-size:4; tab-size:4; font-size: 12px !important; line-height: 15px !important;\"><\/p>\n<p>&#8230; model = LlamaForPretraining(model_config).to(device) model.load_state_dict(checkpoint) model = torch.compile(model) &#8230;<\/textarea><\/p>\n<div class=\"urvanov-syntax-highlighter-main\" style=\"\">\n<table class=\"crayon-table\">\n<tr class=\"urvanov-syntax-highlighter-row\">\n<td class=\"crayon-nums \" data-settings=\"show\">\n<\/td>\n<td class=\"urvanov-syntax-highlighter-code\">\n<div class=\"crayon-pre\" style=\"font-size: 12px !important; line-height: 15px !important; -moz-tab-size:4; -o-tab-size:4; -webkit-tab-size:4; tab-size:4;\">\n<p><span class=\"crayon-sy\">\u3002<\/span><span class=\"crayon-sy\">\u3002<\/span><span class=\"crayon-sy\">\u3002<\/span><\/p>\n<p><span class=\"crayon-v\">\u6a21\u578b<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-o\">=<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-e\">\u9884\u8bad\u7ec3\u9a86\u9a7c<\/span><span class=\"crayon-sy\">\uff08<\/span><span class=\"crayon-v\">\u6a21\u578b\u914d\u7f6e<\/span><span class=\"crayon-sy\">\uff09<\/span><span class=\"crayon-sy\">\u3002<\/span><span class=\"crayon-st\">\u5230<\/span><span class=\"crayon-sy\">\uff08<\/span><span class=\"crayon-v\">\u8bbe\u5907<\/span><span class=\"crayon-sy\">\uff09<\/span><\/p>\n<p><span class=\"crayon-v\">\u6a21\u578b<\/span><span class=\"crayon-sy\">\u3002<\/span><span class=\"crayon-e\">\u52a0\u8f7d\u72b6\u6001\u5b57\u5178<\/span><span class=\"crayon-sy\">\uff08<\/span><span class=\"crayon-v\">\u68c0\u67e5\u7ad9<\/span><span class=\"crayon-sy\">\uff09<\/span><\/p>\n<p><span class=\"crayon-v\">\u6a21\u578b<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-o\">=<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-v\">\u706b\u70ac<\/span><span class=\"crayon-sy\">\u3002<\/span><span class=\"crayon-e\">\u7f16\u8bd1<\/span><span class=\"crayon-sy\">\uff08<\/span><span class=\"crayon-v\">\u6a21\u578b<\/span><span class=\"crayon-sy\">\uff09<\/span><\/p>\n<p><span class=\"crayon-sy\">\u3002<\/span><span class=\"crayon-sy\">\u3002<\/span><span class=\"crayon-sy\">\u3002<\/span><\/p>\n<\/div>\n<\/td>\n<\/tr>\n<\/table><\/div>\n<\/p><\/div>\n<p>\u7f16\u8bd1\u540e\u4e0d\u8981\u52a0\u8f7d\u6a21\u578b\u6743\u91cd\u3002\u8fd9\u662f\u56e0\u4e3a\u7f16\u8bd1\u540e\u7684\u6a21\u578b\u662f\u4e00\u4e2a\u4e0e\u539f\u59cb\u6a21\u578b\u5171\u4eab\u76f8\u540c\u6743\u91cd\u7684\u5bf9\u8c61\u3002\u5728\u7f16\u8bd1\u8fc7\u7a0b\u4e2d\uff0c\u8ba1\u7b97\u56fe\u662f\u53c2\u8003\u539f\u59cb\u6a21\u578b\u7684\u6743\u91cd\u5f20\u91cf\u6784\u5efa\u7684\u3002\u5982\u679c\u5728\u7f16\u8bd1\u540e\u52a0\u8f7d\u6743\u91cd\uff0c\u6a21\u578b\u53ef\u80fd\u65e0\u6cd5\u6309\u9884\u671f\u5de5\u4f5c\u3002<\/p>\n<p>\u540c\u6837\uff0c\u8981\u4fdd\u5b58\u7f16\u8bd1\u540e\u7684\u6a21\u578b\uff0c\u60a8\u5e94\u8be5\u5f15\u7528\u539f\u59cb\u6a21\u578b\u7684\u72b6\u6001\u5b57\u5178\uff0c\u5982\u4e0b\u6240\u793a\uff1a<\/p>\n<div id=\"urvanov-syntax-highlighter-696489205a08b651958123\" class=\"urvanov-syntax-highlighter-syntax crayon-theme-classic urvanov-syntax-highlighter-font-monaco urvanov-syntax-highlighter-os-pc print-yes notranslate\" data-settings=\" minimize scroll-mouseover disable-anim\" style=\" margin-top: 12px; margin-bottom: 12px; font-size: 12px !important; line-height: 15px !important;\">\n<p><textarea wrap=\"soft\" class=\"urvanov-syntax-highlighter-plain print-no\" data-settings=\"dblclick\" readonly=\"readonly\" style=\"-moz-tab-size:4; -o-tab-size:4; -webkit-tab-size:4; tab-size:4; font-size: 12px !important; line-height: 15px !important;\"><\/p>\n<p>torch.save(getattr(model, &#8220;_orig_mod&#8221;, model).state_dict(), &#8220;model.pth&#8221;)<\/textarea><\/p>\n<div class=\"urvanov-syntax-highlighter-main\" style=\"\">\n<table class=\"crayon-table\">\n<tr class=\"urvanov-syntax-highlighter-row\">\n<td class=\"crayon-nums \" data-settings=\"show\">\n<\/td>\n<td class=\"urvanov-syntax-highlighter-code\">\n<div class=\"crayon-pre\" style=\"font-size: 12px !important; line-height: 15px !important; -moz-tab-size:4; -o-tab-size:4; -webkit-tab-size:4; tab-size:4;\">\n<p><span class=\"crayon-v\">\u706b\u70ac<\/span><span class=\"crayon-sy\">\u3002<\/span><span class=\"crayon-e\">\u8282\u7701<\/span><span class=\"crayon-sy\">\uff08<\/span><span class=\"crayon-e\">\u83b7\u53d6\u5c5e\u6027<\/span><span class=\"crayon-sy\">\uff08<\/span><span class=\"crayon-v\">\u6a21\u578b<\/span><span class=\"crayon-sy\">,<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-s\">\u201c_original_mod\u201d<\/span><span class=\"crayon-sy\">,<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-v\">\u6a21\u578b<\/span><span class=\"crayon-sy\">\uff09<\/span><span class=\"crayon-sy\">\u3002<\/span><span class=\"crayon-e\">\u72b6\u6001\u5b57\u5178<\/span><span class=\"crayon-sy\">\uff08<\/span><span class=\"crayon-sy\">\uff09<\/span><span class=\"crayon-sy\">,<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-s\">\u201c\u578b\u53f7.pth\u201d<\/span><span class=\"crayon-sy\">\uff09<\/span><\/p>\n<\/div>\n<\/td>\n<\/tr>\n<\/table><\/div>\n<\/p><\/div>\n<p>\u53ef\u4ee5\u4f7f\u7528\u7f16\u8bd1\u540e\u7684\u6a21\u578b\u8bbf\u95ee\u539f\u59cb\u6a21\u578b <code>model._orig_mod<\/code>\u3002\u5728\u4e0a\u9762\u7684\u4ee3\u7801\u4e2d\uff0c\u6211\u4eec\u4f7f\u7528 <code>getattr(model, \"_orig_mod\", model)<\/code> \u83b7\u53d6\u539f\u59cb\u6a21\u578b\uff08\u5982\u679c\u5b58\u5728\uff09\uff0c\u6216\u4f7f\u7528 <code>model<\/code> \u672c\u8eab\uff0c\u5982\u679c\u6ca1\u6709\u3002\u8fd9\u884c\u4ee3\u7801\u9002\u7528\u4e8e\u7f16\u8bd1\u6a21\u578b\u548c\u539f\u59cb\u6a21\u578b\u3002<\/p>\n<h2>\u68af\u5ea6\u7d2f\u79ef<\/h2>\n<p>\u5f53\u60a8\u8bad\u7ec3\u6a21\u578b\u65f6\uff0c\u60a8\u5728\u5411\u540e\u4f20\u9012\u4e0a\u82b1\u8d39\u7684\u65f6\u95f4\u53ef\u80fd\u6bd4\u5411\u524d\u4f20\u9012\u591a\u4e24\u5230\u4e09\u500d\u3002\u8fd9\u662f\u56e0\u4e3a\u5411\u540e\u4f20\u9012\u7684\u8ba1\u7b97\u91cf\u66f4\u5927\u5e76\u4e14\u4f7f\u7528\u66f4\u591a\u7684\u5185\u5b58\u3002<\/p>\n<p>\u52a0\u5feb\u8bad\u7ec3\u901f\u5ea6\u7684\u4e00\u4e2a\u7b80\u5355\u6280\u5de7\u662f\u51cf\u5c11\u5411\u540e\u4f20\u9012\u3002\u8fd9\u53ef\u4ee5\u901a\u8fc7\u589e\u52a0\u6279\u5927\u5c0f\u6765\u5b9e\u73b0\uff1a\u5728\u76f8\u540c\u6570\u91cf\u7684\u6570\u636e\u6837\u672c\u4e0b\uff0c\u66f4\u5927\u7684\u6279\u5927\u5c0f\u610f\u5473\u7740\u8981\u5904\u7406\u7684\u6279\u66f4\u5c11\u3002<\/p>\n<p>\u4f46\u662f\uff0c\u8f83\u5927\u7684\u6279\u5904\u7406\u5927\u5c0f\u9700\u8981\u66f4\u591a\u7684\u5185\u5b58\u3002\u5728\u5185\u5b58\u53d7\u9650\u7684\u73af\u5883\u4e2d\uff0c\u60a8\u53ef\u4ee5\u901a\u8fc7\u8fd0\u884c\u591a\u4e2a\u524d\u5411\u4f20\u9012\u5e76\u7d2f\u79ef\u68af\u5ea6\u6765\u6a21\u62df\u66f4\u5927\u7684\u6279\u91cf\u5927\u5c0f\u3002\u8fd9\u5c31\u662f\u6240\u8c13\u7684 <strong>\u68af\u5ea6\u7d2f\u79ef<\/strong>\u3002<\/p>\n<p>\u7528\u4ee3\u7801\u6765\u89e3\u91ca\u8fd9\u4e2a\u60f3\u6cd5\u66f4\u5bb9\u6613\uff1a<\/p>\n<div id=\"urvanov-syntax-highlighter-696489205a08e019405992\" class=\"urvanov-syntax-highlighter-syntax crayon-theme-classic urvanov-syntax-highlighter-font-monaco urvanov-syntax-highlighter-os-pc print-yes notranslate\" data-settings=\" minimize scroll-mouseover disable-anim\" style=\" margin-top: 12px; margin-bottom: 12px; font-size: 12px !important; line-height: 15px !important;\">\n<p><textarea wrap=\"soft\" class=\"urvanov-syntax-highlighter-plain print-no\" data-settings=\"dblclick\" readonly=\"readonly\" style=\"-moz-tab-size:4; -o-tab-size:4; -webkit-tab-size:4; tab-size:4; font-size: 12px !important; line-height: 15px !important;\"><\/p>\n<p>..accumulate_steps = 4 for epoch in range(num_epochs): optimizationr.zero_grad() for i, batch in enumerate(dataloader): # \u83b7\u53d6\u6279\u91cf\u6570\u636e input_ids, target_ids = batch # \u521b\u5efa\u6ce8\u610f\u529b\u63a9\u7801\uff1a\u56e0\u679c\u63a9\u7801 + \u586b\u5145\u63a9\u7801 attn_mask = create_causal_mask(input_ids.shape[1]device) + \\ create_padding_mask(input_ids, PAD_TOKEN_ID, device) # \u4ece\u6a21\u578b\u4e2d\u63d0\u53d6\u8f93\u51fa logits = model(input_ids, attn_mask) # \u8ba1\u7b97\u635f\u5931\uff1alogits \u548c\u76ee\u6807\u4e4b\u95f4\u7684\u4ea4\u53c9\u71b5\uff0c\u5ffd\u7565\u586b\u5145\u6807\u8bb0 loss = loss_fn(logits.view(-1, logits.size(-1)), target_ids.view(-1)) loss = loss \/accumulate_steps # \u8fd0\u884c\u5411\u540e\uff0c\u4f46\u4ec5\u6bcf\u4e2a `accumulate_steps` \u6b65\u9aa4\u66f4\u65b0\u4e00\u6b21 loss.backward() if (i + 1) %accumulate_steps == 0: torch.nn.utils.clip_grad_norm_(model.parameters(), 1.0) Optimizer.step() Optimizer.zero_grad() Scheduler.step()<\/textarea><\/p>\n<div class=\"urvanov-syntax-highlighter-main\" style=\"\">\n<table class=\"crayon-table\">\n<tr class=\"urvanov-syntax-highlighter-row\">\n<td class=\"crayon-nums \" data-settings=\"show\">\n<div class=\"urvanov-syntax-highlighter-nums-content\" style=\"font-size: 12px !important; line-height: 15px !important;\">\n<p>1<\/p>\n<p>2<\/p>\n<p>3<\/p>\n<p>4<\/p>\n<p>5<\/p>\n<p>6<\/p>\n<p>7<\/p>\n<p>8<\/p>\n<p>9<\/p>\n<p>10<\/p>\n<p>11<\/p>\n<p>12<\/p>\n<p>13<\/p>\n<p>14<\/p>\n<p>15<\/p>\n<p>16<\/p>\n<p>17 \u53f7<\/p>\n<p>18<\/p>\n<p>19<\/p>\n<p>20<\/p>\n<p>21<\/p>\n<p>22 \u53f7<\/p>\n<\/div>\n<\/td>\n<td class=\"urvanov-syntax-highlighter-code\">\n<div class=\"crayon-pre\" style=\"font-size: 12px !important; line-height: 15px !important; -moz-tab-size:4; -o-tab-size:4; -webkit-tab-size:4; tab-size:4;\">\n<p><span class=\"crayon-sy\">\u3002<\/span><span class=\"crayon-sy\">\u3002<\/span><\/p>\n<p><span class=\"crayon-v\">\u7d2f\u79ef\u6b65\u6570<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-o\">=<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-cn\">4<\/span><\/p>\n<p><span class=\"crayon-st\">\u4e3a\u4e86<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-e\">\u65f6\u4ee3 <\/span><span class=\"crayon-st\">\u5728<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-e\">\u8303\u56f4<\/span><span class=\"crayon-sy\">\uff08<\/span><span class=\"crayon-v\">\u7eaa\u5143\u6570<\/span><span class=\"crayon-sy\">\uff09<\/span><span class=\"crayon-o\">:<\/span><\/p>\n<p><span class=\"crayon-h\">    <\/span><span class=\"crayon-v\">\u4f18\u5316\u5668<\/span><span class=\"crayon-sy\">\u3002<\/span><span class=\"crayon-e\">\u96f6\u68af\u5ea6<\/span><span class=\"crayon-sy\">\uff08<\/span><span class=\"crayon-sy\">\uff09<\/span><\/p>\n<p><span class=\"crayon-h\">    <\/span><span class=\"crayon-st\">\u4e3a\u4e86<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-v\">\u6211<\/span><span class=\"crayon-sy\">,<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-e\">\u6279 <\/span><span class=\"crayon-st\">\u5728<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-e\">\u679a\u4e3e<\/span><span class=\"crayon-sy\">\uff08<\/span><span class=\"crayon-v\">\u6570\u636e\u52a0\u8f7d\u5668<\/span><span class=\"crayon-sy\">\uff09<\/span><span class=\"crayon-o\">:<\/span><\/p>\n<p><span class=\"crayon-h\">        <\/span><span class=\"crayon-p\"># \u83b7\u53d6\u6279\u91cf\u6570\u636e<\/span><\/p>\n<p><span class=\"crayon-h\">        <\/span><span class=\"crayon-v\">\u8f93\u5165ID<\/span><span class=\"crayon-sy\">,<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-v\">\u76ee\u6807ID<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-o\">=<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-i\">\u6279<\/span><\/p>\n<p><span class=\"crayon-h\">        <\/span><span class=\"crayon-p\"># \u521b\u5efa\u6ce8\u610f\u529b\u63a9\u7801\uff1a\u56e0\u679c\u63a9\u7801+\u586b\u5145\u63a9\u7801<\/span><\/p>\n<p><span class=\"crayon-h\">        <\/span><span class=\"crayon-v\">\u5c5e\u6027\u63a9\u7801<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-o\">=<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-e\">\u521b\u5efa\u56e0\u679c\u63a9\u7801<\/span><span class=\"crayon-sy\">\uff08<\/span><span class=\"crayon-v\">\u8f93\u5165ID<\/span><span class=\"crayon-sy\">\u3002<\/span><span class=\"crayon-v\">\u5f62\u72b6<\/span><span class=\"crayon-sy\">[<\/span><span class=\"crayon-cn\">1<\/span><span class=\"crayon-sy\">]<\/span><span class=\"crayon-sy\">,<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-v\">\u8bbe\u5907<\/span><span class=\"crayon-sy\">\uff09<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-o\">+<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-sy\">\\<\/span><\/p>\n<p><span class=\"crayon-h\">                    <\/span><span class=\"crayon-e\">\u521b\u5efa\u586b\u5145\u63a9\u7801<\/span><span class=\"crayon-sy\">\uff08<\/span><span class=\"crayon-v\">\u8f93\u5165ID<\/span><span class=\"crayon-sy\">,<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-v\">PAD_TOKEN_ID<\/span><span class=\"crayon-sy\">,<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-v\">\u8bbe\u5907<\/span><span class=\"crayon-sy\">\uff09<\/span><\/p>\n<p><span class=\"crayon-h\">        <\/span><span class=\"crayon-p\"># \u4ece\u6a21\u578b\u4e2d\u63d0\u53d6\u8f93\u51fa<\/span><\/p>\n<p><span