{"id":29653,"date":"2025-03-31T00:54:29","date_gmt":"2025-03-30T16:54:29","guid":{"rendered":"https:\/\/www.aisharenet.com\/?p=29653"},"modified":"2025-03-31T00:54:29","modified_gmt":"2025-03-30T16:54:29","slug":"omnisql","status":"publish","type":"post","link":"https:\/\/www.kdjingpai.com\/ja\/omnisql\/","title":{"rendered":"OmniSQL\uff1a\u5c06\u81ea\u7136\u8bed\u8a00\u8f6c\u5316\u4e3a\u9ad8\u8d28\u91cfSQL\u67e5\u8be2\u7684\u6a21\u578b"},"content":{"rendered":"<p>OmniSQL \u662f\u4e00\u4e2a\u5f00\u6e90\u9879\u76ee\uff0c\u7531 RUCKBReasoning \u56e2\u961f\u5f00\u53d1\uff0c\u6258\u7ba1\u5728 GitHub \u4e0a\u3002\u5b83\u7684\u6838\u5fc3\u529f\u80fd\u662f\u5c06\u7528\u6237\u8f93\u5165\u7684\u81ea\u7136\u8bed\u8a00\u95ee\u9898\u8f6c\u5316\u4e3a\u9ad8\u8d28\u91cf\u7684 SQL \u67e5\u8be2\u8bed\u53e5\uff0c\u5e2e\u52a9\u7528\u6237\u8f7b\u677e\u4e0e\u6570\u636e\u5e93\u4ea4\u4e92\u3002\u9879\u76ee\u57fa\u4e8e\u4e00\u4e2a\u81ea\u52a8\u5316\u7684\u6587\u672c\u8f6c SQL \u6570\u636e\u751f\u6210\u6846\u67b6\uff0c\u63a8\u51fa\u4e86\u5305\u542b 250 \u4e07\u6761\u6837\u672c\u7684 SynSQL-2.5M \u6570\u636e\u96c6\uff0c\u8fd9\u662f\u76ee\u524d\u6700\u5927\u7684\u8de8\u9886\u57df\u5408\u6210\u6587\u672c\u8f6c SQL \u6570\u636e\u96c6\u3002OmniSQL \u63d0\u4f9b 7B\u300114B\u300132B \u4e09\u79cd\u6a21\u578b\u89c4\u6a21\uff0c\u9002\u5408\u4e0d\u540c\u9700\u6c42\u7684\u7528\u6237\u3002\u65e0\u8bba\u662f\u6570\u636e\u5206\u6790\u3001\u6570\u636e\u5e93\u7ba1\u7406\uff0c\u8fd8\u662f\u6a21\u578b\u7814\u7a76\uff0c\u5b83\u90fd\u80fd\u63d0\u4f9b\u5f3a\u5927\u652f\u6301\u3002\u9879\u76ee\u4f7f\u7528 Apache 2.0 \u534f\u8bae\uff0c\u7528\u6237\u53ef\u4ee5\u514d\u8d39\u4e0b\u8f7d\u5e76\u53c2\u4e0e\u6539\u8fdb\u3002<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" class=\"aligncenter size-full wp-image-29654\" title=\"OmniSQL\uff1a\u5c06\u81ea\u7136\u8bed\u8a00\u8f6c\u5316\u4e3a\u9ad8\u8d28\u91cfSQL\u67e5\u8be2\u7684\u6a21\u578b-1\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/03\/2d910996893e6e4.jpg\" alt=\"OmniSQL\uff1a\u5c06\u81ea\u7136\u8bed\u8a00\u8f6c\u5316\u4e3a\u9ad8\u8d28\u91cfSQL\u67e5\u8be2\u7684\u6a21\u578b-1\" width=\"3076\" height=\"1020\" srcset=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/03\/2d910996893e6e4.jpg 3076w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/03\/2d910996893e6e4-768x255.jpg 768w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/03\/2d910996893e6e4-1536x509.jpg 1536w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/03\/2d910996893e6e4-2048x679.jpg 2048w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/03\/2d910996893e6e4-18x6.jpg 