{"id":17506,"date":"2025-01-07T11:15:17","date_gmt":"2025-01-07T03:15:17","guid":{"rendered":"https:\/\/www.aisharenet.com\/?p=17506"},"modified":"2025-11-18T17:05:43","modified_gmt":"2025-11-18T09:05:43","slug":"vita","status":"publish","type":"post","link":"https:\/\/www.kdjingpai.com\/pt\/vita\/","title":{"rendered":"VITA"},"content":{"rendered":"<p>VITA\u662f\u4e00\u4e2a\u9886\u5148\u7684\u5f00\u6e90\u4ea4\u4e92\u5f0f\u591a\u6a21\u6001\u5927\u8bed\u8a00\u6a21\u578b\u9879\u76ee\uff0c\u7387\u5148\u5b9e\u73b0\u4e86\u771f\u6b63\u7684\u5168\u65b9\u4f4d\u591a\u6a21\u6001\u4ea4\u4e92\u80fd\u529b\u3002\u8be5\u9879\u76ee\u4e8e2024\u5e748\u6708\u63a8\u51faVITA-1.0\u7248\u672c\uff0c\u5f00\u521b\u4e86\u9996\u4e2a\u5f00\u6e90\u4ea4\u4e92\u5f0f\u5168\u6a21\u6001\u5927\u8bed\u8a00\u6a21\u578b\u7684\u5148\u6cb3\u30022024\u5e7412\u6708\uff0c\u9879\u76ee\u63a8\u51fa\u4e86\u91cd\u5927\u5347\u7ea7\u7248\u672cVITA-1.5\uff0c\u5728\u4ea4\u4e92\u4f53\u9a8c\u548c\u6027\u80fd\u65b9\u9762\u90fd\u6709\u663e\u8457\u63d0\u5347\u3002VITA\u6a21\u578b\u652f\u6301\u56fe\u50cf\u3001\u89c6\u9891\u3001\u97f3\u9891\u7b49\u591a\u79cd\u6a21\u6001\u7684\u8f93\u5165\u548c\u8f93\u51fa\uff0c\u5177\u5907\u5b9e\u65f6\u4ea4\u4e92\u80fd\u529b\uff0c\u5c06\u7aef\u5230\u7aef\u7684\u8bed\u97f3\u4ea4\u4e92\u5ef6\u8fdf\u4ece4\u79d2\u663e\u8457\u964d\u4f4e\u52301.5\u79d2\uff0c\u6781\u5927\u63d0\u5347\u4e86\u7528\u6237\u4f53\u9a8c\u3002\u4f5c\u4e3a\u4e00\u4e2a\u5b8c\u5168\u5f00\u6e90\u7684\u9879\u76ee\uff0cVITA\u4e3a\u7814\u7a76\u4eba\u5458\u548c\u5f00\u53d1\u8005\u63d0\u4f9b\u4e86\u63a2\u7d22\u591a\u6a21\u6001AI\u7684\u7406\u60f3\u5e73\u53f0\u3002<\/p>\n<p><img decoding=\"async\" class=\"aligncenter\" title=\"VITA\uff1a\u5f00\u6e90\u89c6\u89c9\u4e0e\u8bed\u97f3\u5b9e\u65f6\u4ea4\u4e92\u7684\u591a\u6a21\u6001\u5927\u8bed\u8a00\u6a21\u578b-1\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/01\/714318e3f51d87c.png\" alt=\"VITA\uff1a\u5f00\u6e90\u89c6\u89c9\u4e0e\u8bed\u97f3\u5b9e\u65f6\u4ea4\u4e92\u7684\u591a\u6a21\u6001\u5927\u8bed\u8a00\u6a21\u578b-1\" \/><\/p>\n<p>&nbsp;<\/p>\n<h2>\u529f\u80fd\u5217\u8868<\/h2>\n<ul>\n<li>\u591a\u6a21\u6001\u8f93\u5165\u5904\u7406\uff1a\u652f\u6301\u56fe\u50cf\u3001\u89c6\u9891\u3001\u97f3\u9891\u7b49\u591a\u79cd\u5f62\u5f0f\u7684\u8f93\u5165<\/li>\n<li>\u5b9e\u65f6\u8bed\u97f3\u4ea4\u4e92\uff1a\u7aef\u5230\u7aef\u8bed\u97f3\u4ea4\u4e92\u5ef6\u8fdf\u4ec51.5\u79d2<\/li>\n<li>\u89c6\u89c9\u5206\u6790\u80fd\u529b\uff1a\u5f3a\u5927\u7684\u56fe\u50cf\u548c\u89c6\u9891\u7406\u89e3\u4e0e\u5206\u6790\u529f\u80fd<\/li>\n<li>\u97f3\u9891\u5904\u7406\uff1a\u652f\u6301\u8bed\u97f3\u8bc6\u522b\u548c\u8bed\u97f3\u5408\u6210<\/li>\n<li>\u8de8\u6a21