class=\"crayon-h\">        <\/span><span class=\"crayon-v\">\u903b\u8f91\u6570<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-o\">=<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-e\">\u6a21\u578b<\/span><span class=\"crayon-sy\">\uff08<\/span><span class=\"crayon-v\">\u8f93\u5165ID<\/span><span class=\"crayon-sy\">,<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-v\">\u5c5e\u6027\u63a9\u7801<\/span><span class=\"crayon-sy\">\uff09<\/span><\/p>\n<p><span class=\"crayon-h\">        <\/span><span class=\"crayon-p\"># \u8ba1\u7b97\u635f\u5931\uff1alogits \u548c\u76ee\u6807\u4e4b\u95f4\u7684\u4ea4\u53c9\u71b5\uff0c\u5ffd\u7565\u586b\u5145\u6807\u8bb0<\/span><\/p>\n<p><span class=\"crayon-h\">        <\/span><span class=\"crayon-v\">\u635f\u5931<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-o\">=<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-e\">\u635f\u5931\u51fd\u6570<\/span><span class=\"crayon-sy\">\uff08<\/span><span class=\"crayon-v\">\u903b\u8f91\u6570<\/span><span class=\"crayon-sy\">\u3002<\/span><span class=\"crayon-e\">\u770b\u6cd5<\/span><span class=\"crayon-sy\">\uff08<\/span><span class=\"crayon-o\">&#8211;<\/span><span class=\"crayon-cn\">1<\/span><span class=\"crayon-sy\">,<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-v\">\u903b\u8f91\u6570<\/span><span class=\"crayon-sy\">\u3002<\/span><span class=\"crayon-e\">\u5c3a\u5bf8<\/span><span class=\"crayon-sy\">\uff08<\/span><span class=\"crayon-o\">&#8211;<\/span><span class=\"crayon-cn\">1<\/span><span class=\"crayon-sy\">\uff09<\/span><span class=\"crayon-sy\">\uff09<\/span><span class=\"crayon-sy\">,<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-v\">\u76ee\u6807ID<\/span><span class=\"crayon-sy\">\u3002<\/span><span class=\"crayon-e\">\u770b\u6cd5<\/span><span class=\"crayon-sy\">\uff08<\/span><span class=\"crayon-o\">&#8211;<\/span><span class=\"crayon-cn\">1<\/span><span class=\"crayon-sy\">\uff09<\/span><span class=\"crayon-sy\">\uff09<\/span><\/p>\n<p><span class=\"crayon-h\">        <\/span><span class=\"crayon-v\">\u635f\u5931<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-o\">=<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-v\">\u635f\u5931<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-o\">\/<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-v\">\u79ef\u7d2f<\/span><span class=\"crayon-sy\">_<\/span>\u6b65\u9aa4<\/p>\n<p><span class=\"crayon-h\">        <\/span><span class=\"crayon-p\"># \u5411\u540e\u8fd0\u884c\uff0c\u4f46\u6bcf\u4e2a `accumulate_steps` \u6b65\u9aa4\u53ea\u66f4\u65b0\u4e00\u6b21<\/span><\/p>\n<p><span class=\"crayon-h\">        <\/span><span class=\"crayon-v\">\u635f\u5931<\/span><span class=\"crayon-sy\">\u3002<\/span><span class=\"crayon-e\">\u843d\u540e<\/span><span class=\"crayon-sy\">\uff08<\/span><span class=\"crayon-sy\">\uff09<\/span><\/p>\n<p><span class=\"crayon-h\">        <\/span><span class=\"crayon-st\">\u5982\u679c<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-sy\">\uff08<\/span><span class=\"crayon-v\">\u6211<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-o\">+<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-cn\">1<\/span><span class=\"crayon-sy\">\uff09<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-o\">%<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-v\">\u7d2f\u79ef\u6b65\u6570<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-o\">==<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-cn\">0<\/span><span class=\"crayon-o\">:<\/span><\/p>\n<p><span class=\"crayon-h\">            <\/span><span class=\"crayon-v\">\u706b\u70ac<\/span><span class=\"crayon-sy\">\u3002<\/span><span class=\"crayon-v\">\u6069<\/span><span class=\"crayon-sy\">\u3002<\/span><span class=\"crayon-v\">\u5b9e\u7528\u7a0b\u5e8f<\/span><span class=\"crayon-sy\">\u3002<\/span><span class=\"crayon-e\">\u526a\u8f91_\u68af\u5ea6_\u8303\u6570_<\/span><span class=\"crayon-sy\">\uff08<\/span><span class=\"crayon-v\">\u6a21\u578b<\/span><span class=\"crayon-sy\">\u3002<\/span><span class=\"crayon-e\">\u53c2\u6570<\/span><span class=\"crayon-sy\">\uff08<\/span><span class=\"crayon-sy\">\uff09<\/span><span class=\"crayon-sy\">,<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-cn\">1.0<\/span><span class=\"crayon-sy\">\uff09<\/span><\/p>\n<p><span class=\"crayon-h\">            <\/span><span class=\"crayon-v\">\u4f18\u5316\u5668<\/span><span class=\"crayon-sy\">\u3002<\/span><span class=\"crayon-e\">\u6b65<\/span><span class=\"crayon-sy\">\uff08<\/span><span class=\"crayon-sy\">\uff09<\/span><\/p>\n<p><span class=\"crayon-h\">            <\/span><span class=\"crayon-v\">\u4f18\u5316\u5668<\/span><span class=\"crayon-sy\">\u3002<\/span><span class=\"crayon-e\">\u96f6\u68af\u5ea6<\/span><span class=\"crayon-sy\">\uff08<\/span><span class=\"crayon-sy\">\uff09<\/span><\/p>\n<p><span class=\"crayon-h\">            <\/span><span class=\"crayon-v\">\u8c03\u5ea6\u7a0b\u5e8f<\/span><span class=\"crayon-sy\">\u3002<\/span><span class=\"crayon-e\">\u6b65<\/span><span class=\"crayon-sy\">\uff08<\/span><span class=\"crayon-sy\">\uff09<\/span><\/p>\n<\/div>\n<\/td>\n<\/tr>\n<\/table><\/div>\n<\/p><\/div>\n<p>\u4e0a\u9762\u7684\u8bad\u7ec3\u5faa\u73af\u6458\u5f55\u81ea <a href=\"https:\/\/machinelearningmastery.com\/pretraining-a-llama-model-on-your-local-gpu\/\">\u4e0a\u4e00\u7bc7\u6587\u7ae0<\/a> \u7528\u4e8e\u5728\u672c\u5730 GPU \u4e0a\u8bad\u7ec3 Llama \u6a21\u578b\u3002<\/p>\n<p>\u901a\u5e38\uff0c\u5f53\u60a8\u8fdb\u884c\u524d\u5411\u4f20\u7403\u65f6\uff0c\u60a8\u4f1a\u8ba1\u7b97\u635f\u5931\u3002\u7136\u540e\u4f60\u6253\u7535\u8bdd <code>loss.backward()<\/code> \u901a\u8fc7\u6a21\u578b\u53c2\u6570\u53cd\u5411\u4f20\u64ad\u635f\u5931\u68af\u5ea6\u3002\u5728 PyTorch \u4e2d\uff0c <code>backward()<\/code> \u65b9\u6cd5\u662f\u7d2f\u79ef\u7684\uff0c\u610f\u5473\u7740\u68af\u5ea6\u76f8\u52a0\u3002\u56e0\u6b64\uff0c\u60a8\u9700\u8981\u8c03\u7528 <code>optimizer.zero_grad()<\/code> \u5728\u8fd0\u884c\u5411\u540e\u4f20\u9012\u4e4b\u524d\u663e\u5f0f\u6e05\u9664\u68af\u5ea6\u3002<\/p>\n<p>\u5728\u4e0a\u9762\u7684\u4ee3\u7801\u4e2d\uff0c\u4f60\u6545\u610f\u4e0d\u8c03\u7528 <code>optimizer.zero_grad()<\/code> \u5728\u6bcf\u6b21\u8fed\u4ee3\u4e2d\u3002\u76f8\u53cd\uff0c\u60a8\u5bf9\u635f\u5931\u8fdb\u884c\u53cd\u5411\u4f20\u64ad\u9664\u4ee5 <code>accumulate_steps<\/code>\u3002\u8fd9\u6837\uff0c\u68af\u5ea6\u4f1a\u6309\u6bd4\u4f8b\u7f29\u5c0f\uff0c\u4f46\u4f1a\u7d2f\u79ef <code>accumulate_steps<\/code> \u8fed\u4ee3\u3002\u6bcf\u4e00\u6b21 <code>accumulate_steps<\/code> \u8fed\u4ee3\u65f6\uff0c\u60a8\u8fd0\u884c\u4f18\u5316\u5668\u6765\u8c03\u6574\u6a21\u578b\u53c2\u6570\u3002<\/p>\n<p>\u8fd9\u79cd\u65b9\u6cd5\u4ea7\u751f\u7684\u7ed3\u679c\u4e0e\u4f7f\u7528\u8f83\u5927\u6279\u91cf\u5927\u5c0f\u83b7\u5f97\u7684\u7ed3\u679c\u76f8\u5f53\u3002\u4f46\u662f\uff0c\u7531\u4e8e\u60a8\u8fd0\u884c\u7684\u4f18\u5316\u5668\u66f4\u65b0\u8f83\u5c11\uff0c\u56e0\u6b64\u5e94\u76f8\u5e94\u8c03\u6574\u5b66\u4e60\u7387\u8ba1\u5212\u3002\u8fd9\u610f\u5473\u7740\u60a8\u9700\u8981\u4f7f\u7528\u4e0d\u540c\u6570\u91cf\u7684\u6b65\u9aa4\u6765\u521d\u59cb\u5316\u8c03\u5ea6\u7a0b\u5e8f\uff1a<\/p>\n<div id=\"urvanov-syntax-highlighter-696489205a091070349743\" class=\"urvanov-syntax-highlighter-syntax crayon-theme-classic urvanov-syntax-highlighter-font-monaco urvanov-syntax-highlighter-os-pc print-yes notranslate\" data-settings=\" minimize scroll-mouseover disable-anim\" style=\" margin-top: 12px; margin-bottom: 12px; font-size: 12px !important; line-height: 15px !important;\">\n<p><textarea wrap=\"soft\" class=\"urvanov-syntax-highlighter-plain print-no\" data-settings=\"dblclick\" readonly=\"readonly\" style=\"-moz-tab-size:4; -o-tab-size:4; -webkit-tab-size:4; tab-size:4; font-size: 12px !important; line-height: 15px !important;\"><\/p>\n<p>&#8230; num_training_steps = (len(dataloader) \/\/ \u7d2f\u79ef\u6b65\u6570) * num_epochs cosine_scheduler = lr_scheduler.CosineAnnealingLR( \u4f18\u5316\u5668, T_max=num_training_steps &#8211; num_warmup_steps, eta_min=0 )<\/textarea><\/p>\n<div class=\"urvanov-syntax-highlighter-main\" style=\"\">\n<table class=\"crayon-table\">\n<tr class=\"urvanov-syntax-highlighter-row\">\n<td class=\"crayon-nums \" data-settings=\"show\">\n<\/td>\n<td class=\"urvanov-syntax-highlighter-code\">\n<div class=\"crayon-pre\" style=\"font-size: 12px !important; line-height: 15px !important; -moz-tab-size:4; -o-tab-size:4; -webkit-tab-size:4; tab-size:4;\">\n<p><span class=\"crayon-sy\">\u3002<\/span><span class=\"crayon-sy\">\u3002<\/span><span class=\"crayon-sy\">\u3002<\/span><\/p>\n<p><span class=\"crayon-v\">\u8bad\u7ec3\u6b65\u6570<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-o\">=<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-sy\">\uff08<\/span><span class=\"crayon-e\">\u4f26<\/span><span class=\"crayon-sy\">\uff08<\/span><span class=\"crayon-v\">\u6570\u636e\u52a0\u8f7d\u5668<\/span><span class=\"crayon-sy\">\uff09<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-c\">\/\/ \u7d2f\u79ef\u6b65\u6570) * num_epochs<\/span><\/p>\n<p><span class=\"crayon-v\">\u4f59\u5f26\u8c03\u5ea6\u5668<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-o\">=<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-v\">lr_\u8c03\u5ea6\u7a0b\u5e8f<\/span><span class=\"crayon-sy\">\u3002<\/span><span class=\"crayon-e\">\u4f59\u5f26\u9000\u706bLR<\/span><span class=\"crayon-sy\">\uff08<\/span><\/p>\n<p><span class=\"crayon-h\">    <\/span><span class=\"crayon-v\">\u4f18\u5316\u5668<\/span><span class=\"crayon-sy\">,<\/span><\/p>\n<p><span class=\"crayon-h\">    <\/span><span class=\"crayon-v\">\u6700\u5927\u6e29\u5ea6<\/span><span class=\"crayon-o\">=<\/span><span class=\"crayon-v\">\u8bad\u7ec3\u6b65\u6570<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-o\">&#8211;<\/span><span class=\"crayon-h\"> <\/span><span class=\"crayon-v\">\u9884\u70ed\u6b65\u9aa4\u6570<\/span><span class=\"crayon-sy\">,<\/span><\/p>\n<p><span class=\"crayon-h\">    <\/span><span class=\"crayon-v\">\u548c\u6700\u5c0f<\/span><span class=\"crayon-o\">=<\/span><span class=\"crayon-cn\">0<\/span><\/p>\n<p><span class=\"crayon-sy\">\uff09<\/span><\/p>\n<\/div>\n<\/td>\n<\/tr>\n<\/table><\/div>\n<\/p><\/div>\n<h2>\u8fdb\u4e00\u6b65\u9605\u8bfb<\/h2>\n<p>\u4ee5\u4e0b\u662f\u4e00\u4e9b\u60a8\u53ef\u80fd\u611f\u5174\u8da3\u7684\u6750\u6599\uff1a<\/p>\n<h2>\u6982\u62ec<\/h2>\n<p>\u5728\u672c\u6587\u4e2d\uff0c\u60a8\u4e86\u89e3\u5230\u4f7f\u7528 <code>torch.compile()<\/code> \u53ef\u4ee5\u901a\u8fc7\u7f16\u8bd1\u8ba1\u7b97\u56fe\u6765\u5e2e\u52a9\u60a8\u52a0\u901f\u6a21\u578b\u3002\u60a8\u8fd8\u4e86\u89e3\u5230\uff0c\u68af\u5ea6\u7d2f\u79ef\u662f\u4e00\u79cd\u901a\u8fc7\u7d2f\u79ef\u591a\u4e2a\u5c0f\u6279\u91cf\u7684\u68af\u5ea6\u6765\u8fdb\u884c\u66f4\u5927\u6709\u6548\u6279\u91cf\u5927\u5c0f\u8bad\u7ec3\u7684\u6280\u672f\u3002\u7531\u4e8e\u901a\u8fc7\u8fd9\u79cd\u65b9\u5f0f\u8fd0\u884c\u7684\u4f18\u5316\u5668\u66f4\u65b0\u8f83\u5c11\uff0c\u56e0\u6b64\u53ef\u4ee5\u8282\u7701\u5411\u540e\u4f20\u9012\u548c\u53c2\u6570\u66f4\u65b0\u7684\u65f6\u95f4\u3002<\/p>\n<\/p><\/div>\n","protected":false},"excerpt":{"rendered":"<p>\u4f7f\u7528\u6df1\u5c42\u8f6c\u6362\u5668\u67b6\u6784\u8bad\u7ec3\u8bed\u8a00\u6a21\u578b\u975e\u5e38\u8017\u65f6\u3002\u4f46\u662f\uff0c\u60a8\u53ef\u4ee5\u4f7f\u7528\u4e00\u4e9b\u6280\u672f\u6765\u52a0\u901f\u8bad\u7ec3\u3002\u5728\u672c\u6587\u4e2d\uff0c\u60a8\u5c06\u4e86\u89e3\uff1a \u4f7f\u7528 torch.compile() \u52a0\u901f\u6a21\u578b \u4f7f\u7528\u68af\u5ea6\u7d2f\u79ef\u8bad\u7ec3\u5177\u6709\u66f4\u5927\u6709\u6548\u6279\u91cf\u5927\u5c0f\u7684\u6a21\u578b \u8ba9\u6211\u4eec\u5f00\u59cb\u5427\uff01 \u4f7f\u7528 torch.compile \u548c\u68af\u5ea6\u7d2f\u79ef\u66f4\u5feb\u5730\u8bad\u7ec3\u6a21\u578b\u6444\u5f71\uff1a \u5f17\u6717\u7d22\u74e6\u00b7\u70ed\u519c\u3002\u4fdd\u7559\u4e00\u4e9b\u6743\u5229\u3002 \u6982\u8ff0 \u672c\u6587\u5206\u4e3a\u4e24\u90e8\u5206\uff1b\u4ed6\u4eec\u662f\uff1a \u4f7f\u7528 torch.compile() \u68af\u5ea6\u7d2f\u79ef \u4f7f\u7528 torch.compile \u5f53\u60a8\u7f16\u5199\u6a21\u578b\u4ee3\u7801\u5e76\u5728 PyTorch \u4e2d\u8fd0\u884c\u5b83\u65f6\uff0c\u5b83\u4f1a\u4ee5 eager \u6a21\u5f0f\u6267\u884c\u3002\u8fd9\u610f\u5473\u7740\u4ee3\u7801\u662f\u9010\u884c\u6267\u884c\u7684\uff0c\u7ed3\u679c\u5b58\u50a8\u5728\u5185\u5b58\u4e2d\u3002\u8fd9\u662f Python \u7684\u672c\u673a\u7279\u6027\uff0c\u56e0\u4e3a\u5b83\u662f\u4e00\u79cd\u89e3\u91ca\u6027\u8bed\u8a00\u3002\u60a8\u77e5\u9053\u662f\u8fd9\u79cd\u60c5\u51b5\uff0c\u56e0\u4e3a\u5f53\u60a8\u5728\u4ee3\u7801\u4e2d\u72af\u9519\u65f6\uff0c\u53ea\u6709\u8fd0\u884c\u8be5\u884c\u4ee3\u7801\u540e\u60a8\u624d\u4f1a\u770b\u5230\u8be5\u9519\u8bef\u3002 \u5728 Eager \u6a21\u5f0f\u4e0b\u8fd0\u884c\u6a21\u578b\u901f\u5ea6\u5f88\u6162\u3002\u4ece PyTorch 2.0 \u5f00\u59cb\uff0c\u60a8\u53ef\u4ee5\u4f7f\u7528 