{"id":30692,"date":"2025-04-22T13:35:16","date_gmt":"2025-04-22T05:35:16","guid":{"rendered":"https:\/\/www.aisharenet.com\/?p=30692"},"modified":"2025-04-22T13:35:16","modified_gmt":"2025-04-22T05:35:16","slug":"internvl","status":"publish","type":"post","link":"https:\/\/www.kdjingpai.com\/de\/internvl\/","title":{"rendered":"InternVL\uff1a\u5f00\u6e90\u591a\u6a21\u6001\u5927\u6a21\u578b\uff0c\u652f\u6301\u56fe\u50cf\u3001\u89c6\u9891\u548c\u6587\u672c\u5904\u7406"},"content":{"rendered":"<p>InternVL \u662f\u7531\u4e0a\u6d77\u4eba\u5de5\u667a\u80fd\u5b9e\u9a8c\u5ba4\uff08OpenGVLab\uff09\u5f00\u53d1\u7684\u4e00\u4e2a\u5f00\u6e90\u591a\u6a21\u6001\u5927\u6a21\u578b\u9879\u76ee\uff0c\u6258\u7ba1\u5728 GitHub \u4e0a\u3002\u5b83\u96c6\u6210\u4e86\u89c6\u89c9\u548c\u8bed\u8a00\u5904\u7406\u80fd\u529b\uff0c\u652f\u6301\u56fe\u50cf\u3001\u89c6\u9891\u548c\u6587\u672c\u7684\u7efc\u5408\u7406\u89e3\u4e0e\u751f\u6210\u3002InternVL \u7684\u76ee\u6807\u662f\u6253\u9020\u4e00\u4e2a\u5ab2\u7f8e\u5546\u4e1a\u6a21\u578b\uff08\u5982 GPT-4o\uff09\u7684\u5f00\u6e90\u66ff\u4ee3\u54c1\uff0c\u5e7f\u6cdb\u5e94\u7528\u4e8e\u89c6\u89c9\u611f\u77e5\u3001\u8de8\u6a21\u6001\u68c0\u7d22\u548c\u591a\u6a21\u6001\u5bf9\u8bdd\u7b49\u4efb\u52a1\u3002\u8be5\u9879\u76ee\u4ee5\u5176\u5f3a\u5927\u7684\u89c6\u89c9\u7f16\u7801\u5668\u3001\u52a8\u6001\u9ad8\u5206\u8fa8\u7387\u652f\u6301\u548c\u9ad8\u6548\u8bad\u7ec3\u7b56\u7565\u8457\u79f0\uff0c\u6a21\u578b\u89c4\u6a21\u4ece 1B \u5230 78B \u53c2\u6570\u4e0d\u7b49\uff0c\u9002\u5408\u4ece\u8fb9\u7f18\u8bbe\u5907\u5230\u9ad8\u6027\u80fd\u670d\u52a1\u5668\u7684\u591a\u79cd\u5e94\u7528\u573a\u666f\u3002\u4ee3\u7801\u3001\u6a21\u578b\u548c\u6570\u636e\u96c6\u5747\u5f00\u653e\uff0c\u9075\u5faa MIT \u8bb8\u53ef\uff0c\u9f13\u52b1\u7814\u7a76\u8005\u548c\u5f00\u53d1\u8005\u81ea\u7531\u4f7f\u7528\u4e0e\u6539\u8fdb\u3002<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" class=\"aligncenter wp-image-30693\" title=\"InternVL\uff1a\u5f00\u6e90\u591a\u6a21\u6001\u5927\u6a21\u578b\uff0c\u652f\u6301\u56fe\u50cf\u3001\u89c6\u9891\u548c\u6587\u672c\u5904\u7406-1\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/04\/565d761c88f50f3.jpg\" alt=\"InternVL\uff1a\u5f00\u6e90\u591a\u6a21\u6001\u5927\u6a21\u578b\uff0c\u652f\u6301\u56fe\u50cf\u3001\u89c6\u9891\u548c\u6587\u672c\u5904\u7406-1\" width=\"1045\" height=\"614\" srcset=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/04\/565d761c88f50f3.jpg 1743w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/04\/565d761c88f50f3-768x451.jpg 