18w\" sizes=\"auto, (max-width: 3076px) 100vw, 3076px\" \/><\/p>\n<p>&nbsp;<\/p>\n<h2>\u529f\u80fd\u5217\u8868<\/h2>\n<ul>\n<li>\u5c06\u81ea\u7136\u8bed\u8a00\u8f6c\u4e3a SQL\uff1a\u7528\u6237\u8f93\u5165\u95ee\u9898\uff0c\u6a21\u578b\u751f\u6210\u51c6\u786e\u7684 SQL \u67e5\u8be2\u3002<\/li>\n<li>\u652f\u6301\u590d\u6742\u67e5\u8be2\uff1a\u751f\u6210\u4ece\u7b80\u5355\u5355\u8868\u67e5\u8be2\u5230\u591a\u8868\u8fde\u63a5\u7684\u9ad8\u7ea7 SQL\u3002<\/li>\n<li>\u6570\u636e\u96c6\u751f\u6210\uff1a\u63d0\u4f9b SynSQL-2.5M\uff0c\u5305\u542b 250 \u4e07\u6761\u9ad8\u8d28\u91cf\u6837\u672c\u3002<\/li>\n<li>\u591a\u89c4\u6a21\u6a21\u578b\uff1a\u63d0\u4f9b 7B\u300114B\u300132B \u4e09\u79cd\u53c2\u6570\u89c4\u6a21\u7684\u6a21\u578b\u3002<\/li>\n<li>\u5f00\u6e90\u514d\u8d39\uff1a\u4ee3\u7801\u548c\u6570\u636e\u96c6\u53ef\u5728 GitHub \u4e0a\u514d\u8d39\u83b7\u53d6\u3002<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<h2>\u4f7f\u7528\u5e2e\u52a9<\/h2>\n<p>OmniSQL \u662f\u4e00\u4e2a\u57fa\u4e8e\u4ee3\u7801\u7684\u5de5\u5177\uff0c\u9002\u5408\u6709\u4e00\u5b9a\u7f16\u7a0b\u57fa\u7840\u7684\u7528\u6237\u3002\u4ee5\u4e0b\u662f\u8be6\u7ec6\u7684\u5b89\u88c5\u548c\u4f7f\u7528\u6307\u5357\uff0c\u5e2e\u52a9\u4f60\u5feb\u901f\u4e0a\u624b\u3002<\/p>\n<h3>\u5b89\u88c5\u6d41\u7a0b<\/h3>\n<ol>\n<li><strong>\u51c6\u5907\u73af\u5883<\/strong><br \/>\n\u786e\u4fdd\u7535\u8111\u5df2\u5b89\u88c5 Python 3.8 \u6216\u66f4\u9ad8\u7248\u672c\u3002\u6253\u5f00\u547d\u4ee4\u884c\uff0c\u8f93\u5165\u00a0<code>python --version<\/code>\u00a0\u68c0\u67e5\u3002\u5982\u679c\u6ca1\u6709\u5b89\u88c5\uff0c\u53ef\u4ece Python \u5b98\u7f51\u4e0b\u8f7d\u3002<\/li>\n<li><strong>\u4e0b\u8f7d\u9879\u76ee<\/strong><br \/>\n\u8bbf\u95ee\u00a0<code>https:\/\/github.com\/RUCKBReasoning\/OmniSQL<\/code>\uff0c\u70b9\u51fb &#8220;Code&#8221; \u6309\u94ae\uff0c\u9009\u62e9 &#8220;Download ZIP&#8221; \u4e0b\u8f7d\u9879\u76ee\u538b\u7f29\u5305\u3002\u89e3\u538b\u540e\u5f97\u5230\u9879\u76ee\u6587\u4ef6\u5939\u3002\u6216\u8005\u7528 Git \u547d\u4ee4\u514b\u9686\uff1a<\/li>\n<\/ol>\n<pre><code>git clone https:\/\/github.com\/RUCKBReasoning\/OmniSQL.git\r\n<\/code><\/pre>\n<ol start=\"3\">\n<li><strong>\u5b89\u88c5\u4f9d\u8d56<\/strong><br \/>\n\u8fdb\u5165\u9879\u76ee\u76ee\u5f55\uff0c\u5728\u547d\u4ee4\u884c\u8fd0\u884c\uff1a<\/li>\n<\/ol>\n<pre><code>pip install -r requirements.txt\r\n<\/code><\/pre>\n<p>\u8fd9\u4f1a\u5b89\u88c5\u8fd0\u884c\u6240\u9700\u7684 Python \u5e93\u3002\u5982\u679c\u9700\u8981\u6a21\u578b\u63a8\u7406\uff0c\u8fd8\u8981\u5b89\u88c5 <a href=\"https:\/\/www.kdjingpai.com\/de\/vllm\/\">vLLM<\/a> \u6216 