{"id":26666,"date":"2025-02-24T10:48:15","date_gmt":"2025-02-24T02:48:15","guid":{"rendered":"https:\/\/www.aisharenet.com\/?p=26666"},"modified":"2025-02-24T10:48:15","modified_gmt":"2025-02-24T02:48:15","slug":"internlm-xcomposer","status":"publish","type":"post","link":"https:\/\/www.kdjingpai.com\/en\/internlm-xcomposer\/","title":{"rendered":"InternLM-XComposer\uff1a\u8f93\u51fa\u8d85\u957f\u6587\u672c\u4e0e\u56fe\u50cf\u89c6\u9891\u7406\u89e3\u7684\u591a\u6a21\u6001\u5927\u6a21\u578b"},"content":{"rendered":"<p>InternLM-XComposer \u662f\u7531 InternLM \u56e2\u961f\u5f00\u53d1\u7684\u4e00\u4e2a\u5f00\u6e90\u56fe\u6587\u591a\u6a21\u6001\u5927\u6a21\u578b\u9879\u76ee\uff0c\u6258\u7ba1\u4e8e GitHub\u3002\u5b83\u57fa\u4e8e InternLM \u8bed\u8a00\u6a21\u578b\uff0c\u80fd\u591f\u5904\u7406\u6587\u672c\u3001\u56fe\u50cf\u3001\u89c6\u9891\u7b49\u591a\u6a21\u6001\u6570\u636e\uff0c\u5e7f\u6cdb\u5e94\u7528\u4e8e\u56fe\u6587\u521b\u4f5c\u3001\u56fe\u50cf\u7406\u89e3\u548c\u89c6\u9891\u5206\u6790\u7b49\u9886\u57df\u3002\u8be5\u9879\u76ee\u4ee5\u5176\u652f\u6301\u9ad8\u8fbe 96K \u957f\u4e0a\u4e0b\u6587\u3001\u5904\u7406 4K \u9ad8\u5206\u8fa8\u7387\u56fe\u50cf\u4ee5\u53ca\u7ec6\u7c92\u5ea6\u89c6\u9891\u7406\u89e3\u80fd\u529b\u800c\u8457\u79f0\uff0c\u4ec5\u4f7f\u7528 7B \u53c2\u6570\u5373\u53ef\u5ab2\u7f8e GPT-4V \u7684\u6027\u80fd\u3002\u7528\u6237\u53ef\u4ee5\u901a\u8fc7 GitHub \u8bbf\u95ee\u4ee3\u7801\u3001\u6a21\u578b\u6743\u91cd\u548c\u8be6\u7ec6\u6587\u6863\uff0c\u9002\u5408\u7814\u7a76\u4eba\u5458\u3001\u5f00\u53d1\u8005\u6216\u5bf9\u591a\u6a21\u6001 AI \u611f\u5174\u8da3\u7684\u7528\u6237\u4f7f\u7528\u3002\u622a\u81f3 2025 \u5e74 2 \u6708\uff0c\u8be5\u9879\u76ee\u5df2\u53d1\u5e03\u591a\u4e2a\u7248\u672c\uff0c\u5305\u62ec InternLM-XComposer-2.5 \u548c OmniLive\uff0c\u6301\u7eed\u4f18\u5316\u591a\u6a21\u6001\u4ea4\u4e92\u4f53\u9a8c\u3002<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" class=\"aligncenter size-full wp-image-26667\" title=\"InternLM-XComposer\uff1a\u8f93\u51fa\u8d85\u957f\u5185\u5bb9\u4e0e\u56fe\u50cf\u89c6\u9891\u7406\u89e3\u7684\u591a\u6a21\u6001\u5927\u6a21\u578b-1\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/02\/35a0f12eb8a74cc.png\" alt=\"InternLM-XComposer\uff1a\u8f93\u51fa\u8d85\u957f\u5185\u5bb9\u4e0e\u56fe\u50cf\u89c6\u9891\u7406\u89e3\u7684\u591a\u6a21\u6001\u5927\u6a21\u578b-1\" width=\"1358\" height=\"832\" srcset=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/02\/35a0f12eb8a74cc.png 1358w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/02\/35a0f12eb8a74cc-768x471.png 768w\" sizes=\"auto, (max-width: 1358px) 100vw, 