768w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/04\/565d761c88f50f3-1536x902.jpg 1536w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/04\/565d761c88f50f3-18x12.jpg 18w\" sizes=\"auto, (max-width: 1045px) 100vw, 1045px\" \/><\/p>\n<p>&nbsp;<\/p>\n<h2>\u529f\u80fd\u5217\u8868<\/h2>\n<ul>\n<li><strong>\u591a\u6a21\u6001\u5bf9\u8bdd<\/strong>\uff1a\u652f\u6301\u56fe\u50cf\u3001\u89c6\u9891\u548c\u6587\u672c\u8f93\u5165\uff0c\u751f\u6210\u81ea\u7136\u8bed\u8a00\u56de\u7b54\uff0c\u9002\u7528\u4e8e\u804a\u5929\u3001\u95ee\u7b54\u548c\u4efb\u52a1\u6307\u5bfc\u3002<\/li>\n<li><strong>\u56fe\u50cf\u5904\u7406<\/strong>\uff1a\u52a8\u6001\u5904\u7406\u9ad8\u8fbe 4K \u5206\u8fa8\u7387\u7684\u56fe\u50cf\uff0c\u652f\u6301\u56fe\u50cf\u5206\u7c7b\u3001\u5206\u5272\u548c\u7269\u4f53\u68c0\u6d4b\u3002<\/li>\n<li><strong>\u89c6\u9891\u7406\u89e3<\/strong>\uff1a\u5206\u6790\u89c6\u9891\u5185\u5bb9\uff0c\u8fdb\u884c\u96f6\u6837\u672c\u89c6\u9891\u5206\u7c7b\u548c\u6587\u672c-\u89c6\u9891\u68c0\u7d22\u3002<\/li>\n<li><strong>\u6587\u6863\u89e3\u6790<\/strong>\uff1a\u5904\u7406\u590d\u6742\u6587\u6863\uff0c\u64c5\u957f OCR\u3001\u8868\u683c\u8bc6\u522b\u548c\u6587\u6863\u95ee\u7b54\uff0c\u9002\u7528\u4e8e DocVQA \u7b49\u4efb\u52a1\u3002<\/li>\n<li><strong>\u591a\u8bed\u8a00\u652f\u6301<\/strong>\uff1a\u5185\u7f6e\u591a\u8bed\u8a00\u6587\u672c\u7f16\u7801\u5668\uff0c\u652f\u6301 110+ \u79cd\u8bed\u8a00\u7684\u751f\u6210\u4efb\u52a1\u3002<\/li>\n<li><strong>\u9ad8\u6548\u63a8\u7406<\/strong>\uff1a\u901a\u8fc7 LMDeploy \u63d0\u4f9b\u7b80\u5316\u7684\u63a8\u7406\u6d41\u7a0b\uff0c\u652f\u6301\u591a\u56fe\u50cf\u548c\u957f\u4e0a\u4e0b\u6587\u5904\u7406\u3002<\/li>\n<li><strong>\u6570\u636e\u96c6\u5f00\u653e<\/strong>\uff1a\u63d0\u4f9b ShareGPT-4o \u7b49\u5927\u89c4\u6a21\u591a\u6a21\u6001\u6570\u636e\u96c6\uff0c\u5305\u542b\u56fe\u50cf\u3001\u89c6\u9891\u548c\u97f3\u9891\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 InternVL\uff0c\u9700\u8981\u914d\u7f6e Python \u73af\u5883\u5e76\u5b89\u88c5\u76f8\u5173\u4f9d\u8d56\u3002\u4ee5\u4e0b\u662f\u8be6\u7ec6\u7684\u5b89\u88c5\u6b65\u9aa4\uff1a<\/p>\n<ol>\n<li><strong>\u514b\u9686\u4ed3\u5e93<\/strong><br \/>\n\u5728\u7ec8\u7aef\u8fd0\u884c\u4ee5\u4e0b\u547d\u4ee4\uff0c\u83b7\u53d6 InternVL \u6e90\u7801\uff1a<\/p>\n<pre><code>git clone https:\/\/github.com\/OpenGVLab\/InternVL.git\r\ncd InternVL\r\n<\/code><\/pre>\n<\/li>\n<li><strong>\u521b\u5efa\u865a\u62df\u73af\u5883<\/strong><br \/>\n\u4f7f\u7528 conda \u521b\u5efa