{"id":30516,"date":"2025-04-14T02:39:03","date_gmt":"2025-04-13T18:39:03","guid":{"rendered":"https:\/\/www.aisharenet.com\/?p=30516"},"modified":"2025-04-14T02:39:03","modified_gmt":"2025-04-13T18:39:03","slug":"minimind-v","status":"publish","type":"post","link":"https:\/\/www.kdjingpai.com\/pt\/minimind-v\/","title":{"rendered":"MiniMind-V\uff1a1\u5c0f\u65f6\u8bad\u7ec326M\u53c2\u6570\u89c6\u89c9\u8bed\u8a00\u6a21\u578b"},"content":{"rendered":"<p>MiniMind-V \u662f\u4e00\u4e2a\u5f00\u6e90\u9879\u76ee\uff0c\u6258\u7ba1\u4e8e GitHub\uff0c\u65e8\u5728\u5e2e\u52a9\u7528\u6237\u5728 1 \u5c0f\u65f6\u5185\u8bad\u7ec3\u4e00\u4e2a\u4ec5 2600 \u4e07\u53c2\u6570\u7684\u8f7b\u91cf\u7ea7\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\uff08VLM\uff09\u3002\u5b83\u57fa\u4e8e MiniMind \u8bed\u8a00\u6a21\u578b\uff0c\u65b0\u589e\u89c6\u89c9\u7f16\u7801\u5668\u548c\u7279\u5f81\u6295\u5f71\u6a21\u5757\uff0c\u652f\u6301\u56fe\u50cf\u548c\u6587\u672c\u8054\u5408\u5904\u7406\u3002\u9879\u76ee\u63d0\u4f9b\u4ece\u6570\u636e\u96c6\u6e05\u6d17\u5230\u6a21\u578b\u63a8\u7406\u7684\u5b8c\u6574\u4ee3\u7801\uff0c\u8bad\u7ec3\u6210\u672c\u4f4e\u81f3\u7ea6 1.3 \u5143\u4eba\u6c11\u5e01\uff0c\u9002\u5408\u5355\u5f20 GPU\uff08\u5982 NVIDIA 3090\uff09\u3002MiniMind-V \u5f3a\u8c03\u7b80\u5355\u6613\u7528\uff0c\u4ee3\u7801\u6539\u52a8\u5c11\u4e8e 50 \u884c\uff0c\u9002\u5408\u5f00\u53d1\u8005\u5b9e\u9a8c\u548c\u5b66\u4e60\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\u7684\u6784\u5efa\u8fc7\u7a0b\u3002<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" class=\"\" title=\"MiniMind-V\uff1a1\u5c0f\u65f6\u8bad\u7ec326M\u53c2\u6570\u89c6\u89c9\u8bed\u8a00\u6a21\u578b-1\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/04\/2bf8046541b5c0a.png\" alt=\"MiniMind-V\uff1a1\u5c0f\u65f6\u8bad\u7ec326M\u53c2\u6570\u89c6\u89c9\u8bed\u8a00\u6a21\u578b-1\" width=\"710\" height=\"622\" \/><\/p>\n<p>&nbsp;<\/p>\n<h2>\u529f\u80fd\u5217\u8868<\/h2>\n<ul>\n<li>\u63d0\u4f9b 2600 \u4e07\u53c2\u6570\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\u7684\u5b8c\u6574\u8bad\u7ec3\u4ee3\u7801\uff0c\u652f\u6301\u5355 GPU \u5feb\u901f\u8bad\u7ec3\u3002<\/li>\n<li>\u4f7f\u7528 CLIP \u89c6\u89c9\u7f16\u7801\u5668\uff0c\u5904\u7406 