1358px\" \/><\/p>\n<p>&nbsp;<\/p>\n<h2>\u529f\u80fd\u5217\u8868<\/h2>\n<ul>\n<li>\u652f\u6301\u8d85\u957f\u4e0a\u4e0b\u6587\u8f93\u51fa\uff1a\u5904\u7406\u957f\u8fbe 96K \u7684\u56fe\u6587\u6df7\u5408\u5185\u5bb9\uff0c\u9002\u5408\u590d\u6742\u4efb\u52a1\u3002<\/li>\n<li>\u9ad8\u5206\u8fa8\u7387\u56fe\u50cf\u7406\u89e3\uff1a\u652f\u6301\u4ece 336 \u50cf\u7d20\u5230 4K \u7684\u56fe\u50cf\u5206\u6790\uff0c\u7ec6\u8282\u6e05\u6670\u3002<\/li>\n<li>\u7ec6\u7c92\u5ea6\u89c6\u9891\u7406\u89e3\uff1a\u5c06\u89c6\u9891\u5206\u89e3\u4e3a\u591a\u5e27\u56fe\u50cf\uff0c\u6355\u6349\u52a8\u6001\u7ec6\u8282\u3002<\/li>\n<li>\u56fe\u6587\u521b\u4f5c\uff1a\u6839\u636e\u6307\u4ee4\u751f\u6210\u56fe\u6587\u5e76\u8302\u7684\u6587\u7ae0\u6216\u7f51\u9875\u5185\u5bb9\u3002<\/li>\n<li>\u591a\u8f6e\u591a\u56fe\u5bf9\u8bdd\uff1a\u652f\u6301\u591a\u5f20\u56fe\u7247\u8f93\u5165\uff0c\u8fdb\u884c\u8fde\u7eed\u5bf9\u8bdd\u5206\u6790\u3002<\/li>\n<li>\u5f00\u6e90\u6a21\u578b\u652f\u6301\uff1a\u63d0\u4f9b\u591a\u79cd\u6a21\u578b\u6743\u91cd\u548c\u5fae\u8c03\u4ee3\u7801\uff0c\u65b9\u4fbf\u4e8c\u6b21\u5f00\u53d1\u3002<\/li>\n<li>\u591a\u6a21\u6001\u6d41\u5a92\u4f53\u4ea4\u4e92\uff1aOmniLive \u7248\u672c\u652f\u6301\u957f\u65f6\u95f4\u89c6\u9891\u548c\u97f3\u9891\u5904\u7406\u3002<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<h2>\u4f7f\u7528\u5e2e\u52a9<\/h2>\n<p>InternLM-XComposer \u662f\u4e00\u4e2a\u57fa\u4e8e GitHub \u7684\u5f00\u6e90\u9879\u76ee\uff0c\u7528\u6237\u9700\u8981\u4e00\u5b9a\u7684\u7f16\u7a0b\u57fa\u7840\u6765\u5b89\u88c5\u548c\u4f7f\u7528\u3002\u4ee5\u4e0b\u662f\u8be6\u7ec6\u7684\u64cd\u4f5c\u6307\u5357\uff0c\u5e2e\u52a9\u7528\u6237\u5feb\u901f\u4e0a\u624b\u3002<\/p>\n<h3>\u5b89\u88c5\u6d41\u7a0b<\/h3>\n<p>&nbsp;<\/p>\n<p><strong>1.\u73af\u5883\u51c6\u5907<\/strong><\/p>\n<p>&nbsp;<\/p>\n<ul>\n<li>\n<ul>\n<li>\u786e\u4fdd\u4f60\u7684\u8bbe\u5907\u5b89\u88c5\u4e86 Python 3.9 \u6216\u4ee5\u4e0a\u7248\u672c\u3002<\/li>\n<li>\u9700\u8981 NVIDIA GPU \u548c CUDA \u652f\u6301\uff08\u63a8\u8350 CUDA 11.x \u6216 12.x\uff09\u3002<\/li>\n<li>\u5b89\u88c5 Git \u4ee5\u514b\u9686\u4ee3\u7801\u5e93\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<p><strong>2.\u514b\u9686\u9879\u76ee<\/strong><br \/>\n\u5728\u7ec8\u7aef\u8fd0\u884c\u4ee5\u4e0b\u547d\u4ee4\uff0c\u5c06\u9879\u76ee\u4e0b\u8f7d\u5230\u672c\u5730\uff1a<\/p>\n<p>&nbsp;<\/p>\n<pre>git clone https:\/\/github.com\/InternLM\/InternLM-XComposer.git\r\ncd InternLM-XComposer<\/pre>\n<p>3.\u00a0<strong>\u521b\u5efa\u865a\u62df\u73af\u5883<\/strong> \u00a0 \u4f7f\u7528 Conda \u6216\u865a\u62df\u73af\u5883\u5de5\u5177\u9694\u79bb\u4f9d\u8d56\uff1a<\/p>\n<pre>conda create -n internlm python=3.9 -y\r\nconda activate internlm<\/pre>\n<p>4.