{"id":26466,"date":"2025-02-23T12:51:52","date_gmt":"2025-02-23T04:51:52","guid":{"rendered":"https:\/\/www.aisharenet.com\/?p=26466"},"modified":"2025-02-23T12:51:52","modified_gmt":"2025-02-23T04:51:52","slug":"evo2","status":"publish","type":"post","link":"https:\/\/www.kdjingpai.com\/pt\/evo2\/","title":{"rendered":"Evo2\uff1a\u652f\u6301\u57fa\u56e0\u7ec4\u5efa\u6a21\u4e0e\u8bbe\u8ba1\u7684\u5f00\u6e90\u751f\u7269AI\u5de5\u5177"},"content":{"rendered":"<p>Arc Institute Evo 2 \u662f\u4e00\u4e2a\u4e13\u6ce8\u4e8e\u57fa\u56e0\u7ec4\u5efa\u6a21\u4e0e\u8bbe\u8ba1\u7684\u5f00\u6e90\u9879\u76ee\uff0c\u7531\u4f4d\u4e8e\u7f8e\u56fd\u52a0\u5dde\u5e15\u6d1b\u963f\u5c14\u6258\u7684\u975e\u8425\u5229\u7814\u7a76\u673a\u6784 Arc Institute \u5f00\u53d1\uff0c\u5e76\u4e0e NVIDIA \u7b49\u5408\u4f5c\u4f19\u4f34\u534f\u4f5c\u63a8\u51fa\u3002\u8be5\u9879\u76ee\u901a\u8fc7\u524d\u6cbf\u6df1\u5ea6\u5b66\u4e60\u6280\u672f\uff0c\u6784\u5efa\u4e86\u80fd\u591f\u5904\u7406 DNA\u3001RNA \u548c\u86cb\u767d\u8d28\u7684\u751f\u7269\u57fa\u7840\u6a21\u578b\uff0c\u9002\u7528\u4e8e\u751f\u547d\u79d1\u5b66\u9886\u57df\u7684\u9884\u6d4b\u4e0e\u751f\u6210\u4efb\u52a1\u3002Evo 2 \u5728\u8d85\u8fc7 9 \u4e07\u4ebf\u6838\u82f7\u9178\u7684\u591a\u6837\u5316\u57fa\u56e0\u7ec4\u6570\u636e\u4e0a\u8bad\u7ec3\uff0c\u62e5\u6709\u9ad8\u8fbe 400 \u4ebf\u53c2\u6570\uff0c\u652f\u6301\u957f\u8fbe 100 \u4e07\u78b1\u57fa\u7684\u4e0a\u4e0b\u6587\u957f\u5ea6\u3002\u5176\u4ee3\u7801\u3001\u8bad\u7ec3\u6570\u636e\u548c\u6a21\u578b\u6743\u91cd\u5b8c\u5168\u5f00\u6e90\uff0c\u6258\u7ba1\u4e8e GitHub\uff0c\u65e8\u5728\u52a0\u901f\u751f\u7269\u5de5\u7a0b\u4e0e\u533b\u5b66\u7814\u7a76\u7684\u53d1\u5c55\u3002\u65e0\u8bba\u662f\u7814\u7a76\u4eba\u5458\u8fd8\u662f\u5f00\u53d1\u8005\uff0c\u90fd\u53ef\u4ee5\u901a\u8fc7\u8be5\u5de5\u5177\u63a2\u7d22\u57fa\u56e0\u7ec4\u7684\u5965\u79d8\uff0c\u8bbe\u8ba1\u65b0\u7684\u751f\u7269\u5e8f\u5217\u3002<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" class=\"aligncenter size-full wp-image-26468\" title=\"Evo2\uff1a\u652f\u6301\u57fa\u56e0\u7ec4\u5efa\u6a21\u4e0e\u8bbe\u8ba1\u7684\u5f00\u6e90\u751f\u7269AI\u5de5\u5177-1\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/02\/c02c428bf5832eb.png\" alt=\"Evo2\uff1a\u652f\u6301\u57fa\u56e0\u7ec4\u5efa\u6a21\u4e0e\u8bbe\u8ba1\u7684\u5f00\u6e90\u751f\u7269AI\u5de5\u5177-1\" width=\"826\" height=\"462\" srcset=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/02\/c02c428bf5832eb.png 826w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/02\/c02c428bf5832eb-768x430.png 768w\" sizes=\"auto, (max-width: 826px) 100vw, 