torch.compile() \u7f16\u8bd1\u6a21\u578b\u4ee5\u63d0\u9ad8\u6027\u80fd\u3002\u8fd9\u4f1a\u751f\u6210\u4e00\u4e2a\u7ecf\u8fc7\u4f18\u5316\u7684\u65b0\u6a21\u578b\u5bf9\u8c61\u3002\u5b83\u4e0e\u60a8\u4f7f\u7528\u521b\u5efa\u7684\u6a21\u578b\u5bf9\u8c61\u4e0d\u540c nn.Module\uff0c\u4f46\u5b83\u4e0e\u539f\u59cb\u6a21\u578b\u5171\u4eab\u76f8\u540c\u7684\u5f20\u91cf\u3002\u60a8\u53ef\u4ee5\u50cf\u5f80\u5e38\u4e00\u6837\u4f7f\u7528\u6b64\u7f16\u8bd1\u6a21\u578b\u8fdb\u884c\u524d\u5411\u4f20\u9012\u3001\u540e\u5411\u4f20\u9012\u548c\u4f18\u5316\u5668\u66f4\u65b0\u3002 \u6784\u5efa\u6a21\u578b\u5e76\u5c06\u5176\u7f16\u8bd1\u4e3a\u8ba1\u7b97\u56fe\u5c31\u662f TensorFlow 1.0 \u7684\u5de5\u4f5c\u65b9\u5f0f\u3002\u8fd9\u4f7f\u5f97\u8c03\u8bd5\u53d8\u5f97\u66f4\u52a0\u56f0\u96be\uff0c\u56e0\u4e3a\u60a8\u6267\u884c\u7684\u6a21\u578b\u65e0\u6cd5\u4e0e\u60a8\u7f16\u5199\u7684\u4ee3\u7801\u9010\u884c\u5339\u914d\u3002\u56e0\u6b64\uff0c\u5728\u8fd0\u884c\u8bd5\u9a8c\u5e76\u786e\u8ba4\u6a21\u578b\u6ca1\u6709\u9519\u8bef\u4e4b\u524d\uff0c\u4e0d\u5e94\u7f16\u8bd1\u6a21\u578b\u3002 \u5e76\u975e\u6240\u6709\u6a21\u578b\u90fd\u53ef\u4ee5\u7f16\u8bd1\u3002\u4f46\u662f\uff0c\u5982\u679c\u60a8\u7684\u6a21\u578b\u652f\u6301\u7f16\u8bd1\uff0c\u60a8\u5c06\u7acb\u5373\u53d7\u76ca\u4e8e\u52a0\u901f\u3002\u8981\u7f16\u8bd1\u6a21\u578b\uff0c\u60a8\u9700\u8981\u505a\u7684\u5c31\u662f\u5728\u51c6\u5907\u4f7f\u7528\u6a21\u578b\u5bf9\u8c61\u4e4b\u524d\u66ff\u6362\u5b83\uff1a &#8230; model = LlamaForPretraining(model_config).to(device) model.load_state_dict(checkpoint) model = torch.compile(model) &#8230; \u3002\u3002\u3002 \u6a21\u578b = \u9884\u8bad\u7ec3\u9a86\u9a7c\uff08\u6a21\u578b\u914d\u7f6e\uff09\u3002\u5230\uff08\u8bbe\u5907\uff09 \u6a21\u578b\u3002\u52a0\u8f7d\u72b6\u6001\u5b57\u5178\uff08\u68c0\u67e5\u7ad9\uff09 \u6a21\u578b = \u706b\u70ac\u3002\u7f16\u8bd1\uff08\u6a21\u578b\uff09<\/p>\n","protected":false},"author":1,"featured_media":177970,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[42],"tags":[],"class_list":{"0":"post-177969","1":"post","2":"type-post","3":"status-publish","4":"format-standard","5":"has-post-thumbnail","7":"category-ai"},"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v26.4 - https:\/\/yoast.com\/wordpress\/plugins\/seo\/ -->\n<title>\u4f7f\u7528 torch.compile \u548c\u68af\u5ea6\u7d2f\u79ef\u66f4\u5feb\u5730\u8bad\u7ec3\u6a21\u578b - Ktromedia<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"http:\/\/ktromedia.com\/?p=177969\" \/>\n<meta property=\"og:locale\" content=\"en_US\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"\u4f7f\u7528 torch.compile \u548c\u68af\u5ea6\u7d2f\u79ef\u66f4\u5feb\u5730\u8bad\u7ec3\u6a21\u578b - Ktromedia\" \/>\n<meta property=\"og:description\" content=\"\u4f7f\u7528\u6df1\u5c42\u8f6c\u6362\u5668\u67b6\u6784\u8bad\u7ec3\u8bed\u8a00\u6a21\u578b\u975e\u5e38\u8017\u65f6\u3002\u4f46\u662f\uff0c\u60a8\u53ef\u4ee5\u4f7f\u7528\u4e00\u4e9b\u6280\u672f\u6765\u52a0\u901f\u8bad\u7ec3\u3002\u5728\u672c\u6587\u4e2d\uff0c\u60a8\u5c06\u4e86\u89e3\uff1a \u4f7f\u7528 torch.compile() \u52a0\u901f\u6a21\u578b \u4f7f\u7528\u68af\u5ea6\u7d2f\u79ef\u8bad\u7ec3\u5177\u6709\u66f4\u5927\u6709\u6548\u6279\u91cf\u5927\u5c0f\u7684\u6a21\u578b \u8ba9\u6211\u4eec\u5f00\u59cb\u5427\uff01 \u4f7f\u7528 torch.compile \u548c\u68af\u5ea6\u7d2f\u79ef\u66f4\u5feb\u5730\u8bad\u7ec3\u6a21\u578b\u6444\u5f71\uff1a \u5f17\u6717\u7d22\u74e6\u00b7\u70ed\u519c\u3002\u4fdd\u7559\u4e00\u4e9b\u6743\u5229\u3002 \u6982\u8ff0 \u672c\u6587\u5206\u4e3a\u4e24\u90e8\u5206\uff1b\u4ed6\u4eec\u662f\uff1a \u4f7f\u7528 torch.compile() \u68af\u5ea6\u7d2f\u79ef \u4f7f\u7528 torch.compile \u5f53\u60a8\u7f16\u5199\u6a21\u578b\u4ee3\u7801\u5e76\u5728 PyTorch \u4e2d\u8fd0\u884c\u5b83\u65f6\uff0c\u5b83\u4f1a\u4ee5 eager \u6a21\u5f0f\u6267\u884c\u3002\u8fd9\u610f\u5473\u7740\u4ee3\u7801\u662f\u9010\u884c\u6267\u884c\u7684\uff0c\u7ed3\u679c\u5b58\u50a8\u5728\u5185\u5b58\u4e2d\u3002\u8fd9\u662f Python \u7684\u672c\u673a\u7279\u6027\uff0c\u56e0\u4e3a\u5b83\u662f\u4e00\u79cd\u89e3\u91ca\u6027\u8bed\u8a00\u3002\u60a8\u77e5\u9053\u662f\u8fd9\u79cd\u60c5\u51b5\uff0c\u56e0\u4e3a\u5f53\u60a8\u5728\u4ee3\u7801\u4e2d\u72af\u9519\u65f6\uff0c\u53ea\u6709\u8fd0\u884c\u8be5\u884c\u4ee3\u7801\u540e\u60a8\u624d\u4f1a\u770b\u5230\u8be5\u9519\u8bef\u3002 \u5728 Eager \u6a21\u5f0f\u4e0b\u8fd0\u884c\u6a21\u578b\u901f\u5ea6\u5f88\u6162\u3002\u4ece PyTorch 2.0 \u5f00\u59cb\uff0c\u60a8\u53ef\u4ee5\u4f7f\u7528 torch.compile() \u7f16\u8bd1\u6a21\u578b\u4ee5\u63d0\u9ad8\u6027\u80fd\u3002\u8fd9\u4f1a\u751f\u6210\u4e00\u4e2a\u7ecf\u8fc7\u4f18\u5316\u7684\u65b0\u6a21\u578b\u5bf9\u8c61\u3002\u5b83\u4e0e\u60a8\u4f7f\u7528\u521b\u5efa\u7684\u6a21\u578b\u5bf9\u8c61\u4e0d\u540c nn.Module\uff0c\u4f46\u5b83\u4e0e\u539f\u59cb\u6a21\u578b\u5171\u4eab\u76f8\u540c\u7684\u5f20\u91cf\u3002\u60a8\u53ef\u4ee5\u50cf\u5f80\u5e38\u4e00\u6837\u4f7f\u7528\u6b64\u7f16\u8bd1\u6a21\u578b\u8fdb\u884c\u524d\u5411\u4f20\u9012\u3001\u540e\u5411\u4f20\u9012\u548c\u4f18\u5316\u5668\u66f4\u65b0\u3002 \u6784\u5efa\u6a21\u578b\u5e76\u5c06\u5176\u7f16\u8bd1\u4e3a\u8ba1\u7b97\u56fe\u5c31\u662f TensorFlow 1.0 \u7684\u5de5\u4f5c\u65b9\u5f0f\u3002\u8fd9\u4f7f\u5f97\u8c03\u8bd5\u53d8\u5f97\u66f4\u52a0\u56f0\u96be\uff0c\u56e0\u4e3a\u60a8\u6267\u884c\u7684\u6a21\u578b\u65e0\u6cd5\u4e0e\u60a8\u7f16\u5199\u7684\u4ee3\u7801\u9010\u884c\u5339\u914d\u3002\u56e0\u6b64\uff0c\u5728\u8fd0\u884c\u8bd5\u9a8c\u5e76\u786e\u8ba4\u6a21\u578b\u6ca1\u6709\u9519\u8bef\u4e4b\u524d\uff0c\u4e0d\u5e94\u7f16\u8bd1\u6a21\u578b\u3002 \u5e76\u975e\u6240\u6709\u6a21\u578b\u90fd\u53ef\u4ee5\u7f16\u8bd1\u3002\u4f46\u662f\uff0c\u5982\u679c\u60a8\u7684\u6a21\u578b\u652f\u6301\u7f16\u8bd1\uff0c\u60a8\u5c06\u7acb\u5373\u53d7\u76ca\u4e8e\u52a0\u901f\u3002\u8981\u7f16\u8bd1\u6a21\u578b\uff0c\u60a8\u9700\u8981\u505a\u7684\u5c31\u662f\u5728\u51c6\u5907\u4f7f\u7528\u6a21\u578b\u5bf9\u8c61\u4e4b\u524d\u66ff\u6362\u5b83\uff1a &#8230; model = LlamaForPretraining(model_config).to(device) model.load_state_dict(checkpoint) model = torch.compile(model) &#8230; \u3002\u3002\u3002 \u6a21\u578b = \u9884\u8bad\u7ec3\u9a86\u9a7c\uff08\u6a21\u578b\u914d\u7f6e\uff09\u3002\u5230\uff08\u8bbe\u5907\uff09 \u6a21\u578b\u3002\u52a0\u8f7d\u72b6\u6001\u5b57\u5178\uff08\u68c0\u67e5\u7ad9\uff09 \u6a21\u578b = \u706b\u70ac\u3002\u7f16\u8bd1\uff08\u6a21\u578b\uff09\" \/>\n<meta property=\"og:url\" content=\"http:\/\/ktromedia.com\/?p=177969\" \/>\n<meta property=\"og:site_name\" content=\"Ktromedia\" \/>\n<meta property=\"article:publisher\" content=\"https:\/\/www.facebook.com\/KTROMedia\/\" \/>\n<meta