\u00a0<strong>\u5b89\u88c5\u4f9d\u8d56<\/strong> \u00a0 \u6839\u636e\u5b98\u65b9\u6587\u6863\u5b89\u88c5\u5fc5\u8981\u5e93\uff1a<\/p>\n<pre>pip install torch==2.0.1+cu117 torchvision==0.15.2+cu117 torchaudio==2.0.2 --index-url https:\/\/download.pytorch.org\/whl\/cu117\r\npip install transformers==4.33.2 timm==0.4.12 sentencepiece==0.1.99 gradio==4.13.0 markdown2==4.4.10 xlsxwriter==3.1.2 einops<\/pre>\n<p>&#8211; \u53ef\u9009\uff1a\u5b89\u88c5 flash-attention2 \u4ee5\u8282\u7701 GPU \u5185\u5b58\uff1a<\/p>\n<pre>pip install flash-attn --no-build-isolation<\/pre>\n<p>5.\u00a0<strong>\u4e0b\u8f7d\u6a21\u578b\u6743\u91cd<\/strong> \u00a0 \u9879\u76ee\u652f\u6301\u4ece Hugging Face \u4e0b\u8f7d\u9884\u8bad\u7ec3\u6a21\u578b\uff0c\u4f8b\u5982\uff1a<\/p>\n<pre>model = AutoModel.from_pretrained('internlm\/internlm-xcomposer2d5-7b', torch_dtype=torch.bfloat16, trust_remote_code=True).cuda().eval()<\/pre>\n<p>6.\u00a0<strong>\u9a8c\u8bc1\u5b89\u88c5<\/strong> \u00a0 \u8fd0\u884c\u793a\u4f8b\u4ee3\u7801\u6d4b\u8bd5\u73af\u5883\u662f\u5426\u6b63\u5e38\uff1a<\/p>\n<pre>python -m torch.distributed.run --nproc_per_node=1 example_code\/simple_chat.py<\/pre>\n<h3>\u4e3b\u8981\u529f\u80fd\u64cd\u4f5c\u6d41\u7a0b<\/h3>\n<h4>1. \u56fe\u6587\u521b\u4f5c<\/h4>\n<ul>\n<li><strong>\u529f\u80fd\u7b80\u4ecb<\/strong>\uff1a\u6839\u636e\u7528\u6237\u6307\u4ee4\u751f\u6210\u5305\u542b\u6587\u672c\u548c\u56fe\u7247\u7684\u5185\u5bb9\uff0c\u4f8b\u5982\u6587\u7ae0\u6216\u7f51\u9875\u3002<\/li>\n<li><strong>\u64cd\u4f5c\u6b65\u9aa4<\/strong>\uff1a<\/li>\n<\/ul>\n<ol>\n<li>\u51c6\u5907\u8f93\u5165\uff1a\u7f16\u5199\u6587\u672c\u6307\u4ee4\uff08\u5982\u201c\u5199\u4e00\u7bc7\u5173\u4e8e\u65c5\u884c\u7684\u6587\u7ae0\uff0c\u5305\u542b\u4e09\u5f20\u56fe\u7247\u201d\uff09\u3002<\/li>\n<li>\u8fd0\u884c\u4ee3\u7801\uff1a<\/li>\n<\/ol>\n<pre><code>from transformers import AutoModel, AutoTokenizer\r\nmodel = AutoModel.from_pretrained('internlm\/internlm-xcomposer2d5-7b', trust_remote_code=True).cuda().eval()\r\ntokenizer = AutoTokenizer.from_pretrained('internlm\/internlm-xcomposer2d5-7b', trust_remote_code=True)\r\nquery = \"\u5199\u4e00\u7bc7\u5173\u4e8e\u65c5\u884c\u7684\u6587\u7ae0\uff0c\u5305\u542b\u4e09\u5f20\u56fe\u7247\"\r\nresponse, _ = model.chat(tokenizer, query, do_sample=False, num_beams=3)\r\nprint(response)\r\n<\/code><\/pre>\n<ol