Python 3.9 \u73af\u5883\u5e76\u6fc0\u6d3b\uff1a<\/p>\n<pre><code>conda create -n internvl python=3.9 -y\r\nconda activate internvl\r\n<\/code><\/pre>\n<\/li>\n<li><strong>\u5b89\u88c5\u4f9d\u8d56<\/strong><br \/>\n\u5b89\u88c5\u9879\u76ee\u6240\u9700\u4f9d\u8d56\uff0c\u9ed8\u8ba4\u5305\u542b\u591a\u6a21\u6001\u5bf9\u8bdd\u548c\u56fe\u50cf\u5904\u7406\u7684\u5fc5\u8981\u5e93\uff1a<\/p>\n<pre><code>pip install -r requirements.txt\r\n<\/code><\/pre>\n<p>\u5982\u679c\u9700\u8981\u989d\u5916\u7684\u529f\u80fd\uff08\u5982\u56fe\u50cf\u5206\u5272\u6216\u5206\u7c7b\uff09\uff0c\u53ef\u624b\u52a8\u5b89\u88c5\u7279\u5b9a\u4f9d\u8d56\uff1a<\/p>\n<pre><code>pip install -r requirements\/segmentation.txt\r\npip install -r requirements\/classification.txt\r\n<\/code><\/pre>\n<\/li>\n<li><strong>\u5b89\u88c5 Flash-Attention\uff08\u53ef\u9009\uff09<\/strong><br \/>\n\u4e3a\u52a0\u901f\u6a21\u578b\u63a8\u7406\uff0c\u5efa\u8bae\u5b89\u88c5 Flash-Attention\uff1a<\/p>\n<pre><code>pip install flash-attn==2.3.6 --no-build-isolation\r\n<\/code><\/pre>\n<p>\u6216\u8005\u4ece\u6e90\u7801\u7f16\u8bd1\uff1a<\/p>\n<pre><code>git clone https:\/\/github.com\/Dao-AILab\/flash-attention.git\r\ncd flash-attention\r\ngit checkout v2.3.6\r\npython setup.py install\r\n<\/code><\/pre>\n<\/li>\n<li><strong>\u5b89\u88c5 MMDeploy\uff08\u53ef\u9009\uff09<\/strong><br \/>\n\u5982\u679c\u9700\u8981\u90e8\u7f72\u6a21\u578b\u5230\u751f\u4ea7\u73af\u5883\uff0c\u5b89\u88c5 MMDeploy\uff1a<\/p>\n<pre><code>pip install -U openmim\r\nmim install mmdeploy\r\n<\/code><\/pre>\n<\/li>\n<\/ol>\n<h3>\u4f7f\u7528\u65b9\u6cd5<\/h3>\n<p>InternVL \u63d0\u4f9b\u591a\u79cd\u4f7f\u7528\u65b9\u5f0f\uff0c\u5305\u62ec\u547d\u4ee4\u884c\u63a8\u7406\u3001API \u670d\u52a1\u548c\u4ea4\u4e92\u5f0f\u6f14\u793a\u3002\u4ee5\u4e0b\u4ee5 InternVL2_5-8B \u6a21\u578b\u4e3a\u4f8b\uff0c\u4ecb\u7ecd\u4e3b\u8981\u529f\u80fd\u7684\u64cd\u4f5c\u6d41\u7a0b\uff1a<\/p>\n<h4>1. \u591a\u6a21\u6001\u5bf9\u8bdd<\/h4>\n<p>InternVL \u652f\u6301\u56fe\u50cf\u548c\u6587\u672c\u8f93\u5165\u7684\u5bf9\u8bdd\u3002\u4ee5\u4e0b\u662f\u4f7f\u7528 LMDeploy \u8fdb\u884c\u63a8\u7406\u7684\u793a\u4f8b\uff1a<\/p>\n<ul>\n<li><strong>\u51c6\u5907\u6a21\u578b\u548c\u56fe\u50cf<\/strong>\uff1a\u786e\u4fdd\u5df2\u4e0b\u8f7d\u6a21\u578b\uff08\u5982\u00a0<code>OpenGVLab\/InternVL2_5-8B<\/code>\uff09\u5e76\u51c6\u5907\u4e00\u5f20\u56fe\u50cf\uff08\u4f8b\u5982\u00a0<code>tiger.jpeg<\/code>\uff09\u3002<\/li>\n<li><strong>\u8fd0\u884c\u63a8\u7406<\/strong>\uff1a\u6267\u884c\u4ee5\u4e0b