224&#215;224 \u50cf\u7d20\u56fe\u50cf\uff0c\u751f\u6210 196 \u4e2a\u89c6\u89c9 token\u3002<\/li>\n<li>\u652f\u6301\u5355\u56fe\u548c\u591a\u56fe\u8f93\u5165\uff0c\u7ed3\u5408\u6587\u672c\u8fdb\u884c\u5bf9\u8bdd\u3001\u56fe\u50cf\u63cf\u8ff0\u6216\u95ee\u7b54\u3002<\/li>\n<li>\u5305\u542b\u6570\u636e\u96c6\u6e05\u6d17\u3001\u9884\u8bad\u7ec3\u548c\u76d1\u7763\u5fae\u8c03\uff08SFT\uff09\u7684\u5168\u6d41\u7a0b\u811a\u672c\u3002<\/li>\n<li>\u63d0\u4f9b PyTorch \u539f\u751f\u5b9e\u73b0\uff0c\u652f\u6301\u591a\u5361\u52a0\u901f\uff0c\u517c\u5bb9\u6027\u5f3a\u3002<\/li>\n<li>\u5305\u542b\u6a21\u578b\u6743\u91cd\u4e0b\u8f7d\uff0c\u652f\u6301 Hugging Face \u548c ModelScope \u5e73\u53f0\u3002<\/li>\n<li>\u63d0\u4f9b Web \u754c\u9762\u548c\u547d\u4ee4\u884c\u63a8\u7406\uff0c\u4fbf\u4e8e\u6d4b\u8bd5\u6a21\u578b\u6548\u679c\u3002<\/li>\n<li>\u652f\u6301 wandb \u5de5\u5177\uff0c\u8bb0\u5f55\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u7684\u635f\u5931\u548c\u6027\u80fd\u3002<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<h2>\u4f7f\u7528\u5e2e\u52a9<\/h2>\n<p>MiniMind-V \u7684\u4f7f\u7528\u6d41\u7a0b\u5305\u62ec\u73af\u5883\u914d\u7f6e\u3001\u6570\u636e\u51c6\u5907\u3001\u6a21\u578b\u8bad\u7ec3\u548c\u6548\u679c\u6d4b\u8bd5\u3002\u4ee5\u4e0b\u8be6\u7ec6\u4ecb\u7ecd\u6bcf\u4e2a\u6b65\u9aa4\uff0c\u5e2e\u52a9\u7528\u6237\u5feb\u901f\u4e0a\u624b\u3002<\/p>\n<h3>\u73af\u5883\u914d\u7f6e<\/h3>\n<p>MiniMind-V \u9700\u8981 Python \u73af\u5883\u548c GPU \u652f\u6301\u3002\u4ee5\u4e0b\u662f\u5b89\u88c5\u6b65\u9aa4\uff1a<\/p>\n<ol>\n<li><strong>\u514b\u9686\u4ee3\u7801<\/strong><br \/>\n\u5728\u7ec8\u7aef\u8fd0\u884c\u4ee5\u4e0b\u547d\u4ee4\uff0c\u4e0b\u8f7d\u9879\u76ee\u4ee3\u7801\uff1a<\/p>\n<pre><code>git clone https:\/\/github.com\/jingyaogong\/minimind-v\r\ncd minimind-v\r\n<\/code><\/pre>\n<\/li>\n<li><strong>\u5b89\u88c5\u4f9d\u8d56<\/strong><br \/>\n\u9879\u76ee\u63d0\u4f9b\u00a0<code>requirements.txt<\/code>\u00a0\u6587\u4ef6\uff0c\u5305\u542b\u6240\u9700\u5e93\u3002\u8fd0\u884c\u4ee5\u4e0b\u547d\u4ee4\uff1a<\/p>\n<pre><code>pip install -r requirements.txt -i https:\/\/pypi.tuna.tsinghua.edu.cn\/simple\r\n<\/code><\/pre>\n<p>\u5efa\u8bae\u4f7f\u7528 Python 3.9 \u6216\u4ee5\u4e0a\u7248\u672c\u3002\u786e\u4fdd PyTorch \u652f\u6301 CUDA\uff08\u82e5\u6709 GPU\uff09\u3002\u53ef\u8fd0\u884c\u4ee5\u4e0b\u4ee3\u7801\u9a8c\u8bc1\uff1a<\/p>\n<pre><code>import torch\r\nprint(torch.cuda.is_available())\r\n<\/code><\/pre>\n<p>\u8f93\u51fa\u00a0<code>True<\/code>\u00a0\u8868\u793a GPU \u53ef\u7528\u3002<\/li>\n<li><strong>\u4e0b\u8f7d CLIP \u6a21\u578b<\/strong><br \/>\nMiniMind-V \u4f7f\u7528 CLIP \u6a21\u578b\uff08<code>clip-vit-base-patch16<\/code>\uff09\u4f5c\u4e3a\u89c6\u89c9\u7f16\u7801\u5668\u3002\u8fd0\u884c\u4ee5\u4e0b\u547d\u4ee4\u4e0b\u8f7d\u5e76\u653e\u7f6e\u5230\u00a0<code>.\/model\/vision_model<\/code>\uff1a<\/p>\n<pre><code>git clone https:\/\/huggingface.co\/openai\/clip-vit-base-patch16 .\/model\/vision_model\r\n<\/code><\/pre>\n<p>\u4e5f\u53ef\u4ece ModelScope \u4e0b\u8f7d\uff1a<\/p>\n<pre><code>git clone https:\/\/www.modelscope.cn\/models\/openai-mirror\/clip-vit-base-patch16 .\/model\/vision_model\r\n<\/code><\/pre>\n<\/li>\n<li><strong>\u4e0b\u8f7d\u57fa\u5ea7\u8bed\u8a00\u6a21\u578b\u6743\u91cd<\/strong><br \/>\nMiniMind-V \u57fa\u4e8e MiniMind \u8bed\u8a00\u6a21\u578b\uff0c\u9700\u4e0b\u8f7d\u8bed\u8a00\u6a21\u578b\u6743\u91cd\u5230\u00a0<code>.\/out<\/code>\u00a0\u76ee\u5f55\u3002\u4f8b\u5982\uff1a<\/p>\n<pre><code>wget https:\/\/huggingface.co\/jingyaogong\/MiniMind2-V-PyTorch\/blob\/main\/lm_512.pth -P .\/out\r\n<\/code><\/pre>\n<p>\u6216\u4e0b\u8f7d\u00a0<code>lm_768.pth<\/code>\uff0c\u5177\u4f53\u89c6\u6a21\u578b\u914d\u7f6e\u800c\u5b9a\u3002<\/li>\n<\/ol>\n<h3>\u6570\u636e\u51c6\u5907<\/h3>\n<p>MiniMind-V \u4f7f\u7528\u7ea6 57 \u4e07\u5f20\u9884\u8bad\u7ec3\u56fe\u50cf\u548c 30 \u4e07\u6761\u6307\u4ee4\u5fae\u8c03\u6570\u636e\uff0c\u5b58\u50a8\u7a7a\u95f4\u7ea6 5GB\u3002\u64cd\u4f5c\u6b65\u9aa4\u5982\u4e0b\uff1a<\/p>\n<ol>\n<li><strong>\u521b\u5efa\u6570\u636e\u96c6\u76ee\u5f55<\/strong><br \/>\n\u5728\u9879\u76ee\u6839\u76ee\u5f55\u521b\u5efa\u00a0<code>.