start=\"3\">\n<li>\u8f93\u51fa\u7ed3\u679c\uff1a\u6a21\u578b\u4f1a\u751f\u6210\u56fe\u6587\u6df7\u5408\u5185\u5bb9\uff0c\u56fe\u7247\u63cf\u8ff0\u4f1a\u81ea\u52a8\u5d4c\u5165\u6587\u672c\u4e2d\u3002<\/li>\n<\/ol>\n<h4>2. \u9ad8\u5206\u8fa8\u7387\u56fe\u50cf\u7406\u89e3<\/h4>\n<ul>\n<li><strong>\u529f\u80fd\u7b80\u4ecb<\/strong>\uff1a\u5206\u6790\u9ad8\u5206\u8fa8\u7387\u56fe\u50cf\u5e76\u63d0\u4f9b\u8be6\u7ec6\u63cf\u8ff0\u3002<\/li>\n<li><strong>\u64cd\u4f5c\u6b65\u9aa4<\/strong>\uff1a<\/li>\n<\/ul>\n<ol>\n<li>\u51c6\u5907\u56fe\u7247\uff1a\u5c06\u56fe\u50cf\u6587\u4ef6\u653e\u5165\u672c\u5730\u76ee\u5f55\uff08\u4f8b\u5982\u00a0<code>examples\/dubai.png<\/code>\uff09\u3002<\/li>\n<li>\u8fd0\u884c\u4ee3\u7801\uff1a<\/li>\n<\/ol>\n<pre><code>query = \"\u8be6\u7ec6\u5206\u6790\u8fd9\u5f20\u56fe\u7247\"\r\nimage = ['examples\/dubai.png']\r\nwith torch.autocast(device_type='cuda', dtype=torch.float16):\r\nresponse, _ = model.chat(tokenizer, query, image, do_sample=False, num_beams=3)\r\nprint(response)\r\n<\/code><\/pre>\n<ol start=\"3\">\n<li>\u8f93\u51fa\u7ed3\u679c\uff1a\u6a21\u578b\u8fd4\u56de\u5bf9\u56fe\u50cf\u5185\u5bb9\u7684\u7ec6\u81f4\u63cf\u8ff0\uff0c\u4f8b\u5982\u5efa\u7b51\u3001\u989c\u8272\u7b49\u7ec6\u8282\u3002<\/li>\n<\/ol>\n<h4>3. \u89c6\u9891\u5206\u6790<\/h4>\n<ul>\n<li><strong>\u529f\u80fd\u7b80\u4ecb<\/strong>\uff1a\u5206\u89e3\u89c6\u9891\u5e27\u5e76\u63cf\u8ff0\u5185\u5bb9\u3002<\/li>\n<li><strong>\u64cd\u4f5c\u6b65\u9aa4<\/strong>\uff1a<\/li>\n<\/ul>\n<ol>\n<li>\u51c6\u5907\u89c6\u9891\uff1a\u4e0b\u8f7d\u793a\u4f8b\u89c6\u9891\uff08\u5982\u00a0<code>liuxiang.mp4<\/code>\uff09\u3002<\/li>\n<li>\u4f7f\u7528 OmniLive \u7248\u672c\uff1a<\/li>\n<\/ol>\n<pre><code>from lmdeploy import pipeline\r\npipe = pipeline('internlm\/internlm-xcomposer2d5-ol-7b')\r\nvideo = load_video('liuxiang.mp4')\r\nquery = \"\u63cf\u8ff0\u8fd9\u6bb5\u89c6\u9891\u5185\u5bb9\"\r\nresponse = pipe((query, video))\r\nprint(response.text)\r\n<\/code><\/pre>\n<ol start=\"3\">\n<li>\u8f93\u51fa\u7ed3\u679c\uff1a\u8fd4\u56de\u89c6\u9891\u5e27\u7684\u8be6\u7ec6\u63cf\u8ff0\uff0c\u4f8b\u5982\u52a8\u4f5c\u6216\u573a\u666f\u3002<\/li>\n<\/ol>\n<h4>4. \u591a\u8f6e\u591a\u56fe\u5bf9\u8bdd<\/h4>\n<ul>\n<li><strong>\u529f\u80fd\u7b80\u4ecb<\/strong>\uff1a\u652f\u6301\u591a\u5f20\u56fe\u7247\u8f93\u5165\uff0c\u8fdb\u884c\u8fde\u7eed\u5bf9\u8bdd\u3002<\/li>\n<li><strong>\u64cd\u4f5c\u6b65\u9aa4<\/strong>\uff1a<\/li>\n<\/ul>\n<ol>\n<li>\u51c6\u5907\u591a\u5f20\u56fe\u7247\uff08\u5982\u00a0<code>cars1.jpg<\/code>,\u00a0<code>cars2.jpg<\/code>,\u00a0<code>cars3.jpg<\/code>\uff09\u3002<\/li>\n<li>\u8fd0\u884c\u4ee3\u7801\uff1a<\/li>\n<\/ol>\n<pre><code>query = \"Image1 &lt;ImageHere&gt;; Image2 &lt;ImageHere&gt;; Image3 &lt;ImageHere&gt;; \u5206\u6790\u8fd9\u4e09\u8f86\u8f66\u7684\u4f18\u7f3a\u70b9\"\r\nimages = ['examples\/cars1.jpg', 