\u6001\u7406\u89e3\uff1a\u5b9e\u73b0\u6587\u672c\u3001\u56fe\u50cf\u3001\u97f3\u9891\u95f4\u7684\u667a\u80fd\u5173\u8054<\/li>\n<li>\u5f00\u6e90\u4ee3\u7801\u652f\u6301\uff1a\u5b8c\u6574\u7684\u8bad\u7ec3\u548c\u63a8\u7406\u4ee3\u7801\u5f00\u653e<\/li>\n<li>\u9884\u8bad\u7ec3\u6a21\u578b\uff1a\u63d0\u4f9b\u591a\u4e2a\u9884\u8bad\u7ec3\u6a21\u578b\u7248\u672c<\/li>\n<li>\u7075\u6d3b\u7684\u90e8\u7f72\u9009\u9879\uff1a\u652f\u6301\u591a\u79cd\u786c\u4ef6\u5e73\u53f0\u90e8\u7f72<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<h2>VITA-1.5 \u6982\u89c8<\/h2>\n<p>\u5728 2024 \u5e74 8 \u6708 12 \u65e5\uff0c\u6211\u4eec\u53d1\u5e03\u4e86\u00a0<strong>VITA-1.0<\/strong>\uff0c\u8fd9\u662f\u00a0<strong>\u9996\u4e2a\u5f00\u6e90\u7684\u4ea4\u4e92\u5f0f\u5168\u80fd\u591a\u6a21\u6001\u5927\u8bed\u8a00\u6a21\u578b<\/strong>\u3002\u73b0\u5728\uff082024 \u5e74 12 \u6708 20 \u65e5\uff09\uff0c\u6211\u4eec\u5e26\u6765\u4e86\u00a0<strong>\u65b0\u7248\u672c VITA-1.5<\/strong>\uff01<\/p>\n<h3>VITA-1.5 \u6709\u54ea\u4e9b\u65b0\u529f\u80fd\uff1f<\/h3>\n<p>\u6211\u4eec\u5f88\u9ad8\u5174\u5730\u4ecb\u7ecd\u00a0<strong>VITA-1.5<\/strong>\uff0c\u5b83\u5f15\u5165\u4e86\u4e00\u7cfb\u5217\u8fdb\u6b65\uff1a<\/p>\n<ol>\n<li><strong>\u663e\u8457\u51cf\u5c11\u4ea4\u4e92\u5ef6\u8fdf<\/strong>\u3002\u7aef\u5230\u7aef\u8bed\u97f3\u4ea4\u4e92\u5ef6\u8fdf\u5df2\u4ece\u00a0<strong>\u7ea6 4 \u79d2<\/strong>\u00a0\u51cf\u5c11\u5230\u00a0<strong>1.5 \u79d2<\/strong>\uff0c\u5b9e\u73b0\u4e86\u51e0\u4e4e\u5373\u65f6\u7684\u4ea4\u4e92\uff0c\u5927\u5927\u63d0\u5347\u4e86\u7528\u6237\u4f53\u9a8c\u3002<\/li>\n<li><strong>\u589e\u5f3a\u7684\u591a\u6a21\u6001\u6027\u80fd<\/strong>\u3002\u5728\u00a0<em>MME<\/em>\u3001<em>MMBench<\/em>\u00a0\u548c\u00a0<em>MathVista<\/em>\u00a0\u7b49\u591a\u6a21\u6001\u57fa\u51c6\u6d4b\u8bd5\u4e2d\u7684\u5e73\u5747\u6027\u80fd\u663e\u8457\u63d0\u5347\uff0c\u4ece\u00a0<strong>59.8<\/strong>\u00a0\u63d0\u9ad8\u5230\u00a0<strong>70.8<\/strong>\u3002<\/li>\n<li><strong>\u8bed\u97f3\u5904\u7406\u80fd\u529b\u7684\u63d0\u5347<\/strong>\u3002\u8bed\u97f3\u5904\u7406\u80fd\u529b\u8fbe\u5230\u65b0\u6c34\u5e73\uff0cASR WER\uff08\u5355\u8bcd\u9519\u8bef\u7387\uff0cTest Other\uff09\u4ece\u00a0<strong>18.4<\/strong>\u00a0\u964d\u4f4e\u5230\u00a0<strong>7.5<\/strong>\u3002\u6b64\u5916\uff0c\u6211\u4eec\u7528\u00a0<strong>\u7aef\u5230\u7aef TTS \u6a21\u5757<\/strong>\u00a0\u66ff\u6362\u4e86 VITA-1.0 \u7684\u72ec\u7acb TTS \u6a21\u5757\uff0c\u8be5\u6a21\u5757\u63a5\u53d7\u5927\u8bed\u8a00\u6a21\u578b\u7684\u5d4c\u5165\u4f5c\u4e3a\u8f93\u5165\u3002<\/li>\n<li><strong>\u6e10\u8fdb\u5f0f\u8bad\u7ec3\u7b56\u7565<\/strong>\u3002\u901a\u8fc7\u8fd9\u79cd\u65b9\u5f0f\uff0c\u8bed\u97f3\u6a21\u5757\u7684\u52a0\u5165\u5bf9\u5176\u4ed6\u591a\u6a21\u6001\u6027\u80fd\uff08\u89c6\u89c9-\u8bed\u8a00\uff09\u51e0\u4e4e\u6ca1\u6709\u5f71\u54cd\u3002\u56fe\u50cf\u7406\u89e3\u7684\u5e73\u5747\u6027\u80fd\u4ec5\u4ece 71.3 \u964d\u4f4e\u5230 70.8\u3002<\/li>\n<\/ol>\n<h2>\u5b9e\u9a8c\u7ed3\u679c<\/h2>\n<ul>\n<li><strong>\u56fe\u50cf\u548c\u89c6\u9891\u7406\u89e3\u57fa\u51c6\u6d4b\u8bd5\u7684\u8bc4\u4f30<\/strong><\/li>\n<\/ul>\n<p><img loading=\"lazy\" decoding=\"async\" class=\"aligncenter size-full wp-image-17509\" title=\"VITA\uff1a\u5f00\u6e90\u89c6\u89c9\u4e0e\u8bed\u97f3\u5b9e\u65f6\u4ea4\u4e92\u7684\u591a\u6a21\u6001\u5927\u8bed\u8a00\u6a21\u578b-1\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/01\/c450ddab2283bda.png\" alt=\"VITA\uff1a\u5f00\u6e90\u89c6\u89c9\u4e0e\u8bed\u97f3\u5b9e\u65f6\u4ea4\u4e92\u7684\u591a\u6a21\u6001\u5927\u8bed\u8a00\u6a21\u578b-1\" width=\"821\" height=\"467\" srcset=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/01\/c450ddab2283bda.png 821w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/01\/c450ddab2283bda-300x171.png 300w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/01\/c450ddab2283bda-768x437.png 768w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/01\/c450ddab2283bda-18x10.png 18w\" sizes=\"auto, (max-width: 821px) 100vw, 821px\" \/><\/p>\n<ul>\n<li><strong>VITA-1.5 \u5728 ASR \u57fa\u51c6\u6d4b\u8bd5\u4e2d\u8868\u73b0\u4f18\u4e8e\u4e13\u4e1a\u8bed\u97f3\u6a21\u578b<\/strong><\/li>\n<\/ul>\n<p><img loading=\"lazy\" decoding=\"async\" class=\"aligncenter size-full wp-image-17510\" title=\"VITA\uff1a\u5f00\u6e90\u89c6\u89c9\u4e0e\u8bed\u97f3\u5b9e\u65f6\u4ea4\u4e92\u7684\u591a\u6a21\u6001\u5927\u8bed\u8a00\u6a21\u578b-1\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/01\/6a9e73d3c453861.png\" alt=\"VITA\uff1a\u5f00\u6e90\u89c6\u89c9\u4e0e\u8bed\u97f3\u5b9e\u65f6\u4ea4\u4e92\u7684\u591a\u6a21\u6001\u5927\u8bed\u8a00\u6a21\u578b-1\" width=\"797\" height=\"376\" srcset=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/01\/6a9e73d3c453861.png 797w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/01\/6a9e73d3c453861-300x142.png 300w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/01\/6a9e73d3c453861-768x362.png 768w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/01\/6a9e73d3c453861-18x8.png 18w\" sizes=\"auto, (max-width: 797px) 100vw, 797px\" \/><\/p>\n<ul>\n<li><strong>\u97f3\u9891\u6a21\u6001\u7684\u52a0\u5165\u5bf9\u56fe\u50cf\u548c\u89c6\u9891\u7406\u89e3\u80fd\u529b\u51e0\u4e4e\u6ca1\u6709\u5f71\u54cd<\/strong><\/li>\n<\/ul>\n<p><img loading=\"lazy\" decoding=\"async\" class=\"aligncenter size-full