826px\" \/><\/p>\n<p>&nbsp;<\/p>\n<h2>\u529f\u80fd\u5217\u8868<\/h2>\n<ul>\n<li>\u652f\u6301\u8de8\u751f\u547d\u9886\u57df\u7684\u57fa\u56e0\u7ec4\u5efa\u6a21\uff1a\u80fd\u591f\u5728\u7ec6\u83cc\u3001\u53e4\u83cc\u548c\u771f\u6838\u751f\u7269\u7684\u57fa\u56e0\u7ec4\u4e0a\u8fdb\u884c\u9884\u6d4b\u4e0e\u8bbe\u8ba1\u3002<\/li>\n<li>\u957f\u5e8f\u5217\u5904\u7406\u80fd\u529b\uff1a\u5904\u7406\u9ad8\u8fbe 100 \u4e07\u78b1\u57fa\u7684 DNA \u5e8f\u5217\uff0c\u9002\u7528\u4e8e\u8d85\u957f\u4e0a\u4e0b\u6587\u7684\u5206\u6790\u4efb\u52a1\u3002<\/li>\n<li>DNA \u751f\u6210\u4e0e\u4f18\u5316\uff1a\u6839\u636e\u8f93\u5165\u7684\u5e8f\u5217\u6216\u7269\u79cd\u63d0\u793a\uff0c\u751f\u6210\u65b0\u7684 DNA \u5e8f\u5217\uff0c\u5e76\u63d0\u4f9b\u7f16\u7801\u533a\u6807\u6ce8\u3002<\/li>\n<li>\u96f6\u6837\u672c\u53d8\u4f53\u6548\u5e94\u9884\u6d4b\uff1a\u65e0\u9700\u989d\u5916\u8bad\u7ec3\u5373\u53ef\u9884\u6d4b\u57fa\u56e0\u53d8\u4f53\u7684\u751f\u7269\u5b66\u5f71\u54cd\uff0c\u4f8b\u5982 BRCA1 \u53d8\u4f53\u7684\u6548\u5e94\u5206\u6790\u3002<\/li>\n<li>\u5f00\u6e90\u6570\u636e\u96c6\u4e0e\u6a21\u578b\uff1a\u63d0\u4f9b\u9884\u8bad\u7ec3\u6a21\u578b\u548c OpenGenome2 \u6570\u636e\u96c6\uff0c\u652f\u6301\u4e8c\u6b21\u5f00\u53d1\u4e0e\u7814\u7a76\u3002<\/li>\n<li>\u591aGPU\u5e76\u884c\u8ba1\u7b97\u652f\u6301\uff1a\u901a\u8fc7 Vortex \u6846\u67b6\u81ea\u52a8\u5206\u914d\u591a\u5757 GPU \u8d44\u6e90\uff0c\u63d0\u5347\u5927\u89c4\u6a21\u8ba1\u7b97\u6548\u7387\u3002<\/li>\n<li>\u4e0e NVIDIA BioNeMo \u96c6\u6210\uff1a\u65e0\u7f1d\u63a5\u5165 NVIDIA \u7684\u751f\u7269\u8ba1\u7b97\u5e73\u53f0\uff0c\u6269\u5c55\u5e94\u7528\u573a\u666f\u3002<\/li>\n<li>\u53ef\u89c6\u5316\u4e0e\u89e3\u91ca\u5de5\u5177\uff1a\u7ed3\u5408 Goodfire \u7684\u89e3\u91ca\u6027\u53ef\u89c6\u5316\u5668\uff0c\u63ed\u793a\u6a21\u578b\u8bc6\u522b\u7684\u751f\u7269\u7279\u5f81\u548c\u6a21\u5f0f\u3002<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<h2>\u4f7f\u7528\u5e2e\u52a9<\/h2>\n<h3>\u5b89\u88c5\u6d41\u7a0b<\/h3>\n<p>\u8981\u5728\u672c\u5730\u4f7f\u7528 Evo 2\uff0c\u9700\u8981\u4e00\u5b9a\u7684\u8ba1\u7b97\u8d44\u6e90\u548c\u73af\u5883\u914d\u7f6e\u3002\u4ee5\u4e0b\u662f\u8be6\u7ec6\u7684\u5b89\u88c5\u6b65\u9aa4\uff1a<\/p>\n<h4>1. \u73af\u5883\u51c6\u5907<\/h4>\n<ul>\n<li><strong>\u64cd\u4f5c\u7cfb\u7edf<\/strong>\uff1a\u63a8\u8350\u4f7f\u7528 Linux\uff08\u5982 Ubuntu\uff09\u6216 macOS\uff0cWindows \u7528\u6237\u9700\u5b89\u88c5 WSL2\u3002<\/li>\n<li><strong>\u786c\u4ef6\u8981\u6c42<\/strong>\uff1a\u81f3\u5c11 1 \u5757 NVIDIA GPU\uff08\u5efa\u8bae\u591a\u5757 GPU \u4ee5\u652f\u6301 40B \u6a21\u578b\uff09\uff0c\u663e\u5b58\u6700\u4f4e 