Python \u4ee3\u7801\uff0c\u63cf\u8ff0\u56fe\u50cf\u5185\u5bb9\uff1a\n<pre><code>from lmdeploy import pipeline, TurbomindEngineConfig\r\nfrom lmdeploy.vl import load_image\r\nmodel = 'OpenGVLab\/InternVL2_5-8B'\r\nimage = load_image('https:\/\/raw.githubusercontent.com\/open-mmlab\/mmdeploy\/main\/tests\/data\/tiger.jpeg')\r\npipe = pipeline(model, backend_config=TurbomindEngineConfig(session_len=8192))\r\nresponse = pipe(('\u63cf\u8ff0\u8fd9\u5f20\u56fe\u7247', image))\r\nprint(response.text)\r\n<\/code><\/pre>\n<\/li>\n<li><strong>\u7ed3\u679c<\/strong>\uff1a\u6a21\u578b\u4f1a\u8f93\u51fa\u56fe\u50cf\u7684\u8be6\u7ec6\u63cf\u8ff0\uff0c\u4f8b\u5982\u201c\u56fe\u7247\u4e2d\u662f\u4e00\u53ea\u7ad9\u7acb\u7684\u8001\u864e\uff0c\u80cc\u666f\u662f\u7eff\u8272\u8349\u5730\u201d\u3002<\/li>\n<\/ul>\n<h4>2. \u591a\u56fe\u50cf\u5904\u7406<\/h4>\n<p>InternVL \u652f\u6301\u540c\u65f6\u5904\u7406\u591a\u5f20\u56fe\u50cf\uff0c\u9002\u5408\u6bd4\u8f83\u6216\u7efc\u5408\u5206\u6790\uff1a<\/p>\n<ul>\n<li><strong>\u4ee3\u7801\u793a\u4f8b<\/strong>\uff1a\n<pre><code>from lmdeploy.vl.constants import IMAGE_TOKEN\r\nimage_urls = [\r\n'https:\/\/raw.githubusercontent.com\/open-mmlab\/mmdeploy\/main\/demo\/resources\/human-pose.jpg',\r\n'https:\/\/raw.githubusercontent.com\/open-mmlab\/mmdeploy\/main\/demo\/resources\/det.jpg'\r\n]\r\nimages = [load_image(url) for url in image_urls]\r\nprompt = f'Image-1: {IMAGE_TOKEN}\\nImage-2: {IMAGE_TOKEN}\\n\u63cf\u8ff0\u8fd9\u4e24\u5f20\u56fe\u7247'\r\nresponse = pipe((prompt, images))\r\nprint(response.text)\r\n<\/code><\/pre>\n<\/li>\n<li><strong>\u7ed3\u679c<\/strong>\uff1a\u6a21\u578b\u4f1a\u5206\u522b\u63cf\u8ff0\u6bcf\u5f20\u56fe\u7247\u7684\u5185\u5bb9\uff0c\u5e76\u53ef\u80fd\u603b\u7ed3\u5b83\u4eec\u4e4b\u95f4\u7684\u5173\u7cfb\u3002<\/li>\n<\/ul>\n<h4>3. \u6587\u6863\u89e3\u6790<\/h4>\n<p>InternVL \u5728\u6587\u6863\u95ee\u7b54\uff08DocVQA\uff09\u548c\u8868\u683c\u8bc6\u522b\u4efb\u52a1\u4e2d\u8868\u73b0\u4f18\u5f02\u3002\u64cd\u4f5c\u6d41\u7a0b\u5982\u4e0b\uff1a<\/p>\n<ul>\n<li><strong>\u51c6\u5907\u6587\u6863\u56fe\u50cf<\/strong>\uff1a\u4e0a\u4f20\u5305\u542b\u6587\u5b57\u3001\u8868\u683c\u6216\u56fe\u8868\u7684\u56fe\u50cf\u3002<\/li>\n<li><strong>\u63d0\u95ee<\/strong>\uff1a\u4f7f\u7528\u7c7b\u4f3c\u201c\u63d0\u53d6\u8868\u683c\u4e2d\u7684\u6570\u636e\u201d\u6216\u201c\u603b\u7ed3\u6587\u6863\u5185\u5bb9\u201d\u7684\u63d0\u793a\u8bcd\u3002<\/li>\n<li><strong>\u4ee3\u7801\u793a\u4f8b<\/strong>\uff1a\n<pre><code>image = load_image('document.jpg')\r\nresponse = pipe(('\u63d0\u53d6\u56fe\u7247\u4e2d\u8868\u683c\u7684\u5185\u5bb9', image))\r\nprint(response.text)\r\n<\/code><\/pre>\n<\/li>\n<li><strong>\u7ed3\u679c<\/strong>\uff1a\u6a21\u578b\u4f1a\u8fd4\u56de\u8868\u683c\u7684\u7ed3\u6784\u5316\u6570\u636e\u6216\u6587\u6863\u7684\u6458\u8981\u3002<\/li>\n<\/ul>\n<h4>4. \u90e8\u7f72 API \u670d\u52a1<\/h4>\n<p>\u4e3a\u65b9\u4fbf\u751f\u4ea7\u73af\u5883\u4f7f\u7528\uff0cInternVL \u652f\u6301\u901a\u8fc7 LMDeploy \u90e8\u7f72 RESTful API\uff1a<\/p>\n<ul>\n<li><strong>\u542f\u52a8\u670d\u52a1<\/strong>\uff1a\n<pre><code>lmdeploy serve api_server OpenGVLab\/InternVL2_5-8B --server-port 23333\r\n<\/code><\/pre>\n<\/li>\n<li><strong>\u8bbf\u95ee API<\/strong>\uff1a\u4f7f\u7528 OpenAI \u517c\u5bb9\u7684\u63a5\u53e3\u53d1\u9001\u8bf7\u6c42\uff0c\u4f8b\u5982\u901a\u8fc7\u00a0<code>curl<\/code>\u00a0\u6216 Python \u5ba2\u6237\u7aef\u8c03\u7528\u6a21\u578b\u3002<\/li>\n<\/ul>\n<h4>5. \u5728\u7ebf\u6f14\u793a<\/h4>\n<p>OpenGVLab \u63d0\u4f9b\u5728\u7ebf\u6f14\u793a\u5e73\u53f0\uff08<code>https:\/\/internvl.opengvlab.com\/<\/code>\uff09\uff0c\u65e0\u9700\u5b89\u88c5\u5373\u53ef\u4f53\u9a8c\uff1a<\/p>\n<ul>\n<li>\u8bbf\u95ee\u7f51\u7ad9\uff0c\u4e0a\u4f20\u56fe\u50cf\u6216\u89c6\u9891\uff0c\u8f93\u5165\u95ee\u9898\u3002<\/li>\n<li>\u6a21\u578b\u4f1a\u5b9e\u65f6\u8fd4\u56de\u7ed3\u679c\uff0c\u9002\u5408\u5feb\u901f\u6d4b\u8bd5\u3002<\/li>\n<\/ul>\n<h3>\u7279\u8272\u529f\u80fd\u64cd\u4f5c<\/h3>\n<ul>\n<li><strong>\u52a8\u6001\u9ad8\u5206\u8fa8\u7387<\/strong>\uff1aInternVL \u81ea\u52a8\u5c06\u56fe\u50cf\u5206\u5272\u4e3a 448&#215;448 \u7684\u5c0f\u5757\uff0c\u652f\u6301\u9ad8\u8fbe 4K \u5206\u8fa8\u7387\u3002\u7528\u6237\u65e0\u9700\u624b\u52a8\u8c03\u6574\u56fe\u50cf\u5927\u5c0f\uff0c\u76f4\u63a5\u4e0a\u4f20\u5373\u53ef\u3002<\/li>\n<li><strong>\u89c6\u9891\u7406\u89e3<\/strong>\uff1a\u4e0a\u4f20\u89c6\u9891\u6587\u4ef6\uff0c\u7ed3\u5408\u63d0\u793a\u8bcd\uff08\u5982\u201c\u603b\u7ed3\u89c6\u9891\u5185\u5bb9\u201d\uff09\uff0c\u6a21\u578b\u4f1a\u5206\u6790\u5173\u952e\u5e27\u5e76\u751f\u6210\u63cf\u8ff0\u3002<\/li>\n<li><strong>\u591a\u8bed\u8a00\u751f\u6210<\/strong>\uff1a\u5728\u63d0\u793a\u8bcd\u4e2d\u6307\u5b9a\u8bed\u8a00\uff08\u5982\u201c\u7528\u6cd5\u8bed\u56de\u7b54\u201d\uff09\uff0c\u6a21\u578b\u53ef\u751f\u6210\u5bf9\u5e94\u8bed\u8a00\u7684\u56de\u7b54\u3002<\/li>\n<\/ul>\n<h3>\u6ce8\u610f\u4e8b\u9879<\/h3>\n<ul>\n<li>\u786e\u4fdd GPU \u5185\u5b58\u5145\u8db3\uff088B \u6a21\u578b\u9700\u7ea6 16GB GPU \u5185\u5b58\uff09\u3002<\/li>\n<li>\u5904\u7406\u591a\u56fe\u50cf\u6216\u957f\u89c6\u9891\u65f6\uff0c\u589e\u5927\u4e0a\u4e0b\u6587\u7a97\u53e3\uff08<code>session_len=16384<\/code>\uff09\u3002<\/li>\n<li>\u68c0\u67e5\u4f9d\u8d56\u7248\u672c\uff0c\u907f\u514d\u517c\u5bb9\u6027\u95ee\u9898\u3002<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<h2>\u5e94\u7528\u573a\u666f<\/h2>\n<ol>\n<li><strong>\u5b66\u672f\u7814\u7a76<\/strong><br \/>\n\u7814\u7a76\u4eba\u5458\u4f7f\u7528 InternVL \u5206\u6790\u79d1\u5b66\u56fe\u8868\u3001\u5904\u7406\u5b9e\u9a8c\u56fe\u50cf\u6216\u89e3\u6790\u8bba\u6587\u4e2d\u7684\u8868\u683c\u6570\u636e\u3002\u6a21\u578b\u7684\u9ad8\u7cbe\u5ea6 OCR \u548c\u6587\u6863\u7406\u89e3\u80fd\u529b\u5927\u5e45\u63d0\u9ad8\u6570\u636e\u63d0\u53d6\u6548\u7387\u3002<\/li>\n<li><strong>\u6559\u80b2\u8f85\u52a9<\/strong><br \/>\n\u6559\u5e08\u548c\u5b66\u751f\u5229\u7528 InternVL \u89e3\u7b54\u56fe\u50cf\u76f8\u5173\u7684\u4f5c\u4e1a\u95ee\u9898\uff0c\u4f8b\u5982\u89e3\u91ca\u5386\u53f2\u56fe\u7247\u6216\u5206\u6790\u5730\u7406\u56fe\u8868\u3002\u6a21\u578b\u7684\u591a\u8bed\u8a00\u652f\u6301\u9002\u5408\u56fd\u9645\u5316\u6559\u80b2\u573a\u666f\u3002<\/li>\n<li><strong>\u4f01\u4e1a\u6587\u6863\u5904\u7406<\/strong><br \/>\n\u4f01\u4e1a\u7528 InternVL \u81ea\u52a8\u5316\u5904\u7406\u626b\u63cf\u6587\u6863\u3001\u5408\u540c\u6216\u53d1\u7968\uff0c\u63d0\u53d6\u5173\u952e\u4fe1\u606f\u5e76\u751f\u6210\u62a5\u544a\uff0c\u8282\u7701\u4eba\u5de5\u6210\u672c\u3002<\/li>\n<li><strong>\u5185\u5bb9\u521b\u4f5c<\/strong><br \/>\n\u5185\u5bb9\u521b\u4f5c\u8005\u901a\u8fc7 InternVL \u5206\u6790\u89c6\u9891\u7d20\u6750\uff0c\u751f\u6210\u811a\u672c\u6216\u5b57\u5e55\uff0c\u63d0\u5347\u521b\u4f5c\u6548\u7387\u3002<\/li>\n<li><strong>\u667a\u80fd\u5ba2\u670d<\/strong><br \/>\n\u5ba2\u670d\u7cfb\u7edf\u96c6\u6210 InternVL\uff0c\u5904\u7406\u7528\u6237\u4e0a\u4f20\u7684\u56fe\u7247\uff08\u5982\u4ea7\u54c1\u6545\u969c\u7167\u7247\uff09\uff0c\u5feb\u901f\u8bca\u65ad\u95ee\u9898\u5e76\u63d0\u4f9b\u89e3\u51b3\u65b9\u6848\u3002<\/li>\n<\/ol>\n<p>&nbsp;<\/p>\n<h2>QA<\/h2>\n<ol>\n<li><strong>InternVL \u652f\u6301\u54ea\u4e9b\u6a21\u578b\u89c4\u6a21\uff1f<\/strong><br \/>\nInternVL \u63d0\u4f9b\u4ece 1B \u5230 78B \u53c2\u6570\u7684\u6a21\u578b\uff0c\u9002\u5408\u4e0d\u540c\u8bbe\u5907\u30021B \u6a21\u578b\u9002\u4e8e\u8fb9\u7f18\u8bbe\u5907\uff0c78B \u6a21\u578b\u6027\u80fd\u5ab2\u7f8e GPT-4o\u3002<\/li>\n<li><strong>\u5982\u4f55\u5904\u7406\u9ad8\u5206\u8fa8\u7387\u56fe\u50cf\uff1f<\/strong><br \/>\n\u6a21\u578b\u81ea\u52a8\u5c06\u56fe\u50cf\u5206\u5272\u4e3a 448&#215;448 \u7684\u5c0f\u5757\uff0c\u652f\u6301 4K \u5206\u8fa8\u7387\u3002\u76f4\u63a5\u4e0a\u4f20\u56fe\u50cf\uff0c\u65e0\u9700\u9884\u5904\u7406\u3002<\/li>\n<li><strong>\u662f\u5426\u652f\u6301\u89c6\u9891\u5206\u6790\uff1f<\/strong><br \/>\n\u662f\u7684\uff0cInternVL \u652f\u6301\u96f6\u6837\u672c\u89c6\u9891\u5206\u7c7b\u548c\u6587\u672c-\u89c6\u9891\u68c0\u7d22\u3002\u4e0a\u4f20\u89c6\u9891\u5e76\u8f93\u5165\u63d0\u793a\u8bcd\u5373\u53ef\u3002<\/li>\n<li><strong>\u6a21\u578b\u662f\u5426\u5f00\u6e90\uff1f<\/strong><br \/>\nInternVL \u5b8c\u5168\u5f00\u6e90\uff0c\u4ee3\u7801\u548c\u6a21\u578b\u6743\u91cd\u5728 GitHub \u4e0a\u63d0\u4f9b\uff0c\u9075\u5faa MIT \u8bb8\u53ef\u3002<\/li>\n<li><strong>\u5982\u4f55\u4f18\u5316\u63a8\u7406\u901f\u5ea6\uff1f<\/strong><br \/>\n\u5b89\u88c5 Flash-Attention \u5e76\u4f7f\u7528 GPU \u52a0\u901f\u3002\u8c03\u6574\u00a0<code>session_len<\/code>\u00a0\u53c2\u6570\u4ee5\u9002\u914d\u957f\u4e0a\u4e0b\u6587\u3002<\/li>\n<\/ol>\n","protected":false},"excerpt":{"rendered":"<p>InternVL \u662f\u7531\u4e0a\u6d77\u4eba\u5de5\u667a\u80fd\u5b9e\u9a8c\u5ba4\uff08OpenGVLab\uff09\u5f00\u53d1\u7684\u4e00\u4e2a\u5f00\u6e90\u591a\u6a21\u6001\u5927\u6a21\u578b\u9879\u76ee\uff0c\u6258\u7ba1\u5728 GitHub \u4e0a\u3002\u5b83\u96c6\u6210\u4e86\u89c6\u89c9\u548c\u8bed\u8a00\u5904\u7406\u80fd\u529b\uff0c\u652f\u6301\u56fe\u50cf\u3001\u89c6\u9891\u548c\u6587\u672c\u7684\u7efc\u5408\u7406\u89e3\u4e0e\u751f\u6210\u3002InternVL \u7684\u76ee\u6807\u662f\u6253\u9020\u4e00\u4e2a\u5ab2\u7f8e\u5546\u4e1a\u6a21\u578b\uff08\u5982 G&#8230;<\/p>\n","protected":false},"author":1,"featured_media":62303,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[20],"tags":[230],"class_list":["post-30692","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-tool","tag-aikaiyuanxiangmu"],"_links":{"self":[{"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/posts\/30692","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/comments?post=30692"}],"version-history":[{"count":0,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/posts\/30692\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/media\/62303"}],"wp:attachment":[{"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/media?parent=30692"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/categories?post=30692"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/tags?post=30692"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}