property=\"article:published_time\" content=\"2026-01-16T14:31:54+00:00\" \/>\n<meta property=\"og:image\" content=\"http:\/\/ktromedia.com\/wp-content\/uploads\/2026\/01\/\u4f7f\u7528-torchcompile-\u548c\u68af\u5ea6\u7d2f\u79ef\u66f4\u5feb\u5730\u8bad\u7ec3\u6a21\u578b.jpg\" \/>\n\t<meta property=\"og:image:width\" content=\"2560\" \/>\n\t<meta property=\"og:image:height\" content=\"1707\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/jpeg\" \/>\n<meta name=\"author\" content=\"KTRO TEAM\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"Written by\" \/>\n\t<meta name=\"twitter:data1\" content=\"KTRO TEAM\" \/>\n\t<meta name=\"twitter:label2\" content=\"Est. reading time\" \/>\n\t<meta name=\"twitter:data2\" content=\"1 minute\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"http:\/\/ktromedia.com\/?p=177969#article\",\"isPartOf\":{\"@id\":\"http:\/\/ktromedia.com\/?p=177969\"},\"author\":{\"name\":\"KTRO TEAM\",\"@id\":\"https:\/\/ktromedia.com\/#\/schema\/person\/612bf2fbac107722ea365932cdd35f5b\"},\"headline\":\"\u4f7f\u7528 torch.compile \u548c\u68af\u5ea6\u7d2f\u79ef\u66f4\u5feb\u5730\u8bad\u7ec3\u6a21\u578b\",\"datePublished\":\"2026-01-16T14:31:54+00:00\",\"mainEntityOfPage\":{\"@id\":\"http:\/\/ktromedia.com\/?p=177969\"},\"wordCount\":174,\"commentCount\":0,\"publisher\":{\"@id\":\"https:\/\/ktromedia.com\/#organization\"},\"image\":{\"@id\":\"http:\/\/ktromedia.com\/?p=177969#primaryimage\"},\"thumbnailUrl\":\"http:\/\/ktromedia.com\/wp-content\/uploads\/2026\/01\/\u4f7f\u7528-torchcompile-\u548c\u68af\u5ea6\u7d2f\u79ef\u66f4\u5feb\u5730\u8bad\u7ec3\u6a21\u578b.jpg\",\"articleSection\":[\"\u4eba\u5de5\u667a\u80fd\"],\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"CommentAction\",\"name\":\"Comment\",\"target\":[\"http:\/\/ktromedia.com\/?p=177969#respond\"]}]},{\"@type\":\"WebPage\",\"@id\":\"http:\/\/ktromedia.com\/?p=177969\",\"url\":\"http:\/\/ktromedia.com\/?p=177969\",\"name\":\"\u4f7f\u7528 torch.compile \u548c\u68af\u5ea6\u7d2f\u79ef\u66f4\u5feb\u5730\u8bad\u7ec3\u6a21\u578b - Ktromedia\",\"isPartOf\":{\"@id\":\"https:\/\/ktromedia.com\/#website\"},\"primaryImageOfPage\":{\"@id\":\"http:\/\/ktromedia.com\/?p=177969#primaryimage\"},\"image\":{\"@id\":\"http:\/\/ktromedia.com\/?p=177969#primaryimage\"},\"thumbnailUrl\":\"http:\/\/ktromedia.com\/wp-content\/uploads\/2026\/01\/\u4f7f\u7528-torchcompile-\u548c\u68af\u5ea6\u7d2f\u79ef\u66f4\u5feb\u5730\u8bad\u7ec3\u6a21\u578b.jpg\",\"datePublished\":\"2026-01-16T14:31:54+00:00\",\"breadcrumb\":{\"@id\":\"http:\/\/ktromedia.com\/?p=177969#breadcrumb\"},\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"http:\/\/ktromedia.com\/?p=177969\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"http:\/\/ktromedia.com\/?p=177969#primaryimage\",\"url\":\"http:\/\/ktromedia.com\/wp-content\/uploads\/2026\/01\/\u4f7f\u7528-torchcompile-\u548c\u68af\u5ea6\u7d2f\u79ef\u66f4\u5feb\u5730\u8bad\u7ec3\u6a21\u578b.jpg\",\"contentUrl\":\"http:\/\/ktromedia.com\/wp-content\/uploads\/2026\/01\/\u4f7f\u7528-torchcompile-\u548c\u68af\u5ea6\u7d2f\u79ef\u66f4\u5feb\u5730\u8bad\u7ec3\u6a21\u578b.jpg\",\"width\":2560,\"height\":1707},{\"@type\":\"BreadcrumbList\",\"@id\":\"http:\/\/ktromedia.com\/?p=177969#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\/\/ktromedia.com\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"\u4f7f\u7528 torch.compile \u548c\u68af\u5ea6\u7d2f\u79ef\u66f4\u5feb\u5730\u8bad\u7ec3\u6a21\u578b\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/ktromedia.com\/#website\",\"url\":\"https:\/\/ktromedia.com\/\",\"name\":\"Ktromedia\",\"description\":\"KTRO MEDIA Crypto News\",\"publisher\":{\"@id\":\"https:\/\/ktromedia.com\/#organization\"},\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/ktromedia.com\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"en-US\"},{\"@type\":\"Organization\",\"@id\":\"https:\/\/ktromedia.com\/#organization\",\"name\":\"Ktromedia\",\"url\":\"https:\/\/ktromedia.com\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\/\/ktromedia.com\/#\/schema\/logo\/image\/\",\"url\":\"https:\/\/ktromedia.com\/wp-content\/uploads\/2025\/11\/ktroicon.png\",\"contentUrl\":\"https:\/\/ktromedia.com\/wp-content\/uploads\/2025\/11\/ktroicon.png\",\"width\":250,\"height\":250,\"caption\":\"Ktromedia\"},\"image\":{\"@id\":\"https:\/\/ktromedia.com\/#\/schema\/logo\/image\/\"},\"sameAs\":[\"https:\/\/www.facebook.com\/KTROMedia\/\",\"https:\/\/www.linkedin.com\/company\/ktro-media\/\",\"https:\/\/t.me\/ktrogroup\"]},{\"@type\":\"Person\",\"@id\":\"https:\/\/ktromedia.com\/#\/schema\/person\/612bf2fbac107722ea365932cdd35f5b\",\"name\":\"KTRO TEAM\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\/\/ktromedia.com\/#\/schema\/person\/image\/\",\"url\":\"http:\/\/ktromedia.com\/wp-content\/uploads\/2025\/10\/cropped-Untitled-design-7-1-150x150.png\",\"contentUrl\":\"http:\/\/ktromedia.com\/wp-content\/uploads\/2025\/10\/cropped-Untitled-design-7-1-150x150.png\",\"caption\":\"KTRO TEAM\"},\"description\":\"KTRO MEDIA \u662f\u4e00\u5bb6\u5168\u7403\u6027\u7684\u534e\u6587WEB3\u5a92\u4f53\u516c\u53f8\u3002\u6211\u4eec\u81f4\u529b\u4e8e\u4e3a\u533a\u5757\u94fe\u548c\u91d1\u878d\u79d1\u6280\u9886\u57df\u63d0\u4f9b\u6700\u65b0\u7684\u65b0\u95fb\u3001\u89c1\u89e3\u548c\u8d8b\u52bf\u5206\u6790\u3002\u6211\u4eec\u7684\u5b97\u65e8\u662f\u4e3a\u5168\u7403\u7528\u6237\u63d0\u4f9b\u9ad8\u8d28\u91cf\u3001\u5168\u9762\u7684\u8d44\u8baf\u670d\u52a1\uff0c\u8ba9\u4ed6\u4eec\u66f4\u597d\u5730\u4e86\u89e3\u533a\u5757\u94fe\u548c\u91d1\u878d\u79d1\u6280\u884c\u4e1a\u7684\u6700\u65b0\u52a8\u6001\u3002\u6211\u4eec\u4e5f\u5e0c\u671b\u80fd\u5e2e\u5230\u66f4\u591a\u4f18\u79c0\u7684WEB3\u4ea7\u54c1\u627e\u5230\u66f4\u591a\u66f4\u597d\u7684\u8d44\u6e90\u597d\u8ba9\u8fd9\u9886\u57df\u53d8\u5f97\u66f4\u6210\u719f\u3002 \u6211\u4eec\u7684\u62a5\u9053\u8303\u56f4\u6db5\u76d6\u4e86\u533a\u5757\u94fe\u3001\u52a0\u5bc6\u8d27\u5e01\u3001\u667a\u80fd\u5408\u7ea6\u3001DeFi\u3001NFT \u548c Web3 \u751f\u6001\u7cfb\u7edf\u7b49\u9886\u57df\u3002\u6211\u4eec\u7684\u62a5\u9053\u4e0d\u4ec5\u6765\u81ea\u884c\u4e1a\u5185\u7684\u4e13\u5bb6\uff0c\u5148\u950b\u8005\u4e5f\u5305\u62ec\u4e86\u6211\u4eec\u81ea\u5df1\u7684\u5206\u6790\u548c\u89c2\u70b9\u3002\u6211\u4eec\u5728\u5404\u4e2a\u56fd\u5bb6\u548c\u5730\u533a\u90fd\u8bbe\u6709\u56e2\u961f\uff0c\u4e3a\u8bfb\u8005\u63d0\u4f9b\u672c\u5730\u5316\u7684\u62a5\u9053\u548c\u5206\u6790\u3002 \u9664\u4e86\u65b0\u95fb\u62a5\u9053\uff0c\u6211\u4eec\u8fd8\u63d0\u4f9b\u5e02\u573a\u7814\u7a76\u548c\u54a8\u8be2\u670d\u52a1\u3002\u6211\u4eec\u7684\u4e13\u4e1a\u56e2\u961f\u53ef\u4ee5\u4e3a\u60a8\u63d0\u4f9b\u6709\u5173\u533a\u5757\u94fe\u548c\u91d1\u878d\u79d1\u6280\u884c\u4e1a\u7684\u6df1\u5165\u5206\u6790\u548c\u5e02\u573a\u8d8b\u52bf\uff0c\u5e2e\u52a9\u60a8\u505a\u51fa\u66f4\u660e\u667a\u7684\u6295\u8d44\u51b3\u7b56\u3002 \u6211\u4eec\u7684\u4f7f\u547d\u662f\u6210\u4e3a\u5168\u7403\u534e\u6587\u533a\u5757\u94fe\u548c\u91d1\u878d\u79d1\u6280\u884c\u4e1a\u6700\u53d7\u4fe1\u8d56\u7684\u4fe1\u606f\u6765\u6e90\u4e4b\u4e00\u3002\u6211\u4eec\u5c06\u7ee7\u7eed\u4e0d\u65ad\u52aa\u529b\uff0c\u4e3a\u8bfb\u8005\u63d0\u4f9b\u6700\u65b0\u3001\u6700\u5168\u9762\u3001\u6700\u53ef\u9760\u7684\u4fe1\u606f\u670d\u52a1\u3002\",\"sameAs\":[\"https:\/\/ktromedia.com\"],\"url\":\"http:\/\/ktromedia.com\/?author=1\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"\u4f7f\u7528 torch.compile \u548c\u68af\u5ea6\u7d2f\u79ef\u66f4\u5feb\u5730\u8bad\u7ec3\u6a21\u578b - Ktromedia","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"http:\/\/ktromedia.com\/?p=177969","og_locale":"en_US","og_type":"article","og_title":"\u4f7f\u7528 torch.compile \u548c\u68af\u5ea6\u7d2f\u79ef\u66f4\u5feb\u5730\u8bad\u7ec3\u6a21\u578b - Ktromedia","og_description":"\u4f7f\u7528\u6df1\u5c42\u8f6c\u6362\u5668\u67b6\u6784\u8bad\u7ec3\u8bed\u8a00\u6a21\u578b\u975e\u5e38\u8017\u65f6\u3002\u4f46\u662f\uff0c\u60a8\u53ef\u4ee5\u4f7f\u7528\u4e00\u4e9b\u6280\u672f\u6765\u52a0\u901f\u8bad\u7ec3\u3002\u5728\u672c\u6587\u4e2d\uff0c\u60a8\u5c06\u4e86\u89e3\uff1a \u4f7f\u7528 torch.compile() \u52a0\u901f\u6a21\u578b \u4f7f\u7528\u68af\u5ea6\u7d2f\u79ef\u8bad\u7ec3\u5177\u6709\u66f4\u5927\u6709\u6548\u6279\u91cf\u5927\u5c0f\u7684\u6a21\u578b \u8ba9\u6211\u4eec\u5f00\u59cb\u5427\uff01 \u4f7f\u7528 torch.compile \u548c\u68af\u5ea6\u7d2f\u79ef\u66f4\u5feb\u5730\u8bad\u7ec3\u6a21\u578b\u6444\u5f71\uff1a \u5f17\u6717\u7d22\u74e6\u00b7\u70ed\u519c\u3002\u4fdd\u7559\u4e00\u4e9b\u6743\u5229\u3002 \u6982\u8ff0 \u672c\u6587\u5206\u4e3a\u4e24\u90e8\u5206\uff1b\u4ed6\u4eec\u662f\uff1a \u4f7f\u7528 torch.compile() \u68af\u5ea6\u7d2f\u79ef \u4f7f\u7528 torch.compile \u5f53\u60a8\u7f16\u5199\u6a21\u578b\u4ee3\u7801\u5e76\u5728 PyTorch \u4e2d\u8fd0\u884c\u5b83\u65f6\uff0c\u5b83\u4f1a\u4ee5 eager \u6a21\u5f0f\u6267\u884c\u3002\u8fd9\u610f\u5473\u7740\u4ee3\u7801\u662f\u9010\u884c\u6267\u884c\u7684\uff0c\u7ed3\u679c\u5b58\u50a8\u5728\u5185\u5b58\u4e2d\u3002\u8fd9\u662f Python \u7684\u672c\u673a\u7279\u6027\uff0c\u56e0\u4e3a\u5b83\u662f\u4e00\u79cd\u89e3\u91ca\u6027\u8bed\u8a00\u3002\u60a8\u77e5\u9053\u662f\u8fd9\u79cd\u60c5\u51b5\uff0c\u56e0\u4e3a\u5f53\u60a8\u5728\u4ee3\u7801\u4e2d\u72af\u9519\u65f6\uff0c\u53ea\u6709\u8fd0\u884c\u8be5\u884c\u4ee3\u7801\u540e\u60a8\u624d\u4f1a\u770b\u5230\u8be5\u9519\u8bef\u3002 \u5728 Eager \u6a21\u5f0f\u4e0b\u8fd0\u884c\u6a21\u578b\u901f\u5ea6\u5f88\u6162\u3002\u4ece PyTorch 2.0 \u5f00\u59cb\uff0c\u60a8\u53ef\u4ee5\u4f7f\u7528 torch.compile() \u7f16\u8bd1\u6a21\u578b\u4ee5\u63d0\u9ad8\u6027\u80fd\u3002\u8fd9\u4f1a\u751f\u6210\u4e00\u4e2a\u7ecf\u8fc7\u4f18\u5316\u7684\u65b0\u6a21\u578b\u5bf9\u8c61\u3002\u5b83\u4e0e\u60a8\u4f7f\u7528\u521b\u5efa\u7684\u6a21\u578b\u5bf9\u8c61\u4e0d\u540c nn.Module\uff0c\u4f46\u5b83\u4e0e\u539f\u59cb\u6a21\u578b\u5171\u4eab\u76f8\u540c\u7684\u5f20\u91cf\u3002\u60a8\u53ef\u4ee5\u50cf\u5f80\u5e38\u4e00\u6837\u4f7f\u7528\u6b64\u7f16\u8bd1\u6a21\u578b\u8fdb\u884c\u524d\u5411\u4f20\u9012\u3001\u540e\u5411\u4f20\u9012\u548c\u4f18\u5316\u5668\u66f4\u65b0\u3002 \u6784\u5efa\u6a21\u578b\u5e76\u5c06\u5176\u7f16\u8bd1\u4e3a\u8ba1\u7b97\u56fe\u5c31\u662f TensorFlow 1.0 \u7684\u5de5\u4f5c\u65b9\u5f0f\u3002\u8fd9\u4f7f\u5f97\u8c03\u8bd5\u53d8\u5f97\u66f4\u52a0\u56f0\u96be\uff0c\u56e0\u4e3a\u60a8\u6267\u884c\u7684\u6a21\u578b\u65e0\u6cd5\u4e0e\u60a8\u7f16\u5199\u7684\u4ee3\u7801\u9010\u884c\u5339\u914d\u3002\u56e0\u6b64\uff0c\u5728\u8fd0\u884c\u8bd5\u9a8c\u5e76\u786e\u8ba4\u6a21\u578b\u6ca1\u6709\u9519\u8bef\u4e4b\u524d\uff0c\u4e0d\u5e94\u7f16\u8bd1\u6a21\u578b\u3002 \u5e76\u975e\u6240\u6709\u6a21\u578b\u90fd\u53ef\u4ee5\u7f16\u8bd1\u3002\u4f46\u662f\uff0c\u5982\u679c\u60a8\u7684\u6a21\u578b\u652f\u6301\u7f16\u8bd1\uff0c\u60a8\u5c06\u7acb\u5373\u53d7\u76ca\u4e8e\u52a0\u901f\u3002\u8981\u7f16\u8bd1\u6a21\u578b\uff0c\u60a8\u9700\u8981\u505a\u7684\u5c31\u662f\u5728\u51c6\u5907\u4f7f\u7528\u6a21\u578b\u5bf9\u8c61\u4e4b\u524d\u66ff\u6362\u5b83\uff1a &#8230; model = LlamaForPretraining(model_config).to(device) model.load_state_dict(checkpoint) model = torch.compile(model) &#8230; \u3002\u3002\u3002 \u6a21\u578b = \u9884\u8bad\u7ec3\u9a86\u9a7c\uff08\u6a21\u578b\u914d\u7f6e\uff09\u3002\u5230\uff08\u8bbe\u5907\uff09 \u6a21\u578b\u3002\u52a0\u8f7d\u72b6\u6001\u5b57\u5178\uff08\u68c0\u67e5\u7ad9\uff09 \u6a21\u578b = \u706b\u70ac\u3002\u7f16\u8bd1\uff08\u6a21\u578b\uff09","og_url":"http:\/\/ktromedia.com\/?p=177969","og_site_name":"Ktromedia","article_publisher":"https:\/\/www.facebook.com\/KTROMedia\/","article_published_time":"2026-01-16T14:31:54+00:00","og_image":[{"width":2560,"height":1707,"url":"http:\/\/ktromedia.com\/wp-content\/uploads\/2026\/01\/\u4f7f\u7528-torchcompile-\u548c\u68af\u5ea6\u7d2f\u79ef\u66f4\u5feb\u5730\u8bad\u7ec3\u6a21\u578b.jpg","type":"image\/jpeg"}],"author":"KTRO TEAM","twitter_card":"summary_large_image","twitter_misc":{"Written by":"KTRO TEAM","Est. reading time":"1 minute"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"http:\/\/ktromedia.com\/?p=177969#article","isPartOf":{"@id":"http:\/\/ktromedia.com\/?p=177969"},"author":{"name":"KTRO TEAM","@id":"https:\/\/ktromedia.com\/#\/schema\/person\/612bf2fbac107722ea365932cdd35f5b"},"headline":"\u4f7f\u7528 torch.compile \u548c\u68af\u5ea6\u7d2f\u79ef\u66f4\u5feb\u5730\u8bad\u7ec3\u6a21\u578b","datePublished":"2026-01-16T14:31:54+00:00","mainEntityOfPage":{"@id":"http:\/\/ktromedia.com\/?p=177969"},"wordCount":174,"commentCount":0,"publisher":{"@id":"https:\/\/ktromedia.com\/#organization"},"image":{"@id":"http:\/\/ktromedia.com\/?p=177969#primaryimage"},"thumbnailUrl":"http:\/\/ktromedia.com\/wp-content\/uploads\/2026\/01\/\u4f7f\u7528-torchcompile-\u548c\u68af\u5ea6\u7d2f\u79ef\u66f4\u5feb\u5730\u8bad\u7ec3\u6a21\u578b.jpg","articleSection":["\u4eba\u5de5\u667a\u80fd"],"inLanguage":"en-US","potentialAction":[{"@type":"CommentAction","name":"Comment","target":["http:\/\/ktromedia.com\/?p=177969#respond"]}]},{"@type":"WebPage","@id":"http:\/\/ktromedia.com\/?p=177969","url":"http:\/\/ktromedia.com\/?p=177969","name":"\u4f7f\u7528 