16GB\uff08\u5982 A100 \u6216 RTX 3090\uff09\u3002<\/li>\n<li><strong>\u8f6f\u4ef6\u4f9d\u8d56<\/strong>\uff1a\u786e\u4fdd\u5b89\u88c5 Git\u3001Python 3.8+\u3001PyTorch\uff08\u652f\u6301 CUDA\uff09\u548c pip\u3002<\/li>\n<\/ul>\n<h4>2. \u514b\u9686\u4ee3\u7801\u4ed3\u5e93<\/h4>\n<p>\u6253\u5f00\u7ec8\u7aef\uff0c\u8fd0\u884c\u4ee5\u4e0b\u547d\u4ee4\u83b7\u53d6 Evo 2 \u6e90\u7801\uff1a<\/p>\n<pre><code>git clone --recurse-submodules git@github.com:ArcInstitute\/evo2.git  \r\ncd evo2\r\n<\/code><\/pre>\n<p>\u6ce8\u610f\uff1a<code>--recurse-submodules<\/code>\u00a0\u786e\u4fdd\u6240\u6709\u5b50\u6a21\u5757\u4e5f\u88ab\u4e0b\u8f7d\u3002<\/p>\n<h4>3. \u5b89\u88c5\u4f9d\u8d56<\/h4>\n<p>\u5728\u9879\u76ee\u6839\u76ee\u5f55\u4e0b\u8fd0\u884c\uff1a<\/p>\n<pre><code>pip install .\r\n<\/code><\/pre>\n<p>\u5982\u679c\u9047\u5230\u95ee\u9898\uff0c\u53ef\u5c1d\u8bd5\u4ece Vortex \u5b89\u88c5\uff08\u5177\u4f53\u89c1 GitHub README\uff09\u3002\u5b89\u88c5\u5b8c\u6210\u540e\uff0c\u8fd0\u884c\u6d4b\u8bd5\u9a8c\u8bc1\uff1a<\/p>\n<pre><code>python -m evo2.test\r\n<\/code><\/pre>\n<p>\u82e5\u8f93\u51fa\u65e0\u62a5\u9519\uff0c\u8bf4\u660e\u5b89\u88c5\u6210\u529f\u3002<\/p>\n<h4>4. \u4e0b\u8f7d\u9884\u8bad\u7ec3\u6a21\u578b<\/h4>\n<p>Evo 2 \u63d0\u4f9b\u591a\u79cd\u6a21\u578b\u7248\u672c\uff08\u5982 1B\u30017B\u300140B \u53c2\u6570\uff09\uff0c\u53ef\u4ece Hugging Face \u6216 GitHub Releases \u4e0b\u8f7d\u3002\u4f8b\u5982\uff1a<\/p>\n<pre><code>wget https:\/\/huggingface.co\/arcinstitute\/evo2_7b\/resolve\/main\/evo2_7b_base.pt\r\n<\/code><\/pre>\n<p>\u5c06\u6a21\u578b\u6587\u4ef6\u653e\u5165\u672c\u5730\u76ee\u5f55\uff0c\u4f9b\u540e\u7eed\u52a0\u8f7d\u4f7f\u7528\u3002<\/p>\n<h3>\u5982\u4f55\u4f7f\u7528<\/h3>\n<p>\u5b89\u88c5\u5b8c\u6210\u540e\uff0cEvo 2 \u7684\u6838\u5fc3\u529f\u80fd\u53ef\u901a\u8fc7 Python \u811a\u672c\u8c03\u7528\u3002\u4ee5\u4e0b\u662f\u4e3b\u8981\u529f\u80fd\u7684\u8be6\u7ec6\u64cd\u4f5c\u6d41\u7a0b\uff1a<\/p>\n<h4>\u529f\u80fd 1\uff1a\u751f\u6210 DNA \u5e8f\u5217<\/h4>\n<p>Evo 2 \u53ef\u4ee5\u6839\u636e\u8f93\u5165\u7684 DNA \u7247\u6bb5\u751f\u6210\u5ef6\u7eed\u5e8f\u5217\u3002\u64cd\u4f5c\u6b65\u9aa4\u5982\u4e0b\uff1a<\/p>\n<ol>\n<li><strong>\u52a0\u8f7d\u6a21\u578b<\/strong>\uff1a\n<pre><code>from evo2 import Evo2  \r\nmodel = Evo2('evo2_7b')  # \u4f7f\u7528 7B \u53c2\u6570\u6a21\u578b\r\n<\/code><\/pre>\n<\/li>\n<li><strong>\u8f93\u5165\u63d0\u793a\u5e76\u751f\u6210<\/strong>\uff1a\n<pre><code>prompt = [\"ACGT\"]  # \u8f93\u5165\u521d\u59cb DNA \u5e8f\u5217  \r\noutput = model.generate(prompt_seqs=prompt, n_tokens=400, temperature=1.0, top_k=4)  \r\nprint(output.sequences[0])  # \u8f93\u51fa\u751f\u6210\u7684 400 \u4e2a\u6838\u82f7\u9178\u5e8f\u5217\r\n<\/code><\/pre>\n<\/li>\n<li><strong>\u7ed3\u679c\u89e3\u91ca<\/strong>\uff1a\u751f\u6210\u7684\u5e8f\u5217\u53ef\u7528\u4e8e\u4e0b\u6e38\u751f\u7269\u5b66\u5206\u6790\uff0ctemperature \u53c2\u6570\u63a7\u5236\u968f\u673a\u6027\uff0ctop_k \u9650\u5236\u91c7\u6837\u8303\u56f4\u3002<\/li>\n<\/ol>\n<h4>\u529f\u80fd 2\uff1a\u96f6\u6837\u672c\u53d8\u4f53\u6548\u5e94\u9884\u6d4b<\/h4>\n<p>\u4ee5 BRCA1 \u57fa\u56e0\u4e3a\u4f8b\uff0c\u9884\u6d4b\u53d8\u4f53\u7684\u751f\u7269\u5b66\u5f71\u54cd\uff1a<\/p>\n<ol>\n<li><strong>\u51c6\u5907\u6570\u636e<\/strong>\uff1a\u5c06\u53c2\u8003\u5e8f\u5217\u548c\u53d8\u4f53\u5e8f\u5217\u5b58\u5165\u5217\u8868\u3002<\/li>\n<li><strong>\u8fd0\u884c\u9884\u6d4b<\/strong>\uff1a\n<pre><code>ref_seqs = [\"ATCG...\"]  # \u53c2\u8003\u5e8f\u5217  \r\nvar_seqs = [\"ATGG...\"]  # \u53d8\u4f53\u5e8f\u5217  \r\nref_scores = model.score_sequences(ref_seqs)  \r\nvar_scores = model.score_sequences(var_seqs)  \r\nprint(f\"Reference likelihood: {ref_scores}, Variant likelihood: {var_scores}\")\r\n<\/code><\/pre>\n<\/li>\n<li><strong>\u5206\u6790\u7ed3\u679c<\/strong>\uff1a\u6bd4\u8f83\u5f97\u5206\u5dee\u5f02\uff0c\u8bc4\u4f30\u53d8\u4f53\u5bf9\u529f\u80fd\u7684\u6f5c\u5728\u5f71\u54cd\u3002<\/li>\n<\/ol>\n<h4>\u529f\u80fd 3\uff1a\u957f\u5e8f\u5217\u5904\u7406<\/h4>\n<p>\u5bf9\u4e8e\u8d85\u957f\u5e8f\u5217\uff0cEvo 2 \u652f\u6301\u5206\u5757\u52a0\u8f7d\u548c\u8ba1\u7b97\uff1a<\/p>\n<ol>\n<li><strong>\u52a0\u8f7d\u5927\u6a21\u578b<\/strong>\uff1a\n<pre><code>model = Evo2('evo2_40b')  # \u9700\u8981\u591a GPU \u652f\u6301\r\n<\/code><\/pre>\n<\/li>\n<li><strong>\u5904\u7406\u957f\u5e8f\u5217<\/strong>\uff1a\n<pre><code>long_seq = \"ATCG...\" * 100000  # \u6a21\u62df 100 \u4e07\u78b1\u57fa\u5e8f\u5217  \r\noutput = model.generate([long_seq], n_tokens=1000)  \r\nprint(output.sequences[0])\r\n<\/code><\/pre>\n<\/li>\n<li><strong>\u6ce8\u610f\u4e8b\u9879<\/strong>\uff1a\u76ee\u524d\u957f\u5e8f\u5217\u524d\u5411\u4f20\u64ad\u53ef\u80fd\u8f83\u6162\uff0c\u5efa\u8bae\u4f18\u5316\u786c\u4ef6\u914d\u7f6e\u6216\u4f7f\u7528\u6559\u5e08\u63d0\u793a\u6cd5\uff08teacher prompting\uff09\u3002<\/li>\n<\/ol>\n<h4>\u529f\u80fd 4\uff1a\u6570\u636e\u96c6\u4e0e\u4e8c\u6b21\u5f00\u53d1<\/h4>\n<ul>\n<li><strong>\u83b7\u53d6\u6570\u636e\u96c6<\/strong>\uff1a\u4ece Hugging Face \u4e0b\u8f7d OpenGenome2 \u6570\u636e\u96c6\uff08FASTA \u6216 JSONL \u683c\u5f0f\uff09\u3002<\/li>\n<li><strong>\u81ea\u5b9a\u4e49\u8bad\u7ec3<\/strong>\uff1a\u57fa\u4e8e Savanna \u6846\u67b6\u4fee\u6539\u6a21\u578b\u67b6\u6784\u6216\u5fae\u8c03\u53c2\u6570\uff0c\u9002\u7528\u4e8e\u7279\u5b9a\u7814\u7a76\u9700\u6c42\u3002<\/li>\n<\/ul>\n<h3>\u64cd\u4f5c\u6280\u5de7\u4e0e\u6ce8\u610f\u4e8b\u9879<\/h3>\n<ul>\n<li><strong>\u591a GPU \u914d\u7f6e<\/strong>\uff1a\u82e5\u4f7f\u7528 40B \u6a21\u578b\uff0c\u9700\u786e\u4fdd Vortex \u6b63\u786e\u8bc6\u522b\u591a\u5757 GPU\uff0c\u53ef\u901a\u8fc7\u00a0<code>nvidia-smi<\/code>\u00a0\u68c0\u67e5\u8d44\u6e90\u5206\u914d\u3002<\/li>\n<li><strong>\u6027\u80fd\u4f18\u5316<\/strong>\uff1a\u957f\u5e8f\u5217\u5904\u7406\u65f6\uff0c\u53ef\u964d\u4f4e\u00a0<code>temperature<\/code>\u00a0\u503c\u4ee5\u51cf\u5c11\u8ba1\u7b97\u8d1f\u62c5\u3002<\/li>\n<li><strong>\u793e\u533a\u652f\u6301<\/strong>\uff1a\u9047\u5230\u95ee\u9898\u53ef\u524d\u5f80 GitHub Issues \u677f\u5757\u63d0\u95ee\uff0cArc Institute \u56e2\u961f\u548c\u793e\u533a\u4f1a\u63d0\u4f9b\u5e2e\u52a9\u3002<\/li>\n<\/ul>\n<p>\u901a\u8fc7\u4ee5\u4e0a\u6b65\u9aa4\uff0c\u7528\u6237\u5373\u53ef\u5feb\u901f\u4e0a\u624b Evo 2\uff0c\u65e0\u8bba\u662f\u751f\u6210 DNA \u5e8f\u5217\u8fd8\u662f\u8fdb\u884c\u57fa\u56e0\u53d8\u4f53\u5206\u6790\uff0c\u90fd\u80fd\u9ad8\u6548\u5b8c\u6210\u4efb\u52a1\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"<p>Arc Institute Evo 2 \u662f\u4e00\u4e2a\u4e13\u6ce8\u4e8e\u57fa\u56e0\u7ec4\u5efa\u6a21\u4e0e\u8bbe\u8ba1\u7684\u5f00\u6e90\u9879\u76ee\uff0c\u7531\u4f4d\u4e8e\u7f8e\u56fd\u52a0\u5dde\u5e15\u6d1b\u963f\u5c14\u6258\u7684\u975e\u8425\u5229\u7814\u7a76\u673a\u6784 Arc Institute \u5f00\u53d1\uff0c\u5e76\u4e0e NVIDIA \u7b49\u5408\u4f5c\u4f19\u4f34\u534f\u4f5c\u63a8\u51fa\u3002\u8be5\u9879\u76ee\u901a\u8fc7\u524d\u6cbf\u6df1\u5ea6\u5b66\u4e60\u6280\u672f\uff0c\u6784\u5efa\u4e86\u80fd\u591f\u5904\u7406 &#8230;<\/p>\n","protected":false},"author":1,"featured_media":61896,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[20],"tags":[230],"class_list":["post-26466","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-tool","tag-aikaiyuanxiangmu"],"_links":{"self":[{"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/posts\/26466","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/comments?post=26466"}],"version-history":[{"count":0,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/posts\/26466\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/media\/61896"}],"wp:attachment":[{"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/media?parent=26466"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/categories?post=26466"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/tags?post=26466"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}