\/dataset<\/code>\u00a0\u6587\u4ef6\u5939\uff1a<\/p>\n<pre><code>mkdir dataset\r\n<\/code><\/pre>\n<\/li>\n<li><strong>\u4e0b\u8f7d\u6570\u636e\u96c6<\/strong><br \/>\n\u4ece Hugging Face \u6216 ModelScope \u4e0b\u8f7d\u6570\u636e\u96c6\uff0c\u5305\u542b\u00a0<code>*.jsonl<\/code>\u00a0\u95ee\u7b54\u6570\u636e\u548c\u00a0<code>*images<\/code>\u00a0\u56fe\u7247\u6570\u636e\uff1a<\/p>\n<ul>\n<li>Hugging Face:\u00a0https:\/\/huggingface.co\/datasets\/jingyaogong\/minimind-v_dataset<\/li>\n<li>ModelScope:\u00a0https:\/\/www.modelscope.cn\/datasets\/gongjy\/minimind-v_dataset<br \/>\n\u4e0b\u8f7d\u540e\u89e3\u538b\u56fe\u7247\u6570\u636e\u5230\u00a0<code>.\/dataset<\/code>\uff1a<\/li>\n<\/ul>\n<pre><code>unzip pretrain_images.zip -d .\/dataset\r\nunzip sft_images.zip -d .\/dataset\r\n<\/code><\/pre>\n<\/li>\n<li><strong>\u9a8c\u8bc1\u6570\u636e\u96c6<\/strong><br \/>\n\u786e\u4fdd\u00a0<code>.\/dataset<\/code>\u00a0\u5305\u542b\u4ee5\u4e0b\u6587\u4ef6\uff1a<\/p>\n<ul>\n<li><code>pretrain_vlm_data.jsonl<\/code>\uff1a\u9884\u8bad\u7ec3\u6570\u636e\uff0c\u7ea6 57 \u4e07\u6761\u3002<\/li>\n<li><code>sft_vlm_data.jsonl<\/code>\uff1a\u5355\u56fe\u5fae\u8c03\u6570\u636e\uff0c\u7ea6 30 \u4e07\u6761\u3002<\/li>\n<li><code>sft_vlm_data_multi.jsonl<\/code>\uff1a\u591a\u56fe\u5fae\u8c03\u6570\u636e\uff0c\u7ea6 1.36 \u4e07\u6761\u3002<\/li>\n<li>\u56fe\u7247\u6587\u4ef6\u5939\uff1a\u5305\u542b\u9884\u8bad\u7ec3\u548c\u5fae\u8c03\u7684\u56fe\u50cf\u6587\u4ef6\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ol>\n<h3>\u6a21\u578b\u8bad\u7ec3<\/h3>\n<p>MiniMind-V \u8bad\u7ec3\u5206\u4e3a\u9884\u8bad\u7ec3\u548c\u76d1\u7763\u5fae\u8c03\uff0c\u652f\u6301\u5355\u5361\u6216\u591a\u5361\u52a0\u901f\u3002<\/p>\n<ol>\n<li><strong>\u914d\u7f6e\u53c2\u6570<\/strong><br \/>\n\u7f16\u8f91\u00a0<code>.\/model\/LMConfig.py<\/code>\uff0c\u8bbe\u7f6e\u6a21\u578b\u53c2\u6570\u3002\u4f8b\u5982\uff1a<\/p>\n<ul>\n<li>\u5c0f\u578b\u6a21\u578b\uff1a<code>dim=512<\/code>,\u00a0<code>n_layers=8<\/code><\/li>\n<li>\u4e2d\u578b\u6a21\u578b\uff1a<code>dim=768<\/code>,\u00a0<code>n_layers=16<\/code><br \/>\n\u8fd9\u4e9b\u53c2\u6570\u51b3\u5b9a\u6a21\u578b\u5927\u5c0f\u548c\u6027\u80fd\u3002<\/li>\n<\/ul>\n<\/li>\n<li><strong>\u9884\u8bad\u7ec3<\/strong><br \/>\n\u8fd0\u884c\u9884\u8bad\u7ec3\u811a\u672c\uff0c\u5b66\u4e60\u56fe\u50cf\u63cf\u8ff0\u80fd\u529b\uff1a<\/p>\n<pre><code>python train_pretrain_vlm.py --epochs 4\r\n<\/code><\/pre>\n<p>\u8f93\u51fa\u6743\u91cd\u4fdd\u5b58\u4e3a\u00a0<code>.