'examples\/cars2.jpg', 'examples\/cars3.jpg']\r\nresponse, _ = model.chat(tokenizer, query, images, do_sample=False, num_beams=3)\r\nprint(response)\r\n<\/code><\/pre>\n<ol start=\"3\">\n<li>\u8f93\u51fa\u7ed3\u679c\uff1a\u6a21\u578b\u9010\u4e00\u5206\u6790\u6bcf\u5f20\u56fe\u7247\u5bf9\u5e94\u7684\u8f66\u8f86\u7279\u70b9\u3002<\/li>\n<\/ol>\n<h3>\u6ce8\u610f\u4e8b\u9879<\/h3>\n<ul>\n<li><strong>\u786c\u4ef6\u8981\u6c42<\/strong>\uff1a\u63a8\u8350\u81f3\u5c11 24GB GPU \u5185\u5b58\uff0c\u4f4e\u914d\u8bbe\u5907\u53ef\u5c1d\u8bd5 4-bit \u91cf\u5316\u7248\u672c\u3002<\/li>\n<li><strong>\u8c03\u8bd5\u6280\u5de7<\/strong>\uff1a\u82e5\u9047\u5230\u663e\u5b58\u4e0d\u8db3\uff0c\u964d\u4f4e\u00a0<code>hd_num<\/code>\u00a0\u53c2\u6570\uff08\u9ed8\u8ba4 18\uff09\u3002<\/li>\n<li><strong>\u793e\u533a\u652f\u6301<\/strong>\uff1a\u8bbf\u95ee GitHub Issues \u9875\u9762\u67e5\u770b\u5e38\u89c1\u95ee\u9898\u6216\u63d0\u4ea4\u53cd\u9988\u3002<\/li>\n<\/ul>\n<p>\u901a\u8fc7\u4ee5\u4e0a\u6b65\u9aa4\uff0c\u7528\u6237\u53ef\u4ee5\u8f7b\u677e\u5b89\u88c5\u5e76\u4f7f\u7528 InternLM-XComposer \u7684\u5f3a\u5927\u529f\u80fd\uff0c\u65e0\u8bba\u662f\u7814\u7a76\u8fd8\u662f\u5f00\u53d1\u90fd\u80fd\u5f97\u5fc3\u5e94\u624b\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"<p>InternLM-XComposer \u662f\u7531 InternLM \u56e2\u961f\u5f00\u53d1\u7684\u4e00\u4e2a\u5f00\u6e90\u56fe\u6587\u591a\u6a21\u6001\u5927\u6a21\u578b\u9879\u76ee\uff0c\u6258\u7ba1\u4e8e GitHub\u3002\u5b83\u57fa\u4e8e InternLM \u8bed\u8a00\u6a21\u578b\uff0c\u80fd\u591f\u5904\u7406\u6587\u672c\u3001\u56fe\u50cf\u3001\u89c6\u9891\u7b49\u591a\u6a21\u6001\u6570\u636e\uff0c\u5e7f\u6cdb\u5e94\u7528\u4e8e\u56fe\u6587\u521b\u4f5c\u3001\u56fe\u50cf\u7406\u89e3\u548c\u89c6\u9891\u5206\u6790\u7b49\u9886&#8230;<\/p>\n","protected":false},"author":1,"featured_media":61907,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[20],"tags":[230],"class_list":["post-26666","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-tool","tag-aikaiyuanxiangmu"],"_links":{"self":[{"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/posts\/26666","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/comments?post=26666"}],"version-history":[{"count":0,"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/posts\/26666\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/media\/61907"}],"wp:attachment":[{"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/media?parent=26666"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/categories?post=26666"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/tags?post=26666"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}