{"id":27234,"date":"2025-02-28T19:34:46","date_gmt":"2025-02-28T11:34:46","guid":{"rendered":"https:\/\/www.aisharenet.com\/?p=27234"},"modified":"2025-02-28T19:34:46","modified_gmt":"2025-02-28T11:34:46","slug":"r1-onevision","status":"publish","type":"post","link":"https:\/\/www.kdjingpai.com\/pt\/r1-onevision\/","title":{"rendered":"R1-Onevision\uff1a\u652f\u6301\u591a\u6a21\u6001\u63a8\u7406\u7684\u5f00\u6e90\u89c6\u89c9\u8bed\u8a00\u6a21\u578b"},"content":{"rendered":"<p>R1-Onevision \u662f\u4e00\u4e2a\u7531 Fancy-MLLM \u56e2\u961f\u5f00\u53d1\u7684\u5f00\u6e90\u591a\u6a21\u6001\u5927\u8bed\u8a00\u6a21\u578b\uff0c\u4e13\u6ce8\u4e8e\u89c6\u89c9\u4e0e\u8bed\u8a00\u7684\u6df1\u5ea6\u7ed3\u5408\uff0c\u80fd\u591f\u5904\u7406\u56fe\u50cf\u3001\u6587\u672c\u7b49\u591a\u6a21\u6001\u8f93\u5165\uff0c\u5e76\u5728\u89c6\u89c9\u63a8\u7406\u3001\u56fe\u50cf\u7406\u89e3\u3001\u6570\u5b66\u89e3\u9898\u7b49\u9886\u57df\u8868\u73b0\u51fa\u8272\u3002\u57fa\u4e8e <a href=\"https:\/\/www.kdjingpai.com\/en\/qwen25-vl\/\">Qwen2.5-VL<\/a> \u6a21\u578b\u4f18\u5316\uff0cR1-Onevision \u5728\u591a\u4e2a\u57fa\u51c6\u6d4b\u8bd5\u4e2d\u8d85\u8d8a\u4e86\u540c\u7c7b\u6a21\u578b\u5982 Qwen2.5-VL-7B\uff0c\u751a\u81f3\u6311\u6218\u4e86 GPT-4V \u7684\u80fd\u529b\u3002\u9879\u76ee\u6258\u7ba1\u4e8e GitHub\uff0c\u63d0\u4f9b\u6a21\u578b\u6743\u91cd\u3001\u6570\u636e\u96c6\u53ca\u4ee3\u7801\uff0c\u9002\u5408\u5f00\u53d1\u8005\u3001\u7814\u7a76\u4eba\u5458\u7528\u4e8e\u5b66\u672f\u63a2\u7d22\u6216\u5b9e\u9645\u5e94\u7528\u30022025\u5e742\u670824\u65e5\u53d1\u5e03\u4ee5\u6765\uff0c\u53d7\u5230\u5e7f\u6cdb\u5173\u6ce8\uff0c\u5c24\u5176\u5728\u89c6\u89c9\u63a8\u7406\u4efb\u52a1\u4e2d\u8868\u73b0\u62a2\u773c\u3002<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" class=\"aligncenter size-full wp-image-27235\" title=\"R1-Onevision\uff1a\u652f\u6301\u591a\u6a21\u6001\u63a8\u7406\u7684\u5f00\u6e90\u89c6\u89c9\u8bed\u8a00\u6a21\u578b-1\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/02\/155b0d9c9ff53af.jpg\" alt=\"R1-Onevision\uff1a\u652f\u6301\u591a\u6a21\u6001\u63a8\u7406\u7684\u5f00\u6e90\u89c6\u89c9\u8bed\u8a00\u6a21\u578b-1\" width=\"1727\" height=\"1513\" srcset=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/02\/155b0d9c9ff53af.jpg 1727w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/02\/155b0d9c9ff53af-768x673.jpg 768w\" sizes=\"auto, (max-width: 1727px) 100vw, 