wp-image-17511\" title=\"VITA\uff1a\u5f00\u6e90\u89c6\u89c9\u4e0e\u8bed\u97f3\u5b9e\u65f6\u4ea4\u4e92\u7684\u591a\u6a21\u6001\u5927\u8bed\u8a00\u6a21\u578b-1\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/01\/047c547080a1b92.png\" alt=\"VITA\uff1a\u5f00\u6e90\u89c6\u89c9\u4e0e\u8bed\u97f3\u5b9e\u65f6\u4ea4\u4e92\u7684\u591a\u6a21\u6001\u5927\u8bed\u8a00\u6a21\u578b-1\" width=\"930\" height=\"278\" srcset=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/01\/047c547080a1b92.png 930w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/01\/047c547080a1b92-300x90.png 300w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/01\/047c547080a1b92-768x230.png 768w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/01\/047c547080a1b92-18x5.png 18w\" sizes=\"auto, (max-width: 930px) 100vw, 930px\" \/><\/p>\n<p>&nbsp;<\/p>\n<h2>\u4f7f\u7528\u5e2e\u52a9<\/h2>\n<h3>1. \u73af\u5883\u914d\u7f6e\u4e0e\u5b89\u88c5<\/h3>\n<p>1.1 \u57fa\u7840\u8981\u6c42\uff1a<\/p>\n<ul>\n<li>Python\u73af\u5883<\/li>\n<li>PyTorch\u6846\u67b6<\/li>\n<li>CUDA\u652f\u6301\uff08\u63a8\u8350\u4f7f\u7528GPU\u52a0\u901f\uff09<\/li>\n<\/ul>\n<p>1.2 \u5b89\u88c5\u6b65\u9aa4\uff1a<\/p>\n<pre><code># \u514b\u9686\u9879\u76ee\u4ed3\u5e93\r\ngit clone https:\/\/github.com\/VITA-MLLM\/VITA.git\r\ncd VITA\r\n# \u5b89\u88c5\u4f9d\u8d56\r\npip install -r requirements.txt\r\n<\/code><\/pre>\n<h3>2. \u6a21\u578b\u4f7f\u7528<\/h3>\n<p>2.1 \u52a0\u8f7d\u9884\u8bad\u7ec3\u6a21\u578b\uff1a<\/p>\n<pre><code>from vita.model.builder import load_pretrained_model\r\nfrom vita.conversation import conv_templates\r\nfrom vita.util.mm_utils import get_model_name_from_path\r\n# \u52a0\u8f7d\u6a21\u578b\r\nmodel_path = 'VITA\/vita'\r\nmodel_name = get_model_name_from_path(model_path)\r\ntokenizer, model, image_processor, _ = load_pretrained_model(\r\nmodel_path, \r\nNone, \r\nmodel_name, \r\nmodel_type='mixtral-8x7b', \r\ndevice_map='auto'\r\n)\r\n<\/code><\/pre>\n<p>2.2 \u97f3\u9891\u5904\u7406\u914d\u7f6e\uff1a<\/p>\n<pre><code># \u521d\u59cb\u5316\u97f3\u9891\u7f16\u7801\u5668\r\naudio_encoder = model.get_audio_encoder()\r\naudio_encoder.to(dtype=torch.float16)\r\naudio_processor = audio_encoder.audio_processor\r\n<\/code><\/pre>\n<h3>3. \u5b9e\u65f6\u4ea4\u4e92\u529f\u80fd<\/h3>\n<ul>\n<li>\u652f\u6301\u5b9e\u65f6\u8bed\u97f3\u8f93\u5165\u548c\u54cd\u5e94<\/li>\n<li>\u96c6\u6210\u56fe\u50cf\u8bc6\u522b\u548c\u5206\u6790<\/li>\n<li>\u652f\u6301\u591a\u8f6e\u5bf9\u8bdd\u4ea4\u4e92<\/li>\n<li>\u63d0\u4f9b\u5b8c\u6574\u7684\u5bf9\u8bdd\u6a21\u677f\u7cfb\u7edf<\/li>\n<\/ul>\n<h3>4. \u9ad8\u7ea7\u529f\u80fd\u4f7f\u7528<\/h3>\n<p>4.1 \u591a\u6a21\u6001\u8f93\u5165\u5904\u7406\uff1a<\/p>\n<ul>\n<li>\u652f\u6301\u6279\u91cf\u56fe\u50cf\u5904\u7406<\/li>\n<li>\u89c6\u9891\u6d41\u5b9e\u65f6\u5206\u6790<\/li>\n<li>\u97f3\u9891\u6d41\u5904\u7406\u4e0e\u5408\u6210<\/li>\n<\/ul>\n<p>4.2 \u6a21\u578b\u8bad\u7ec3\u4e0e\u5fae\u8c03\uff1a<\/p>\n<ul>\n<li>\u63d0\u4f9b\u5b8c\u6574\u7684\u8bad\u7ec3\u811a\u672c<\/li>\n<li>\u652f\u6301\u6301\u7eed\u5b66\u4e60\u529f\u80fd<\/li>\n<li>\u81ea\u5b9a\u4e49\u6570\u636e\u96c6\u8bad\u7ec3\u652f\u6301<\/li>\n<\/ul>\n<h3>5. \u8bc4\u4f30\u4e0e\u6d4b\u8bd5<\/h3>\n<ul>\n<li>\u652f\u6301\u4e3b\u6d41\u591a\u6a21\u6001\u8bc4\u6d4b\u57fa\u51c6<\/li>\n<li>\u96c6\u6210VLMEvalKit\u8bc4\u4f30\u5de5\u5177<\/li>\n<li>\u63d0\u4f9b\u8be6\u7ec6\u7684\u6027\u80fd\u6d4b\u8bd5\u6307\u6807<\/li>\n<\/ul>\n<h3>6. \u6ce8\u610f\u4e8b\u9879<\/h3>\n<ul>\n<li>\u5efa\u8bae\u4f7f\u7528GPU\u8fdb\u884c\u6a21\u578b\u63a8\u7406<\/li>\n<li>\u6ce8\u610f\u5185\u5b58\u7ba1\u7406\uff0c\u7279\u522b\u662f\u5904\u7406\u5927\u578b\u591a\u6a21\u6001\u8f93\u5165\u65f6<\/li>\n<li>\u5b9a\u671f\u68c0\u67e5\u9879\u76ee\u66f4\u65b0\u4ee5\u83b7\u53d6\u6700\u65b0\u529f\u80fd\u548c\u4f18\u5316<\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"<p>VITA\u662f\u4e00\u4e2a\u9886\u5148\u7684\u5f00\u6e90\u4ea4\u4e92\u5f0f\u591a\u6a21\u6001\u5927\u8bed\u8a00\u6a21\u578b\u9879\u76ee\uff0c\u7387\u5148\u5b9e\u73b0\u4e86\u771f\u6b63\u7684\u5168\u65b9\u4f4d\u591a\u6a21\u6001\u4ea4\u4e92\u80fd\u529b\u3002\u8be5\u9879\u76ee\u4e8e2024\u5e748\u6708\u63a8\u51faVITA-1.0\u7248\u672c\uff0c\u5f00\u521b\u4e86\u9996\u4e2a\u5f00\u6e90\u4ea4\u4e92\u5f0f\u5168\u6a21\u6001\u5927\u8bed\u8a00\u6a21\u578b\u7684\u5148\u6cb3\u30022024\u5e7412\u6708\uff0c\u9879\u76ee\u63a8\u51fa\u4e86\u91cd\u5927\u5347\u7ea7\u7248\u672cVITA-1.5\uff0c&#8230;<\/p>\n","protected":false},"author":1,"featured_media":32782,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[392,402],"tags":[230,374],"class_list":["post-17506","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-models","category-multimodal-models","tag-aikaiyuanxiangmu","tag-duandaoduanyuyinzhubei"],"_links":{"self":[{"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/posts\/17506","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=17506"}],"version-history":[{"count":0,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/posts\/17506\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/media\/32782"}],"wp:attachment":[{"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/media?parent=17506"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/categories?post=17506"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/tags?post=17506"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}