\/out\/pretrain_vlm_512.pth<\/code>\uff08\u6216\u00a0<code>768.pth<\/code>\uff09\u3002\u5355\u5f20 NVIDIA 3090 \u7ea6\u9700 1 \u5c0f\u65f6\u5b8c\u6210 1 \u4e2a epoch\u3002\u51bb\u7ed3 CLIP \u6a21\u578b\uff0c\u4ec5\u8bad\u7ec3\u6295\u5f71\u5c42\u548c\u8bed\u8a00\u6a21\u578b\u6700\u540e\u4e00\u5c42\u3002<\/li>\n<li><strong>\u76d1\u7763\u5fae\u8c03\uff08SFT\uff09<\/strong><br \/>\n\u4f7f\u7528\u9884\u8bad\u7ec3\u6743\u91cd\u8fdb\u884c\u5fae\u8c03\uff0c\u4f18\u5316\u5bf9\u8bdd\u80fd\u529b\uff1a<\/p>\n<pre><code>python train_sft_vlm.py --epochs 4\r\n<\/code><\/pre>\n<p>\u8f93\u51fa\u6743\u91cd\u4fdd\u5b58\u4e3a\u00a0<code>.\/out\/sft_vlm_512.pth<\/code>\u3002\u6b64\u6b65\u9aa4\u8bad\u7ec3\u6295\u5f71\u5c42\u548c\u8bed\u8a00\u6a21\u578b\u5168\u90e8\u53c2\u6570\u3002<\/li>\n<li><strong>\u591a\u5361\u8bad\u7ec3\uff08\u53ef\u9009\uff09<\/strong><br \/>\n\u82e5\u6709 N \u5f20\u663e\u5361\uff0c\u4f7f\u7528\u4ee5\u4e0b\u547d\u4ee4\u52a0\u901f\uff1a<\/p>\n<pre><code>torchrun --nproc_per_node N train_pretrain_vlm.py --epochs 4\r\n<\/code><\/pre>\n<p>\u66ff\u6362\u00a0<code>train_pretrain_vlm.py<\/code>\u00a0\u4e3a\u5176\u4ed6\u8bad\u7ec3\u811a\u672c\uff08\u5982\u00a0<code>train_sft_vlm.py<\/code>\uff09\u3002<\/li>\n<li><strong>\u76d1\u63a7\u8bad\u7ec3<\/strong><br \/>\n\u53ef\u4f7f\u7528 wandb \u8bb0\u5f55\u8bad\u7ec3\u635f\u5931\uff1a<\/p>\n<pre><code>python train_pretrain_vlm.py --epochs 4 --use_wandb\r\n<\/code><\/pre>\n<p>\u5728 wandb \u5b98\u7f51\u67e5\u770b\u5b9e\u65f6\u6570\u636e\u3002<\/li>\n<\/ol>\n<h3>\u6548\u679c\u6d4b\u8bd5<\/h3>\n<p>\u8bad\u7ec3\u5b8c\u6210\u540e\uff0c\u53ef\u6d4b\u8bd5\u6a21\u578b\u7684\u56fe\u50cf\u5bf9\u8bdd\u80fd\u529b\u3002<\/p>\n<ol>\n<li><strong>\u547d\u4ee4\u884c\u63a8\u7406<\/strong><br \/>\n\u8fd0\u884c\u4ee5\u4e0b\u547d\u4ee4\u52a0\u8f7d\u6a21\u578b\uff1a<\/p>\n<pre><code>python eval_vlm.py --load 1 --model_mode 1\r\n<\/code><\/pre>\n<ul>\n<li><code>--load 1<\/code>\uff1a\u4ece Hugging Face \u52a0\u8f7d transformers \u683c\u5f0f\u6a21\u578b\u3002<\/li>\n<li><code>--load 0<\/code>\uff1a\u4ece\u00a0<code>.