1727px\" \/><\/p>\n<p>&nbsp;<\/p>\n<h2>\u529f\u80fd\u5217\u8868<\/h2>\n<ul>\n<li><strong>\u591a\u6a21\u6001\u63a8\u7406<\/strong>\uff1a\u652f\u6301\u56fe\u50cf\u4e0e\u6587\u672c\u7ed3\u5408\u7684\u590d\u6742\u63a8\u7406\u4efb\u52a1\uff0c\u5982\u6570\u5b66\u9898\u89e3\u7b54\u3001\u79d1\u5b66\u95ee\u9898\u5206\u6790\u3002<\/li>\n<li><strong>\u56fe\u50cf\u7406\u89e3<\/strong>\uff1a\u80fd\u591f\u5206\u6790\u56fe\u7247\u5185\u5bb9\u5e76\u751f\u6210\u8be6\u7ec6\u63cf\u8ff0\u6216\u56de\u7b54\u76f8\u5173\u95ee\u9898\u3002<\/li>\n<li><strong>\u6570\u636e\u96c6\u652f\u6301<\/strong>\uff1a\u63d0\u4f9b R1-Onevision \u6570\u636e\u96c6\uff0c\u5305\u542b\u81ea\u7136\u573a\u666f\u3001OCR\u3001\u56fe\u8868\u7b49\u591a\u9886\u57df\u6570\u636e\u3002<\/li>\n<li><strong>\u6a21\u578b\u8bad\u7ec3<\/strong>\uff1a\u4f7f\u7528\u5f00\u6e90 LLama-Factory \u6846\u67b6\uff0c\u652f\u6301\u5168\u6a21\u578b\u76d1\u7763\u5fae\u8c03\uff08SFT\uff09\u3002<\/li>\n<li><strong>\u9ad8\u6027\u80fd\u8bc4\u4f30<\/strong>\uff1a\u5728 Mathvision\u3001Mathverse \u7b49\u6d4b\u8bd5\u4e2d\u5c55\u73b0\u4f18\u4e8e\u540c\u884c\u7684\u63a8\u7406\u80fd\u529b\u3002<\/li>\n<li><strong>\u5f00\u6e90\u8d44\u6e90<\/strong>\uff1a\u63d0\u4f9b\u6a21\u578b\u6743\u91cd\u548c\u4ee3\u7801\uff0c\u65b9\u4fbf\u4e8c\u6b21\u5f00\u53d1\u6216\u7814\u7a76\u3002<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<h2>\u4f7f\u7528\u5e2e\u52a9<\/h2>\n<h3>\u5b89\u88c5\u6d41\u7a0b<\/h3>\n<p>R1-Onevision \u662f\u4e00\u4e2a\u57fa\u4e8e GitHub \u7684\u5f00\u6e90\u9879\u76ee\uff0c\u9700\u8981\u4e00\u5b9a\u7684\u7f16\u7a0b\u57fa\u7840\u548c\u73af\u5883\u914d\u7f6e\u624d\u80fd\u8fd0\u884c\u3002\u4ee5\u4e0b\u662f\u8be6\u7ec6\u7684\u5b89\u88c5\u4e0e\u4f7f\u7528\u6307\u5357\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 Windows\uff08\u914d\u5408 WSL\uff09\u3002<\/li>\n<li><strong>\u786c\u4ef6\u8981\u6c42<\/strong>\uff1a\u5efa\u8bae\u914d\u5907 NVIDIA GPU\uff08\u81f3\u5c11 16GB \u663e\u5b58\uff0c\u5982 A100 \u6216 RTX 3090\uff09\uff0c\u4ee5\u652f\u6301\u6a21\u578b\u63a8\u7406\u548c\u8bad\u7ec3\u3002<\/li>\n<li><strong>\u4f9d\u8d56\u8f6f\u4ef6<\/strong>\uff1a\n<ul>\n<li>Python 3.8 \u6216\u66f4\u9ad8\u7248\u672c\u3002<\/li>\n<li>PyTorch\uff08\u63a8\u8350\u5b89\u88c5 GPU \u7248\u672c\uff0c\u53c2\u8003\u00a0PyTorch \u5b98\u7f51\uff09\u3002<\/li>\n<li>Git\uff08\u7528\u4e8e\u514b\u9686\u4ee3\u7801\u4ed3\u5e93\uff09\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<h4>2. \u514b\u9686\u4ed3\u5e93<\/h4>\n<p>\u6253\u5f00\u7ec8\u7aef\uff0c\u8fd0\u884c\u4ee5\u4e0b\u547d\u4ee4\u83b7\u53d6 R1-Onevision \u9879\u76ee\u4ee3\u7801\uff1a<\/p>\n<pre><code>git clone https:\/\/github.com\/Fancy-MLLM\/R1-Onevision.git\r\ncd R1-Onevision\r\n<\/code><\/pre>\n<h4>3. \u5b89\u88c5\u4f9d\u8d56<\/h4>\n<p>\u9879\u76ee\u4f9d\u8d56\u591a\u4e2a