torch.compile \u548c\u68af\u5ea6\u7d2f\u79ef\u66f4\u5feb\u5730\u8bad\u7ec3\u6a21\u578b - Ktromedia","isPartOf":{"@id":"https:\/\/ktromedia.com\/#website"},"primaryImageOfPage":{"@id":"http:\/\/ktromedia.com\/?p=177969#primaryimage"},"image":{"@id":"http:\/\/ktromedia.com\/?p=177969#primaryimage"},"thumbnailUrl":"http:\/\/ktromedia.com\/wp-content\/uploads\/2026\/01\/\u4f7f\u7528-torchcompile-\u548c\u68af\u5ea6\u7d2f\u79ef\u66f4\u5feb\u5730\u8bad\u7ec3\u6a21\u578b.jpg","datePublished":"2026-01-16T14:31:54+00:00","breadcrumb":{"@id":"http:\/\/ktromedia.com\/?p=177969#breadcrumb"},"inLanguage":"en-US","potentialAction":[{"@type":"ReadAction","target":["http:\/\/ktromedia.com\/?p=177969"]}]},{"@type":"ImageObject","inLanguage":"en-US","@id":"http:\/\/ktromedia.com\/?p=177969#primaryimage","url":"http:\/\/ktromedia.com\/wp-content\/uploads\/2026\/01\/\u4f7f\u7528-torchcompile-\u548c\u68af\u5ea6\u7d2f\u79ef\u66f4\u5feb\u5730\u8bad\u7ec3\u6a21\u578b.jpg","contentUrl":"http:\/\/ktromedia.com\/wp-content\/uploads\/2026\/01\/\u4f7f\u7528-torchcompile-\u548c\u68af\u5ea6\u7d2f\u79ef\u66f4\u5feb\u5730\u8bad\u7ec3\u6a21\u578b.jpg","width":2560,"height":1707},{"@type":"BreadcrumbList","@id":"http:\/\/ktromedia.com\/?p=177969#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/ktromedia.com\/"},{"@type":"ListItem","position":2,"name":"\u4f7f\u7528 torch.compile \u548c\u68af\u5ea6\u7d2f\u79ef\u66f4\u5feb\u5730\u8bad\u7ec3\u6a21\u578b"}]},{"@type":"WebSite","@id":"https:\/\/ktromedia.com\/#website","url":"https:\/\/ktromedia.com\/","name":"Ktromedia","description":"KTRO MEDIA Crypto News","publisher":{"@id":"https:\/\/ktromedia.com\/#organization"},"potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/ktromedia.com\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"en-US"},{"@type":"Organization","@id":"https:\/\/ktromedia.com\/#organization","name":"Ktromedia","url":"https:\/\/ktromedia.com\/","logo":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/ktromedia.com\/#\/schema\/logo\/image\/","url":"https:\/\/ktromedia.com\/wp-content\/uploads\/2025\/11\/ktroicon.png","contentUrl":"https:\/\/ktromedia.com\/wp-content\/uploads\/2025\/11\/ktroicon.png","width":250,"height":250,"caption":"Ktromedia"},"image":{"@id":"https:\/\/ktromedia.com\/#\/schema\/logo\/image\/"},"sameAs":["https:\/\/www.facebook.com\/KTROMedia\/","https:\/\/www.linkedin.com\/company\/ktro-media\/","https:\/\/t.me\/ktrogroup"]},{"@type":"Person","@id":"https:\/\/ktromedia.com\/#\/schema\/person\/612bf2fbac107722ea365932cdd35f5b","name":"KTRO TEAM","image":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/ktromedia.com\/#\/schema\/person\/image\/","url":"http:\/\/ktromedia.com\/wp-content\/uploads\/2025\/10\/cropped-Untitled-design-7-1-150x150.png","contentUrl":"http:\/\/ktromedia.com\/wp-content\/uploads\/2025\/10\/cropped-Untitled-design-7-1-150x150.png","caption":"KTRO TEAM"},"description":"KTRO MEDIA \u662f\u4e00\u5bb6\u5168\u7403\u6027\u7684\u534e\u6587WEB3\u5a92\u4f53\u516c\u53f8\u3002\u6211\u4eec\u81f4\u529b\u4e8e\u4e3a\u533a\u5757\u94fe\u548c\u91d1\u878d\u79d1\u6280\u9886\u57df\u63d0\u4f9b\u6700\u65b0\u7684\u65b0\u95fb\u3001\u89c1\u89e3\u548c\u8d8b\u52bf\u5206\u6790\u3002\u6211\u4eec\u7684\u5b97\u65e8\u662f\u4e3a\u5168\u7403\u7528\u6237\u63d0\u4f9b\u9ad8\u8d28\u91cf\u3001\u5168\u9762\u7684\u8d44\u8baf\u670d\u52a1\uff0c\u8ba9\u4ed6\u4eec\u66f4\u597d\u5730\u4e86\u89e3\u533a\u5757\u94fe\u548c\u91d1\u878d\u79d1\u6280\u884c\u4e1a\u7684\u6700\u65b0\u52a8\u6001\u3002\u6211\u4eec\u4e5f\u5e0c\u671b\u80fd\u5e2e\u5230\u66f4\u591a\u4f18\u79c0\u7684WEB3\u4ea7\u54c1\u627e\u5230\u66f4\u591a\u66f4\u597d\u7684\u8d44\u6e90\u597d\u8ba9\u8fd9\u9886\u57df\u53d8\u5f97\u66f4\u6210\u719f\u3002 \u6211\u4eec\u7684\u62a5\u9053\u8303\u56f4\u6db5\u76d6\u4e86\u533a\u5757\u94fe\u3001\u52a0\u5bc6\u8d27\u5e01\u3001\u667a\u80fd\u5408\u7ea6\u3001DeFi\u3001NFT \u548c Web3 \u751f\u6001\u7cfb\u7edf\u7b49\u9886\u57df\u3002\u6211\u4eec\u7684\u62a5\u9053\u4e0d\u4ec5\u6765\u81ea\u884c\u4e1a\u5185\u7684\u4e13\u5bb6\uff0c\u5148\u950b\u8005\u4e5f\u5305\u62ec\u4e86\u6211\u4eec\u81ea\u5df1\u7684\u5206\u6790\u548c\u89c2\u70b9\u3002\u6211\u4eec\u5728\u5404\u4e2a\u56fd\u5bb6\u548c\u5730\u533a\u90fd\u8bbe\u6709\u56e2\u961f\uff0c\u4e3a\u8bfb\u8005\u63d0\u4f9b\u672c\u5730\u5316\u7684\u62a5\u9053\u548c\u5206\u6790\u3002 \u9664\u4e86\u65b0\u95fb\u62a5\u9053\uff0c\u6211\u4eec\u8fd8\u63d0\u4f9b\u5e02\u573a\u7814\u7a76\u548c\u54a8\u8be2\u670d\u52a1\u3002\u6211\u4eec\u7684\u4e13\u4e1a\u56e2\u961f\u53ef\u4ee5\u4e3a\u60a8\u63d0\u4f9b\u6709\u5173\u533a\u5757\u94fe\u548c\u91d1\u878d\u79d1\u6280\u884c\u4e1a\u7684\u6df1\u5165\u5206\u6790\u548c\u5e02\u573a\u8d8b\u52bf\uff0c\u5e2e\u52a9\u60a8\u505a\u51fa\u66f4\u660e\u667a\u7684\u6295\u8d44\u51b3\u7b56\u3002 \u6211\u4eec\u7684\u4f7f\u547d\u662f\u6210\u4e3a\u5168\u7403\u534e\u6587\u533a\u5757\u94fe\u548c\u91d1\u878d\u79d1\u6280\u884c\u4e1a\u6700\u53d7\u4fe1\u8d56\u7684\u4fe1\u606f\u6765\u6e90\u4e4b\u4e00\u3002\u6211\u4eec\u5c06\u7ee7\u7eed\u4e0d\u65ad\u52aa\u529b\uff0c\u4e3a\u8bfb\u8005\u63d0\u4f9b\u6700\u65b0\u3001\u6700\u5168\u9762\u3001\u6700\u53ef\u9760\u7684\u4fe1\u606f\u670d\u52a1\u3002","sameAs":["https:\/\/ktromedia.com"],"url":"http:\/\/ktromedia.com\/?author=1"}]}},"_links":{"self":[{"href":"http:\/\/ktromedia.com\/index.php?rest_route=\/wp\/v2\/posts\/177969","targetHints":{"allow":["GET"]}}],"collection":[{"href":"http:\/\/ktromedia.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/ktromedia.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"http:\/\/ktromedia.com\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"http:\/\/ktromedia.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=177969"}],"version-history":[{"count":1,"href":"http:\/\/ktromedia.com\/index.php?rest_route=\/wp\/v2\/posts\/177969\/revisions"}],"predecessor-version":[{"id":177971,"href":"http:\/\/ktromedia.com\/index.php?rest_route=\/wp\/v2\/posts\/177969\/revisions\/177971"}],"wp:featuredmedia":[{"embeddable":true,"href":"http:\/\/ktromedia.com\/index.php?rest_route=\/wp\/v2\/media\/177970"}],"wp:attachment":[{"href":"http:\/\/ktromedia.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=177969"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/ktromedia.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=177969"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/ktromedia.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=177969"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}