{"id":2624,"date":"2025-02-20T16:56:22","date_gmt":"2025-02-20T08:56:22","guid":{"rendered":"https:\/\/aws-oncloudai.com\/?p=2624"},"modified":"2025-02-20T16:56:22","modified_gmt":"2025-02-20T08:56:22","slug":"zaslydtsscsaxn","status":"publish","type":"post","link":"https:\/\/aws-oncloudai.com\/zh_hk\/zaslydtsscsaxn\/","title":{"rendered":"\u5728AWS \u4e0a\u5229\u7528DeepSeek-R1 \u63d0\u5347\u751f\u6210\u5f0fAI \u6548\u80fd"},"content":{"rendered":"<p>\u96a8\u8457\u751f\u6210\u5f0fAI \u6280\u8853\u7684\u5feb\u901f\u767c\u5c55\uff0c\u5982\u4f55\u6709\u6548\u7387\u3001\u4f4e\u6210\u672c\u5730\u90e8\u7f72\u548c\u904b\u884c\u9019\u4e9b\u5f37\u5927\u7684\u6a21\u578b\u6210\u70ba\u4e86\u4f01\u696d\u548c\u958b\u767c\u8005\u9762\u81e8\u7684\u91cd\u8981\u6311\u6230\u3002<strong>DeepSeek-R1<\/strong>\uff0c\u4f5c\u70ba\u4e00\u6b3e\u7a81\u7834\u6027\u7684\u751f\u6210\u5f0fAI \u57fa\u790e\u6a21\u578b\uff0c\u4e0d\u50c5\u5177\u5099\u5f37\u5927\u7684\u63a8\u7406\u80fd\u529b\uff0c\u5728\u6027\u80fd\u548c\u8cc7\u6e90\u4f7f\u7528\u65b9\u9762\u4e5f\u8868\u73fe\u51fa\u8272\u3002\u501f\u52a9 <strong>Amazon Web Services\uff08AWS\uff09<\/strong> \u7684\u5f37\u5927\u96f2\u7aef\u5e73\u53f0\uff0c\u4f7f\u7528\u8005\u53ef\u4ee5\u900f\u904e\u9748\u6d3b\u7684\u90e8\u7f72\u65b9\u5f0f\uff0c\u7121\u7e2b\u6574\u5408DeepSeek-R1\uff0c\u70ba\u5404\u7a2e\u7522\u696d\u7684AI \u61c9\u7528\u63d0\u4f9b\u652f\u63f4\u3002\u672c\u6587\u5c07\u8a73\u7d30\u4ecb\u7d39\u5982\u4f55\u5728AWS \u4e0a\u6709\u6548\u904b\u7528DeepSeek-R1 \u6a21\u578b\uff0c\u5354\u52a9\u60a8\u5728\u751f\u6210\u5f0fAI \u9818\u57df\u53d6\u5f97\u66f4\u5927\u7a81\u7834\u3002<\/p>\n<p>&nbsp;<\/p>\n<h4><strong>DeepSeek-R1 \u7c21\u4ecb<\/strong><\/h4>\n<p>DeepSeek-R1 \u662f\u4e00\u7a2e\u5148\u9032\u7684\u751f\u6210\u5f0fAI \u6a21\u578b\uff0c\u5b83\u900f\u904e\u5c07\u5f37\u5316\u5b78\u7fd2\u8207\u6df7\u5408\u5c08\u5bb6\u67b6\u69cb\u7d50\u5408\uff0c\u63d0\u4f9b\u9ad8\u6548\u7684\u63a8\u7406\u548c\u8655\u7406\u80fd\u529b\uff0c\u4e14\u6210\u672c\u4f4e\u5ec9\u3002\u8a72\u6a21\u578b\u4e0d\u50c5\u80fd\u5728\u8cc7\u6e90\u6d88\u8017\u4e0a\u4fdd\u6301\u4f4e\u8abf\uff0c\u9084\u80fd\u4fdd\u7559\u5f37\u5927\u7684\u6578\u5b78\u548c\u7a0b\u5f0f\u8a2d\u8a08\u80fd\u529b\uff0c\u975e\u5e38\u9069\u5408\u9700\u8981\u9ad8\u6548\u904b\u7b97\u7684\u5834\u666f\u3002<\/p>\n<p>\u672c\u6587\u5c07\u6df1\u5165\u63a2\u8a0e\u5982\u4f55\u5728AWS \u74b0\u5883\u4e2d\u6709\u6548\u5229\u7528DeepSeek-R1 \u6a21\u578b\u8b8a\u9ad4\uff0c\u501f\u52a9\u5168\u7403AWS \u5c08\u5bb6\u7684\u8cc7\u6e90\u548c\u5be6\u969b\u7a0b\u5f0f\u78bc\u7bc4\u4f8b\u4f86\u5be6\u73fe\u6700\u4f73\u61c9\u7528\u3002<\/p>\n<p>&nbsp;<\/p>\n<h4>\u7121\u4f3a\u670d\u5668\u63a8\u7406\uff1a\u4f7f\u7528Amazon Bedrock Custom Model Import<\/h4>\n<p>DeepSeek-R1 \u63a8\u51fa\u4e86\u8207Llama Grouped Query Attention\uff08GQA\uff09\u67b6\u69cb\u76f8\u5bb9\u7684\u7cbe\u7c21\u7248\u672c\uff0c\u5305\u62ec8B \u548c70B \u53c3\u6578\u6a21\u578b\uff08DeepSeek-R1-Distill-Llama-8B \u548cDeepSeek-R1-Distill-Llama-70B\uff09\u3002\u900f\u904e\u5c0e\u51fa\u6a21\u578b\u6b0a\u91cd\u4e26\u4f7f\u7528 <strong>Amazon Bedrock Custom Model Import<\/strong>\uff0c\u4f7f\u7528\u8005\u53ef\u4ee5\u5728AWS \u4e0a\u9032\u884c\u7121\u4f3a\u670d\u5668\u63a8\u7406\uff0c\u4e26\u4e14\u53ef\u4ee5\u5229\u7528Bedrock \u7684\u7d71\u4e00API \u548c\u5148\u9032\u5de5\u5177\uff08\u5982Guardrails\uff09\u4f86\u78ba\u4fddAI \u7684\u8cac\u4efb\u6027\u548c\u5b89\u5168\u6027\u3002\u7279\u5225\u662f\uff0c\u900f\u904e\u9019\u7a2e\u65b9\u5f0f\u90e8\u7f72\u7684\u6a21\u578b\u5c07\u4ee5\u63a8\u7406\u91cf\u8a08\u8cbb\uff0c\u6a21\u578b\u7684\u6210\u672c\u6548\u76ca\u986f\u8457\u3002\u4f8b\u5982\uff0c8B \u6a21\u578b\u9700\u89812 \u500b\u6a21\u578b\u55ae\u5143\uff0c\u800c70B \u6a21\u578b\u5247\u9700\u89818 \u500b\u55ae\u5143\u3002\u5c0e\u5165\u904e\u7a0b\u6700\u591a\u9700\u898130 \u5206\u9418\uff0c\u4e14\u51b7\u555f\u52d5\u5ef6\u9072\u70ba10 \u79d2\u3002<\/p>\n<p>&nbsp;<\/p>\n<h4>\u5728CPU \u4e0a\u9032\u884c\u5373\u6642\u63a8\u7406\uff1a\u5229\u7528AWS Graviton \u548cAmazon SageMaker<\/h4>\n<p>\u900f\u904e\u91cf\u5316\u65b9\u6cd5\uff0cDeepSeek-R1 \u53ef\u4ee5\u7121\u9700GPU\uff0c\u5728\u57fa\u65bcCPU \u7684\u5be6\u4f8b\u4e0a\u9032\u884c\u9ad8\u6548\u7684\u5373\u6642\u63a8\u7406\u3002\u6b64\u904e\u7a0b\u652f\u63f4\u4f7f\u7528\u5982 <strong>LLama.cpp<\/strong> \u7684\u6846\u67b6\u4f86\u76f8\u5bb9CPU \u63a8\u7406\u3002\u900f\u904e\u9078\u64c7\u9069\u5408\u7684\u91cf\u5316\u65b9\u5f0f\uff08\u4f8b\u59824 \u4f4d\u62165 \u4f4d\u91cf\u5316\uff09\uff0c\u53ef\u4ee5\u7372\u5f97\u6700\u4f73\u7684\u901f\u5ea6\u548c\u6e96\u78ba\u5ea6\u3002<\/p>\n<p>AWS \u63d0\u4f9b\u4e86\u773e\u591a\u512a\u5316\u5f8c\u7684\u958b\u6e90\u91cf\u5316\u7248\u672c\uff0c\u5305\u62ec\u4f86\u81eaLMStudio \u548cUnsloth \u7b49\u793e\u7fa4\u767c\u5e03\u7684\u6a21\u578b\u7248\u672c\uff0c\u5e6b\u52a9\u4f7f\u7528\u8005\u66f4\u65b9\u4fbf\u90e8\u7f72DeepSeek-R1\u3002<\/p>\n<p>&nbsp;<\/p>\n<h4>GPU \u4e0a\u7684\u5373\u6642\u63a8\u7406\uff1aAmazon SageMaker \u548cAmazon Bedrock Marketplace<\/h4>\n<p>\u5c0d\u65bc\u66f4\u9ad8\u7aef\u7684\u4f7f\u7528\u5834\u666f\uff0cAWS \u9084\u652f\u63f4\u900f\u904e <strong>Amazon SageMaker<\/strong> \u548c <strong>Amazon Bedrock Marketplace<\/strong> \u5728GPU \u5be6\u4f8b\u4e0a\u9032\u884cDeepSeek-R1 \u6a21\u578b\u7684\u90e8\u7f72\u3002\u900f\u904e <strong>Bedrock Marketplace<\/strong>\uff0c\u4f7f\u7528\u8005\u53ef\u4ee5\u8f15\u9b06\u5730\u5728\u96f2\u7aef\u90e8\u7f72671B \u53c3\u6578\u7684DeepSeek-R1 \u6a21\u578b\uff0c\u50c5\u9700\u5e7e\u6b65\u64cd\u4f5c\u5373\u53ef\u5b8c\u6210\u3002\u5efa\u8b70\u7684\u5be6\u4f8b\u914d\u7f6e\u5982p5e.48xlarge\uff0c\u9069\u7528\u65bc\u5927\u898f\u6a21\u6a21\u578b\u63a8\u7406\u3002<\/p>\n<p>&nbsp;<\/p>\n<h4>AWS Trainium \u4e0a\u7684\u5373\u6642\u63a8\u7406<\/h4>\n<p>AWS Trainium \u662f\u5c08\u70ba\u751f\u6210\u5f0fAI \u5de5\u4f5c\u8ca0\u8f09\u6700\u4f73\u5316\u7684AI \u904b\u7b97\u6676\u7247\u3002\u4f7f\u7528 <strong>AWS Neuron SDK<\/strong>\uff0cDeepSeek-R1 \u53ef\u4ee5\u5728Trainium \u5be6\u4f8b\u4e0a\u5be6\u73fe\u9ad8\u6548\u63a8\u7406\u3002\u5118\u7ba1\u652f\u63f4\u4e00\u4e9b\u6d41\u884c\u7684\u63a8\u7406\u512a\u5316\u5eab\uff0c\u5982vLLM\uff0c\u4f46Trainium \u4ecd\u5728\u4e0d\u65b7\u512a\u5316\uff0c\u4ee5\u63d0\u4f9b\u66f4\u597d\u7684\u63a8\u7406\u6548\u7387\u548c\u9748\u6d3b\u6027\uff0c\u7279\u5225\u662f\u5c0d\u65bc\u5927\u898f\u6a21AI \u4efb\u52d9\u3002<\/p>\n<p>&nbsp;<\/p>\n<h4>\u64f4\u5c55\u63a8\u7406\u80fd\u529b\uff1aRay on EKS \u548cAmazon EKS Auto Mode<\/h4>\n<p>\u5982\u679c\u60a8\u5e0c\u671b\u5728\u63a8\u7406\u57fa\u790e\u67b6\u69cb\u4e0a\u7372\u5f97\u66f4\u5927\u7684\u9748\u6d3b\u6027\u548c\u63a7\u5236\u6b0a\uff0c\u53ef\u4ee5\u9078\u64c7 <strong>Amazon EKS Auto Mode<\/strong> \u6216\u4f7f\u7528 <strong>Ray on EKS<\/strong> \u4f86\u64f4\u5c55DeepSeek-R1 \u7684\u63a8\u7406\u80fd\u529b\u3002\u9019\u4e9bKubernetes \u5de5\u5177\u53ef\u4ee5\u5e6b\u52a9\u60a8\u9748\u6d3b\u7ba1\u7406\u548c\u64f4\u5c55AI \u5de5\u4f5c\u8ca0\u8f09\uff0c\u81ea\u52d5\u5316\u904b\u7b97\u3001\u5132\u5b58\u548c\u7db2\u8def\u7684\u7ba1\u7406\u3002<\/p>\n<p>&nbsp;<\/p>\n<h4>Hugging Face \u90e8\u7f72\u8207\u5fae\u8abf<\/h4>\n<p>DeepSeek-R1 \u5df2\u5728 <strong>Hugging Face<\/strong> \u4e0a\u767c\u5e03\uff0c\u4e26\u8207AWS \u6df1\u5ea6\u96c6\u6210\uff0c\u4f7f\u958b\u767c\u8005\u80fd\u5920\u5728AWS \u5e73\u53f0\u4e0a\u65b9\u4fbf\u5730\u90e8\u7f72\u548c\u5fae\u8abfDeepSeek \u6a21\u578b\u3002\u900f\u904e <strong>Hugging Face Text Generation Inference\uff08TGI\uff09<\/strong>\uff0c\u958b\u767c\u8005\u53ef\u4ee5\u5feb\u901f\u5c07\u5176\u751f\u6210\u5f0fAI \u61c9\u7528\u7a0b\u5f0f\u90e8\u7f72\u5230\u96f2\u7aef\uff0c\u4e26\u5229\u7528Hugging Face \u5f37\u5927\u7684\u793e\u7fa4\u652f\u63f4\u548c\u8cc7\u6e90\u3002<\/p>\n<p>&nbsp;<\/p>\n<h4>\u5fae\u8abf\uff1a\u5728Amazon SageMaker \u4e0a\u9032\u884c\u9ad8\u6548\u80fd\u5fae\u8abf<\/h4>\n<p>\u7531\u65bcDeepSeek-R1 \u7684\u9ad8\u6548\u6027\uff0c\u5b83\u662f\u9032\u884c\u5fae\u8abf\u7684\u7406\u60f3\u9078\u64c7\u3002 AWS \u63d0\u4f9b\u77ad\u5982 <strong>PyTorch FSDP<\/strong> \u548c <strong>QLoRa<\/strong> \u7b49\u6d41\u884c\u5eab\uff0c\u5e6b\u52a9\u7528\u6236\u5728 <strong>Amazon SageMaker<\/strong> \u4e0a\u9ad8\u6548\u57f7\u884c\u5fae\u8abf\uff0c\u5ba2\u88fd\u5316\u6a21\u578b\u4ee5\u6eff\u8db3\u7279\u5b9a\u7684\u4efb\u52d9\u9700\u6c42\u3002<\/p>\n<p>&nbsp;<\/p>\n<h4>\u603b\u7ed3<\/h4>\n<p>DeepSeek-R1 \u6a21\u578b\u6191\u85c9\u5176\u5728AI \u63a8\u7406\u548c\u8a08\u7b97\u4e0a\u7684\u51fa\u8272\u6027\u80fd\uff0c\u5df2\u7d93\u5728\u751f\u6210\u5f0fAI \u9818\u57df\u5f15\u767c\u4e86\u5ee3\u6cdb\u95dc\u6ce8\u3002 AWS \u7684\u5f37\u5927\u57fa\u790e\u67b6\u69cb\u70baDeepSeek-R1 \u63d0\u4f9b\u4e86\u9748\u6d3b\u3001\u4f4e\u6210\u672c\u7684\u90e8\u7f72\u548c\u63a8\u7406\u65b9\u6848\uff0c\u7121\u8ad6\u662f\u900f\u904e <strong>Amazon Bedrock<\/strong> \u9084\u662f <strong>Amazon SageMaker<\/strong>\uff0c\u90fd\u80fd\u5e6b\u52a9\u958b\u767c\u8005\u5145\u5206\u767c\u63ee\u6f5b\u529b\uff0c\u5275\u9020\u9ad8\u6548\u7684\u751f\u6210\u5f0fAI \u61c9\u7528\u3002<\/p>\n<p>\u5c0d\u65bc\u9700\u8981\u8655\u7406\u5927\u898f\u6a21AI \u4efb\u52d9\u7684\u7528\u6236\uff0c\u5efa\u8b70\u5f9eDeepSeek-R1 \u7cbe\u7c21\u7248\uff08\u59828B \u621670B \u53c3\u6578\u6a21\u578b\uff09\u958b\u59cb\u8a55\u4f30\uff0c\u6839\u64da\u5be6\u969b\u9700\u6c42\u9010\u6b65\u64f4\u5c55\u5230\u66f4\u5927\u898f\u6a21\u7684671B \u53c3\u6578\u6a21\u578b\u3002<\/p>","protected":false},"excerpt":{"rendered":"<p>\u96a8\u8457\u751f\u6210\u5f0fAI \u6280\u8853\u7684\u5feb\u901f\u767c\u5c55\uff0c\u5982\u4f55\u6709\u6548\u7387\u3001\u4f4e\u6210\u672c\u5730\u90e8\u7f72\u548c\u904b\u884c\u9019\u4e9b\u5f37\u5927\u7684\u6a21\u578b\u6210\u70ba\u4e86\u4f01\u696d\u548c\u958b\u767c\u8005\u9762\u81e8\u7684\u91cd\u8981\u6311\u6230\u3002 DeepSeek-R1\uff0c\u4f5c\u70ba\u4e00\u6b3e\u7a81\u7834\u6027\u7684\u751f\u6210\u5f0fAI \u57fa\u790e\u6a21\u578b\uff0c\u4e0d\u50c5\u5177\u5099\u5f37\u5927\u7684\u63a8\u7406\u80fd\u529b\uff0c\u5728\u6027\u80fd\u548c\u8cc7\u6e90\u4f7f\u7528\u65b9\u9762\u4e5f\u8868\u73fe\u51fa\u8272\u3002\u900f\u904eAmazon Web Services\uff08AWS\uff09 \u7684\u5f37\u5927\u96f2\u7aef\u5e73\u53f0\uff0c\u4f7f\u7528\u8005\u53ef\u4ee5\u900f\u904e\u9748\u6d3b\u7684\u90e8\u7f72\u65b9\u5f0f\uff0c\u7121\u7e2b\u6574\u5408DeepSeek-R1\uff0c\u70ba\u5404\u7a2e\u7522\u696d\u7684AI \u61c9\u7528\u63d0\u4f9b\u652f\u63f4\u3002<\/p>","protected":false},"author":1,"featured_media":2627,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[65],"tags":[],"class_list":["post-2624","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-technical-sharing"],"_links":{"self":[{"href":"https:\/\/aws-oncloudai.com\/zh_hk\/wp-json\/wp\/v2\/posts\/2624","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/aws-oncloudai.com\/zh_hk\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/aws-oncloudai.com\/zh_hk\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/aws-oncloudai.com\/zh_hk\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/aws-oncloudai.com\/zh_hk\/wp-json\/wp\/v2\/comments?post=2624"}],"version-history":[{"count":0,"href":"https:\/\/aws-oncloudai.com\/zh_hk\/wp-json\/wp\/v2\/posts\/2624\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/aws-oncloudai.com\/zh_hk\/wp-json\/wp\/v2\/media\/2627"}],"wp:attachment":[{"href":"https:\/\/aws-oncloudai.com\/zh_hk\/wp-json\/wp\/v2\/media?parent=2624"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/aws-oncloudai.com\/zh_hk\/wp-json\/wp\/v2\/categories?post=2624"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/aws-oncloudai.com\/zh_hk\/wp-json\/wp\/v2\/tags?post=2624"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}