Transformers\uff0c\u547d\u4ee4\u5982\u4e0b\uff1a<\/p>\n<pre><code>pip install vllm\r\n<\/code><\/pre>\n<p>\u6216<\/p>\n<pre><code>pip install transformers torch\r\n<\/code><\/pre>\n<ol start=\"4\">\n<li><strong>\u4e0b\u8f7d\u6a21\u578b\u548c\u6570\u636e\u96c6<\/strong><br \/>\nOmniSQL \u63d0\u4f9b\u4e09\u79cd\u6a21\u578b\u548c SynSQL-2.5M \u6570\u636e\u96c6\uff0c\u53ef\u4ece\u4ee5\u4e0b\u94fe\u63a5\u4e0b\u8f7d\uff1a<\/li>\n<\/ol>\n<ul>\n<li>SynSQL-2.5M:\u00a0<a href=\"https:\/\/huggingface.co\/datasets\/seeklhy\/SynSQL-2.5M\">HuggingFace<\/a><\/li>\n<li>OmniSQL-7B:\u00a0<a href=\"https:\/\/huggingface.co\/seeklhy\/OmniSQL-7B\">HuggingFace<\/a><\/li>\n<li>OmniSQL-14B:\u00a0<a href=\"https:\/\/huggingface.co\/seeklhy\/OmniSQL-14B\">HuggingFace<\/a><\/li>\n<li>OmniSQL-32B:\u00a0<a href=\"https:\/\/huggingface.co\/seeklhy\/OmniSQL-32B\">HuggingFace<\/a><br \/>\n\u4e0b\u8f7d\u540e\uff0c\u5c06\u6587\u4ef6\u653e\u5165\u9879\u76ee\u76ee\u5f55\u3002<\/li>\n<\/ul>\n<ol start=\"5\">\n<li><strong>\u8fd0\u884c\u9879\u76ee<\/strong><br \/>\n\u8fdb\u5165\u9879\u76ee\u76ee\u5f55\uff0c\u8fd0\u884c\u00a0<code>python omnisql.py<\/code>\u00a0\u68c0\u67e5\u73af\u5883\u662f\u5426\u6b63\u5e38\u3002\u5b9e\u9645\u4f7f\u7528\u65f6\u9700\u52a0\u8f7d\u6a21\u578b\uff0c\u5177\u4f53\u89c1\u4e0b\u6587\u3002<\/li>\n<\/ol>\n<h3>\u4e3b\u8981\u529f\u80fd\u64cd\u4f5c<\/h3>\n<h4>1. \u5c06\u81ea\u7136\u8bed\u8a00\u8f6c\u4e3a SQL<\/h4>\n<p>OmniSQL \u7684\u6838\u5fc3\u529f\u80fd\u662f\u5c06\u95ee\u9898\u8f6c\u5316\u4e3a SQL \u67e5\u8be2\u3002\u4ee5 vLLM \u4e3a\u4f8b\uff0c\u8fd0\u884c\u4ee5\u4e0b\u4ee3\u7801\uff1a<\/p>\n<pre><code>from vllm import LLM, SamplingParams\r\nfrom transformers import AutoTokenizer\r\n# \u5b9a\u4e49\u63d0\u793a\u6a21\u677f\r\nprompt = '''Task Overview:\r\nYou are a data science expert. Below, you are provided with a database schema and a natural language question. Your task is to generate a valid SQL query.\r\nDatabase Engine: SQLite\r\nDatabase Schema:\r\nCREATE TABLE users (id INTEGER PRIMARY KEY, name TEXT, age INTEGER);\r\nQuestion:\r\n\u67e5\u627e users \u8868\u4e2d\u5e74\u9f84\u5927\u4e8e 30 \u7684\u4eba\u7684\u540d\u5b57\r\nInstructions:\r\n- \u53ea\u8f93\u51fa\u95ee\u9898\u8981\u6c42\u7684\u4fe1\u606f\u3002\r\n- \u9010\u6b65\u601d\u8003\u540e\u751f\u6210 SQL\u3002\r\nOutput Format:\r\n<\/code><\/pre>\n<p>&#8212; Your SQL query<\/p>\n<pre><code>'''\r\n# \u52a0\u8f7d\u6a21\u578b\r\nmodel_path = \"seeklhy\/OmniSQL-7B\"  # \u66ff\u6362\u4e3a\u4f60\u7684\u6a21\u578b\u8def\u5f84\r\ntokenizer = AutoTokenizer.from_pretrained(model_path)\r\nllm = LLM(model=model_path, dtype=\"float16\")\r\n# \u751f\u6210 SQL\r\nsampling_params = SamplingParams(temperature=0, max_tokens=2048)\r\nchat_prompt = tokenizer.apply_chat_template([{\"role\": \"user\", \"content\": prompt}], add_generation_prompt=True, tokenize=False)\r\noutputs = llm.generate([chat_prompt], sampling_params)\r\nprint(outputs[0].outputs[0].text)\r\n<\/code><\/pre>\n<p>\u8f93\u51fa\u53ef\u80fd\u662f\uff1a<\/p>\n<pre><code>SELECT name FROM users WHERE age &gt; 30;\r\n<\/code><\/pre>\n<h4>2. \u4f7f\u7528 SynSQL-2.5M \u6570\u636e\u96c6<\/h4>\n<p>\u6570\u636e\u96c6\u5305\u542b 250 \u4e07\u6761\u6837\u672c\uff0c\u6bcf\u6761\u5305\u62ec\u6570\u636e\u5e93\u7ed3\u6784\u3001\u95ee\u9898\u3001SQL \u67e5\u8be2\u548c\u601d\u8003\u8fc7\u7a0b\u3002\u4e0b\u8f7d\u540e\uff0c\u53ef\u76f4\u63a5\u7528\u4e8e\u8bad\u7ec3\u6216\u7814\u7a76\u3002\u67e5\u770b\u6837\u672c\uff1a<\/p>\n<ul>\n<li>\u89e3\u538b\u6570\u636e\u96c6\u6587\u4ef6\u3002<\/li>\n<li>\u6253\u5f00\u4efb\u610f JSON \u6587\u4ef6\uff0c\u683c\u5f0f\u5982\u00a0<code>{\"db\": ..., \"question\": ..., \"sql\": ..., \"cot\": ...}<\/code>\u3002<\/li>\n<\/ul>\n<h4>3. \u8bad\u7ec3\u548c\u8bc4\u4f30<\/h4>\n<p>\u9879\u76ee\u63d0\u4f9b\u8bad\u7ec3\u811a\u672c\uff0c\u4f4d\u4e8e\u00a0<code>train_and_evaluate<\/code>\u00a0\u6587\u4ef6\u5939\u3002\u8fd0\u884c\u793a\u4f8b\uff1a<\/p>\n<pre><code>python train.py --model OmniSQL-7B --data SynSQL-2.5M\r\n<\/code><\/pre>\n<p>\u8bc4\u4f30\u811a\u672c\u4e5f\u5728\u540c\u4e00\u6587\u4ef6\u5939\uff0c\u53ef\u590d\u73b0\u5b98\u65b9\u7ed3\u679c\u3002<\/p>\n<h3>\u4f7f\u7528\u6280\u5de7<\/h3>\n<ul>\n<li><strong>\u6570\u636e\u5e93\u652f\u6301<\/strong>\uff1a\u76ee\u524d\u4ec5\u652f\u6301 SQLite\u3002\u82e5\u9700\u5176\u4ed6\u6570\u636e\u5e93\uff0c\u53ef\u7528\u6570\u636e\u751f\u6210\u6846\u67b6\u5408\u6210\u65b0\u6570\u636e\u3002<\/li>\n<li><strong>\u786c\u4ef6\u8981\u6c42<\/strong>\uff1a7B \u6a21\u578b\u9700\u7ea6 14GB \u663e\u5b58\uff0c32B \u9700\u66f4\u9ad8\u914d\u7f6e\u3002<\/li>\n<li><strong>\u67e5\u770b\u793a\u4f8b<\/strong>\uff1a\u9879\u76ee\u00a0<code>examples<\/code>\u00a0\u6587\u4ef6\u5939\u63d0\u4f9b\u63d0\u793a\u6a21\u677f\u793a\u4f8b\u3002<\/li>\n<\/ul>\n<p>\u901a\u8fc7\u8fd9\u4e9b\u6b65\u9aa4\uff0c\u4f60\u53ef\u4ee5\u7528 OmniSQL \u5feb\u901f\u751f\u6210 SQL \u6216\u7814\u7a76\u6587\u672c\u8f6c SQL \u6280\u672f\u3002<\/p>\n<p>&nbsp;<\/p>\n<h2>\u5e94\u7528\u573a\u666f<\/h2>\n<ol>\n<li><strong>\u6570\u636e\u5206\u6790<\/strong><br \/>\n\u6570\u636e\u5206\u6790\u5e08\u8f93\u5165\u95ee\u9898\uff0c\u5982\u201c\u67e5\u627e\u9500\u552e\u989d\u524d\u5341\u7684\u5546\u54c1\u201d\uff0cOmniSQL \u751f\u6210\u5bf9\u5e94 SQL\uff0c\u8282\u7701\u65f6\u95f4\u3002<\/li>\n<li><strong>\u6a21\u578b\u7814\u7a76<\/strong><br \/>\n\u7814\u7a76\u8005\u7528 SynSQL-2.5M \u8bad\u7ec3\u65b0\u6a21\u578b\uff0c\u63d0\u5347\u6587\u672c\u8f6c SQL \u80fd\u529b\u3002<\/li>\n<li><strong>\u6559\u80b2\u5b66\u4e60<\/strong><br \/>\n\u5b66\u751f\u901a\u8fc7\u8f93\u5165\u95ee\u9898\u89c2\u5bdf\u751f\u6210\u7684 SQL\uff0c\u5b66\u4e60\u6570\u636e\u5e93\u64cd\u4f5c\u3002<\/li>\n<\/ol>\n<p>&nbsp;<\/p>\n<h2>QA<\/h2>\n<ol>\n<li><strong>OmniSQL \u652f\u6301\u54ea\u4e9b\u6570\u636e\u5e93\uff1f<\/strong><br \/>\n\u5f53\u524d\u4ec5\u652f\u6301 SQLite\uff0c\u672a\u6765\u53ef\u901a\u8fc7\u5408\u6210\u6570\u636e\u6269\u5c55\u3002<\/li>\n<li><strong>\u6570\u636e\u96c6\u6709\u591a\u5927\uff1f<\/strong><br \/>\nSynSQL-2.5M \u5305\u542b 250 \u4e07\u6761\u6837\u672c\uff0c\u8986\u76d6 1.6 \u4e07\u4e2a\u6570\u636e\u5e93\u3002<\/li>\n<li><strong>\u6a21\u578b\u6709\u591a\u5f3a\uff1f<\/strong><br \/>\n\u5728 Spider\u3001BIRD \u7b49\u57fa\u51c6\u6d4b\u8bd5\u4e2d\uff0cOmniSQL \u8d85\u8d8a\u4e86 GPT-4o \u7b49\u6a21\u578b\u3002<\/li>\n<\/ol>\n","protected":false},"excerpt":{"rendered":"<p>OmniSQL \u662f\u4e00\u4e2a\u5f00\u6e90\u9879\u76ee\uff0c\u7531 RUCKBReasoning \u56e2\u961f\u5f00\u53d1\uff0c\u6258\u7ba1\u5728 GitHub \u4e0a\u3002\u5b83\u7684\u6838\u5fc3\u529f\u80fd\u662f\u5c06\u7528\u6237\u8f93\u5165\u7684\u81ea\u7136\u8bed\u8a00\u95ee\u9898\u8f6c\u5316\u4e3a\u9ad8\u8d28\u91cf\u7684 SQL \u67e5\u8be2\u8bed\u53e5\uff0c\u5e2e\u52a9\u7528\u6237\u8f7b\u677e\u4e0e\u6570\u636e\u5e93\u4ea4\u4e92\u3002\u9879\u76ee\u57fa\u4e8e\u4e00\u4e2a\u81ea\u52a8\u5316\u7684\u6587\u672c\u8f6c SQL \u6570\u636e&#8230;<\/p>\n","protected":false},"author":1,"featured_media":62169,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[20],"tags":[230,244],"class_list":["post-29653","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-tool","tag-aikaiyuanxiangmu","tag-aishujufenxi"],"_links":{"self":[{"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/posts\/29653","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/comments?post=29653"}],"version-history":[{"count":0,"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/posts\/29653\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/media\/62169"}],"wp:attachment":[{"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/media?parent=29653"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/categories?post=29653"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/ja\/wp-json\/wp\/v2\/tags?post=29653"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}