Python \u5e93\uff0c\u53ef\u901a\u8fc7\u4ee5\u4e0b\u547d\u4ee4\u5b89\u88c5\uff1a<\/p>\n<pre><code>pip install -r requirements.txt\r\n<\/code><\/pre>\n<p>\u82e5\u9700\u8981\u52a0\u901f\u63a8\u7406\uff0c\u63a8\u8350\u5b89\u88c5 Flash Attention\uff1a<\/p>\n<pre><code>pip install flash-attn --no-build-isolation\r\n<\/code><\/pre>\n<h4>4. \u4e0b\u8f7d\u6a21\u578b\u6743\u91cd<\/h4>\n<p>R1-Onevision \u63d0\u4f9b\u9884\u8bad\u7ec3\u6a21\u578b\uff0c\u53ef\u4ece Hugging Face \u4e0b\u8f7d\uff1a<\/p>\n<ul>\n<li>\u8bbf\u95ee\u00a0Hugging Face \u6a21\u578b\u9875\u9762\u3002<\/li>\n<li>\u4e0b\u8f7d\u6a21\u578b\u6587\u4ef6\uff08\u5982\u00a0<code>R1-Onevision-7B<\/code>\uff09\u5e76\u89e3\u538b\u81f3\u9879\u76ee\u76ee\u5f55\u4e0b\u7684\u00a0<code>models<\/code>\u00a0\u6587\u4ef6\u5939\uff08\u9700\u624b\u52a8\u521b\u5efa\uff09\u3002<\/li>\n<\/ul>\n<h4>5. \u914d\u7f6e\u73af\u5883<\/h4>\n<p>\u786e\u4fdd CUDA \u5df2\u6b63\u786e\u5b89\u88c5\u5e76\u4e0e PyTorch \u517c\u5bb9\uff0c\u53ef\u8fd0\u884c\u4ee5\u4e0b\u4ee3\u7801\u9a8c\u8bc1\uff1a<\/p>\n<pre><code>import torch\r\nprint(torch.cuda.is_available())  # \u8f93\u51fa True \u8868\u793a GPU \u53ef\u7528\r\n<\/code><\/pre>\n<h3>\u4f7f\u7528\u65b9\u6cd5<\/h3>\n<h4>\u57fa\u7840\u63a8\u7406\uff1a\u56fe\u50cf\u4e0e\u6587\u672c\u5206\u6790<\/h4>\n<p>R1-Onevision \u652f\u6301\u901a\u8fc7 Python \u811a\u672c\u8fd0\u884c\u63a8\u7406\u4efb\u52a1\u3002\u4ee5\u4e0b\u662f\u52a0\u8f7d\u6a21\u578b\u5e76\u5904\u7406\u56fe\u50cf\u4e0e\u6587\u672c\u7684\u793a\u4f8b\uff1a<\/p>\n<ol>\n<li><strong>\u7f16\u5199\u63a8\u7406\u811a\u672c<\/strong>\uff1a<br \/>\n\u5728\u9879\u76ee\u6839\u76ee\u5f55\u4e0b\u521b\u5efa\u4e00\u4e2a\u6587\u4ef6\uff08\u5982\u00a0<code>infer.py<\/code>\uff09\uff0c\u8f93\u5165\u4ee5\u4e0b\u4ee3\u7801\uff1a<\/li>\n<\/ol>\n<pre><code>from transformers import AutoProcessor, Qwen2_5_VLForConditionalGeneration\r\nimport torch\r\nfrom qwen_vl_utils import process_vision_info\r\n# \u52a0\u8f7d\u6a21\u578b\u548c\u5904\u7406\u5668\r\nMODEL_ID = \"models\/R1-Onevision-7B\"  # \u66ff\u6362\u4e3a\u6a21\u578b\u5b9e\u9645\u8def\u5f84\r\nprocessor = AutoProcessor.from_pretrained(MODEL_ID, trust_remote_code=True)\r\nmodel = Qwen2_5_VLForConditionalGeneration.from_pretrained(\r\nMODEL_ID, trust_remote_code=True, torch_dtype=torch.bfloat16\r\n).to(\"cuda\").eval()\r\n# \u8f93\u5165\u56fe\u50cf\u548c\u6587\u672c\r\nmessages = [\r\n{\r\n\"role\": \"user\",\r\n\"content\": [\r\n{\"type\": \"image\", \"image\": \"path\/to\/your\/image.jpg\"},  # \u66ff\u6362\u4e3a\u672c\u5730\u56fe\u50cf\u8def\u5f84\r\n{\"type\": \"text\", \"text\": \"\u8bf7\u63cf\u8ff0\u8fd9\u5f20\u56fe\u7247\u7684\u5185\u5bb9\u5e76\u56de\u7b54\uff1a\u56fe\u4e2d\u6709\u51e0\u4e2a\u4eba\uff1f\"}\r\n]\r\n}\r\n]\r\n# \u5904\u7406\u8f93\u5165\r\ninputs = processor(messages, return_tensors=\"pt\").to(\"cuda\")\r\noutputs = model.generate(**inputs, max_new_tokens=512)\r\nresponse = processor.decode(outputs[0], skip_special_tokens=True)\r\nprint(response)\r\n<\/code><\/pre>\n<ol start=\"2\">\n<li><strong>\u8fd0\u884c\u811a\u672c<\/strong>\uff1a<\/li>\n<\/ol>\n<pre><code>python infer.py\r\n<\/code><\/pre>\n<p>\u811a\u672c\u5c06\u8f93\u51fa\u56fe\u50cf\u63cf\u8ff0\u548c\u56de\u7b54\u3002\u4f8b\u5982\uff0c\u82e5\u56fe\u7247\u4e2d\u6709\u4e24\u4eba\uff0c\u6a21\u578b\u53ef\u80fd\u8fd4\u56de\uff1a\u201c\u56fe\u7247\u663e\u793a\u4e00\u4e2a\u516c\u56ed\u573a\u666f\uff0c\u6709\u4e24\u4e2a\u4eba\u5750\u5728\u957f\u6905\u4e0a\u3002\u201d<\/p>\n<h4>\u7279\u8272\u529f\u80fd\uff1a\u6570\u5b66\u63a8\u7406<\/h4>\n<p>R1-Onevision \u5728\u6570\u5b66\u89c6\u89c9\u63a8\u7406\u4e0a\u8868\u73b0\u7a81\u51fa\u3002\u5047\u8bbe\u6709\u4e00\u5f20\u5305\u542b\u6570\u5b66\u9898\u7684\u56fe\u7247\uff08\u5982\u201c2x + 3 = 7\uff0c\u6c42 x\u201d\uff09\uff0c\u53ef\u6309\u4ee5\u4e0b\u6b65\u9aa4\u64cd\u4f5c\uff1a<\/p>\n<ol>\n<li>\u4fee\u6539\u00a0<code>messages<\/code>\u00a0\u4e2d\u7684\u6587\u672c\u4e3a\uff1a\u201c\u8bf7\u89e3\u7b54\u8fd9\u5f20\u56fe\u7247\u4e2d\u7684\u6570\u5b66\u9898\uff0c\u5e76\u7ed9\u51fa\u8ba1\u7b97\u8fc7\u7a0b\u3002\u201d<\/li>\n<li>\u8fd0\u884c\u811a\u672c\uff0c\u6a21\u578b\u5c06\u8fd4\u56de\u7c7b\u4f3c\u4ee5\u4e0b\u7ed3\u679c\uff1a<\/li>\n<\/ol>\n<pre><code>\u56fe\u7247\u4e2d\u7684\u9898\u76ee\u662f\uff1a2x + 3 = 7\r\n\u89e3\u9898\u8fc7\u7a0b\uff1a\r\n1. \u4e24\u8fb9\u540c\u65f6\u51cf\u53bb 3\uff1a2x + 3 - 3 = 7 - 3\r\n2. \u7b80\u5316\u5f97\uff1a2x = 4\r\n3. \u4e24\u8fb9\u540c\u65f6\u9664\u4ee5 2\uff1a2x \/ 2 = 4 \/ 2\r\n4. \u5f97\u51fa\uff1ax = 2\r\n\u6700\u7ec8\u7b54\u6848\uff1ax = 2\r\n<\/code><\/pre>\n<h4>\u6570\u636e\u96c6\u4f7f\u7528<\/h4>\n<p>R1-Onevision \u63d0\u4f9b\u4e13\u7528\u6570\u636e\u96c6\uff0c\u53ef\u7528\u4e8e\u6a21\u578b\u5fae\u8c03\u6216\u6d4b\u8bd5\uff1a<\/p>\n<ul>\n<li>\u4e0b\u8f7d\u6570\u636e\u96c6\uff1aHugging Face \u6570\u636e\u96c6\u9875\u9762\u3002<\/li>\n<li>\u6570\u636e\u5305\u542b\u56fe\u50cf\u548c\u6587\u672c\u5bf9\uff0c\u89e3\u538b\u540e\u53ef\u76f4\u63a5\u7528\u4e8e\u8bad\u7ec3\u6216\u9a8c\u8bc1\u3002<\/li>\n<\/ul>\n<h4>\u6a21\u578b\u5fae\u8c03<\/h4>\n<p>\u82e5\u9700\u81ea\u5b9a\u4e49\u6a21\u578b\uff0c\u53ef\u4f7f\u7528 LLama-Factory \u8fdb\u884c\u76d1\u7763\u5fae\u8c03\uff1a<\/p>\n<ol>\n<li>\u5b89\u88c5 LLama-Factory\uff1a<\/li>\n<\/ol>\n<pre><code>git clone https:\/\/github.com\/hiyouga\/LLaMA-Factory.git\r\ncd LLaMA-Factory\r\npip install -r requirements.txt\r\n<\/code><\/pre>\n<ol start=\"2\">\n<li>\u914d\u7f6e\u8bad\u7ec3\u53c2\u6570\uff08\u53c2\u8003\u9879\u76ee\u6587\u6863\uff09\uff0c\u8fd0\u884c\uff1a<\/li>\n<\/ol>\n<pre><code>python train.py --model_name models\/R1-Onevision-7B --dataset path\/to\/dataset\r\n<\/code><\/pre>\n<h3>\u64cd\u4f5c\u6d41\u7a0b\u603b\u7ed3<\/h3>\n<ul>\n<li><strong>\u56fe\u50cf\u5206\u6790<\/strong>\uff1a\u51c6\u5907\u56fe\u50cf\u8def\u5f84\uff0c\u7f16\u5199\u811a\u672c\uff0c\u8fd0\u884c\u83b7\u53d6\u7ed3\u679c\u3002<\/li>\n<li><strong>\u6570\u5b66\u63a8\u7406<\/strong>\uff1a\u4e0a\u4f20\u9898\u76ee\u56fe\u7247\uff0c\u8f93\u5165\u95ee\u9898\uff0c\u67e5\u770b\u8be6\u7ec6\u89e3\u7b54\u3002<\/li>\n<li><strong>\u81ea\u5b9a\u4e49\u5f00\u53d1<\/strong>\uff1a\u4e0b\u8f7d\u6570\u636e\u96c6\u548c\u6a21\u578b\uff0c\u8c03\u6574\u53c2\u6570\u8fdb\u884c\u8bad\u7ec3\u3002<br \/>\n\u4f7f\u7528\u65f6\u9700\u6ce8\u610f GPU \u5185\u5b58\u5360\u7528\uff0c\u63a8\u8350\u81f3\u5c11 16GB \u663e\u5b58\u4ee5\u786e\u4fdd\u6d41\u7545\u8fd0\u884c\u3002<\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"<p>R1-Onevision \u662f\u4e00\u4e2a\u7531 Fancy-MLLM \u56e2\u961f\u5f00\u53d1\u7684\u5f00\u6e90\u591a\u6a21\u6001\u5927\u8bed\u8a00\u6a21\u578b\uff0c\u4e13\u6ce8\u4e8e\u89c6\u89c9\u4e0e\u8bed\u8a00\u7684\u6df1\u5ea6\u7ed3\u5408\uff0c\u80fd\u591f\u5904\u7406\u56fe\u50cf\u3001\u6587\u672c\u7b49\u591a\u6a21\u6001\u8f93\u5165\uff0c\u5e76\u5728\u89c6\u89c9\u63a8\u7406\u3001\u56fe\u50cf\u7406\u89e3\u3001\u6570\u5b66\u89e3\u9898\u7b49\u9886\u57df\u8868\u73b0\u51fa\u8272\u3002\u57fa\u4e8e Qwen2.5-VL \u6a21\u578b\u4f18\u5316\uff0cR1&#8230;<\/p>\n","protected":false},"author":1,"featured_media":61937,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[20],"tags":[230],"class_list":["post-27234","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\/27234","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=27234"}],"version-history":[{"count":0,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/posts\/27234\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/media\/61937"}],"wp:attachment":[{"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/media?parent=27234"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/categories?post=27234"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/tags?post=27234"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}