\/out<\/code>\u00a0\u52a0\u8f7d PyTorch \u6743\u91cd\u3002<\/li>\n<li><code>--model_mode 1<\/code>\uff1a\u6d4b\u8bd5\u5fae\u8c03\u6a21\u578b\uff1b<code>0<\/code>\u00a0\u6d4b\u8bd5\u9884\u8bad\u7ec3\u6a21\u578b\u3002<\/li>\n<\/ul>\n<\/li>\n<li><strong>Web \u754c\u9762\u6d4b\u8bd5<\/strong><br \/>\n\u542f\u52a8 Web \u754c\u9762\uff1a<\/p>\n<pre><code>python web_demo_vlm.py\r\n<\/code><\/pre>\n<p>\u8bbf\u95ee\u00a0<code>http:\/\/localhost:8000<\/code>\uff0c\u4e0a\u4f20\u56fe\u7247\u5e76\u8f93\u5165\u6587\u672c\u6d4b\u8bd5\u3002<\/li>\n<li><strong>\u8f93\u5165\u683c\u5f0f<\/strong><br \/>\nMiniMind-V \u4f7f\u7528 196 \u4e2a\u00a0<code>@@@<\/code>\u00a0\u5360\u4f4d\u7b26\u8868\u793a\u4e00\u5f20\u56fe\u7247\u3002\u4f8b\u5982\uff1a<\/p>\n<pre><code>@@@...@@@\\n\u8fd9\u5f20\u56fe\u7247\u662f\u4ec0\u4e48\u5185\u5bb9\uff1f\r\n<\/code><\/pre>\n<p>\u591a\u56fe\u8f93\u5165\u793a\u4f8b\uff1a<\/p>\n<pre><code>@@@...@@@\\n\u7b2c\u4e00\u5f20\u56fe\u662f\u4ec0\u4e48\uff1f\\n@@@...@@@\\n\u7b2c\u4e8c\u5f20\u56fe\u662f\u4ec0\u4e48\uff1f\r\n<\/code><\/pre>\n<\/li>\n<li><strong>\u4e0b\u8f7d\u9884\u8bad\u7ec3\u6743\u91cd<\/strong><br \/>\n\u82e5\u4e0d\u8bad\u7ec3\uff0c\u53ef\u76f4\u63a5\u4e0b\u8f7d\u5b98\u65b9\u6743\u91cd\uff1a<\/p>\n<ul>\n<li>PyTorch \u683c\u5f0f\uff1a<a href=\"https:\/\/huggingface.co\/jingyaogong\/MiniMind2-V-PyTorch\">https:\/\/huggingface.co\/jingyaogong\/MiniMind2-V-PyTorch<\/a><\/li>\n<li>Transformers \u683c\u5f0f\uff1a<a href=\"https:\/\/huggingface.co\/collections\/jingyaogong\/minimind-v-67000833fb60b3a2e1f3597d\">https:\/\/huggingface.co\/collections\/jingyaogong\/minimind-v-67000833fb60b3a2e1f3597d<\/a><\/li>\n<\/ul>\n<\/li>\n<\/ol>\n<h3>\u6ce8\u610f\u4e8b\u9879<\/h3>\n<ul>\n<li>\u63a8\u8350\u663e\u5b58 24GB\uff08\u5982 RTX 3090\uff09\u3002\u82e5\u663e\u5b58\u4e0d\u8db3\uff0c\u51cf\u5c0f\u6279\u6b21\u5927\u5c0f\uff08<code>batch_size<\/code>\uff09\u3002<\/li>\n<li>\u786e\u4fdd\u6570\u636e\u96c6\u8def\u5f84\u6b63\u786e\uff0c<code>*.jsonl<\/code>\u00a0\u548c\u56fe\u7247\u6587\u4ef6\u9700\u653e\u5728\u00a0<code>.\/dataset<\/code>\u3002<\/li>\n<li>\u8bad\u7ec3\u65f6\u51bb\u7ed3 CLIP \u6a21\u578b\uff0c\u964d\u4f4e\u7b97\u529b\u9700\u6c42\u3002<\/li>\n<li>\u591a\u56fe\u5bf9\u8bdd\u6548\u679c\u6709\u9650\uff0c\u5efa\u8bae\u4f18\u5148\u6d4b\u8bd5\u5355\u56fe\u573a\u666f\u3002<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<h2>\u5e94\u7528\u573a\u666f<\/h2>\n<ol>\n<li><strong>AI \u7b97\u6cd5\u5b66\u4e60<\/strong><br \/>\nMiniMind-V \u63d0\u4f9b\u7b80\u6d01\u7684\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\u4ee3\u7801\uff0c\u9002\u5408\u5b66\u751f\u7406\u89e3\u8de8\u6a21\u6001\u6a21\u578b\u539f\u7406\u3002\u7528\u6237\u53ef\u4fee\u6539\u4ee3\u7801\uff0c\u5b9e\u9a8c\u4e0d\u540c\u53c2\u6570\u6216\u6570\u636e\u96c6\u3002<\/li>\n<li><strong>\u5feb\u901f\u539f\u578b\u9a8c\u8bc1<\/strong><br \/>\n\u5f00\u53d1\u8005\u53ef\u57fa\u4e8e MiniMind-V \u6784\u5efa\u56fe\u50cf\u5bf9\u8bdd\u5e94\u7528\u539f\u578b\u3002\u5b83\u8f7b\u91cf\u9ad8\u6548\uff0c\u9002\u5408\u4f4e\u7b97\u529b\u8bbe\u5907\uff0c\u5982\u4e2a\u4eba\u7535\u8111\u6216\u5d4c\u5165\u5f0f\u7cfb\u7edf\u3002<\/li>\n<li><strong>\u6559\u80b2\u57f9\u8bad\u5de5\u5177<\/strong><br \/>\n\u9ad8\u6821\u6216\u673a\u6784\u53ef\u5c06 MiniMind-V \u7528\u4e8e AI \u8bfe\u7a0b\uff0c\u5c55\u793a\u6a21\u578b\u8bad\u7ec3\u5168\u6d41\u7a0b\u3002\u4ee3\u7801\u6ce8\u91ca\u6e05\u6670\uff0c\u9002\u5408\u8bfe\u5802\u5b9e\u8df5\u3002<\/li>\n<li><strong>\u4f4e\u6210\u672c\u5b9e\u9a8c<\/strong><br \/>\n\u9879\u76ee\u8bad\u7ec3\u6210\u672c\u4f4e\uff0c\u9002\u5408\u9884\u7b97\u6709\u9650\u7684\u56e2\u961f\u6d4b\u8bd5\u591a\u6a21\u6001\u6a21\u578b\u6548\u679c\uff0c\u65e0\u9700\u9ad8\u6027\u80fd\u670d\u52a1\u5668\u3002<\/li>\n<\/ol>\n<p>&nbsp;<\/p>\n<h2>QA<\/h2>\n<ol>\n<li><strong>MiniMind-V \u652f\u6301\u591a\u5927\u56fe\u7247\uff1f<\/strong><br \/>\n\u9ed8\u8ba4\u5904\u7406 224&#215;224 \u50cf\u7d20\u56fe\u7247\uff0c\u53d7 CLIP \u6a21\u578b\u9650\u5236\u3002\u6570\u636e\u96c6\u56fe\u7247\u53ef\u80fd\u538b\u7f29\u81f3 128&#215;128 \u4ee5\u8282\u7701\u7a7a\u95f4\u3002\u672a\u6765\u53ef\u5c1d\u8bd5\u66f4\u5927\u5206\u8fa8\u7387 CLIP \u6a21\u578b\u3002<\/li>\n<li><strong>\u8bad\u7ec3\u9700\u8981\u591a\u5c11\u65f6\u95f4\uff1f<\/strong><br \/>\n\u5355\u5f20 NVIDIA 3090 \u4e0a\uff0c\u9884\u8bad\u7ec3 1 \u4e2a epoch \u7ea6 1 \u5c0f\u65f6\uff0c\u5fae\u8c03\u7a0d\u5feb\u3002\u5177\u4f53\u65f6\u95f4\u56e0\u786c\u4ef6\u548c\u6570\u636e\u91cf\u800c\u5f02\u3002<\/li>\n<li><strong>\u53ef\u4ee5\u4e0d\u9884\u8bad\u7ec3\u76f4\u63a5\u5fae\u8c03\u5417\uff1f<\/strong><br \/>\n\u53ef\u4ee5\u3002\u76f4\u63a5\u4e0b\u8f7d\u5b98\u65b9\u9884\u8bad\u7ec3\u6743\u91cd\uff0c\u8fd0\u884c\u00a0<code>train_sft_vlm.py<\/code>\u00a0\u8fdb\u884c\u5fae\u8c03\u3002<\/li>\n<li><strong>\u652f\u6301\u54ea\u4e9b\u8bed\u8a00\uff1f<\/strong><br \/>\n\u4e3b\u8981\u652f\u6301\u4e2d\u6587\u548c\u82f1\u6587\uff0c\u6548\u679c\u53d6\u51b3\u4e8e\u6570\u636e\u96c6\u3002\u7528\u6237\u53ef\u901a\u8fc7\u5fae\u8c03\u6269\u5c55\u5176\u4ed6\u8bed\u8a00\u3002<\/li>\n<li><strong>\u591a\u56fe\u5bf9\u8bdd\u6548\u679c\u5982\u4f55\uff1f<\/strong><br \/>\n\u5f53\u524d\u591a\u56fe\u5bf9\u8bdd\u80fd\u529b\u6709\u9650\uff0c\u5efa\u8bae\u4f18\u5148\u4f7f\u7528\u5355\u56fe\u573a\u666f\u3002\u672a\u6765\u53ef\u901a\u8fc7\u66f4\u5927\u6a21\u578b\u548c\u6570\u636e\u96c6\u6539\u8fdb\u3002<\/li>\n<\/ol>\n","protected":false},"excerpt":{"rendered":"<p>MiniMind-V \u662f\u4e00\u4e2a\u5f00\u6e90\u9879\u76ee\uff0c\u6258\u7ba1\u4e8e GitHub\uff0c\u65e8\u5728\u5e2e\u52a9\u7528\u6237\u5728 1 \u5c0f\u65f6\u5185\u8bad\u7ec3\u4e00\u4e2a\u4ec5 2600 \u4e07\u53c2\u6570\u7684\u8f7b\u91cf\u7ea7\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\uff08VLM\uff09\u3002\u5b83\u57fa\u4e8e MiniMind \u8bed\u8a00\u6a21\u578b\uff0c\u65b0\u589e\u89c6\u89c9\u7f16\u7801\u5668\u548c\u7279\u5f81\u6295\u5f71\u6a21\u5757\uff0c\u652f\u6301\u56fe\u50cf\u548c\u6587\u672c\u8054\u5408\u5904\u7406\u3002\u9879\u76ee&#8230;<\/p>\n","protected":false},"author":1,"featured_media":62267,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[20],"tags":[230,365],"class_list":["post-30516","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-tool","tag-aikaiyuanxiangmu","tag-damoxingweidiao"],"_links":{"self":[{"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/posts\/30516","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=30516"}],"version-history":[{"count":0,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/posts\/30516\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/media\/62267"}],"wp:attachment":[{"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/media?parent=30516"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/categories?post=30516"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/pt\/wp-json\/wp\/v2\/tags?post=30516"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}