{"id":21416,"date":"2025-02-17T15:22:31","date_gmt":"2025-02-17T07:22:31","guid":{"rendered":"https:\/\/www.aisharenet.com\/?p=21416"},"modified":"2025-02-17T16:08:02","modified_gmt":"2025-02-17T08:08:02","slug":"pandianyu-ollama-lei","status":"publish","type":"post","link":"https:\/\/www.kdjingpai.com\/en\/pandianyu-ollama-lei\/","title":{"rendered":"\u76d8\u70b9\u4e0e Ollama \u7c7b\u4f3c\u7684 LLM \u6846\u67b6\uff1a\u672c\u5730\u90e8\u7f72\u5927\u6a21\u578b\u7684\u591a\u5143\u9009\u62e9"},"content":{"rendered":"<p>\u5728\u4eba\u5de5\u667a\u80fd\u548c\u5927\u578b\u8bed\u8a00\u6a21\u578b (LLMs) \u9886\u57df\uff0cOllama \u6846\u67b6\u7684\u51fa\u73b0\u65e0\u7591\u5438\u5f15\u4e86\u4f17\u591a\u76ee\u5149\u3002\u8fd9\u6b3e\u5f00\u6e90\u6846\u67b6\u4e13\u6ce8\u4e8e\u7b80\u5316\u5927\u578b\u8bed\u8a00\u6a21\u578b\u5728\u672c\u5730\u7684\u90e8\u7f72\u548c\u8fd0\u884c\uff0c\u8ba9\u66f4\u591a\u5f00\u53d1\u8005\u80fd\u591f\u8f7b\u677e\u4f53\u9a8c LLM \u7684\u9b45\u529b\u3002\u7136\u800c\uff0c\u653e\u773c\u5e02\u573a\uff0cOllama \u5e76\u975e\u5b64\u519b\u594b\u6218\uff0c\u8fd8\u6709\u4e0d\u5c11\u540c\u7c7b\u578b\u7684\u5de5\u5177\u6d8c\u73b0\uff0c\u5b83\u4eec\u5171\u540c\u4e3a\u5f00\u53d1\u8005\u63d0\u4f9b\u4e86\u66f4\u52a0\u591a\u6837\u5316\u7684\u9009\u62e9\u3002\u672c\u6587\u5c06\u6df1\u5165\u76d8\u70b9\u4e0e <a href=\"https:\/\/www.kdjingpai.com\/en\/ollama\/\">Ollama<\/a> \u7c7b\u4f3c\u7684\u51e0\u6b3e\u5927\u6a21\u578b\u6846\u67b6\u5de5\u5177\uff0c\u5e0c\u671b\u80fd\u5e2e\u52a9\u8bfb\u8005\u66f4\u5168\u9762\u5730\u4e86\u89e3\u8fd9\u4e00\u9886\u57df\u7684\u6280\u672f\u751f\u6001\uff0c\u627e\u5230\u6700\u9002\u5408\u81ea\u5df1\u7684\u5229\u5668\u3002<\/p>\n<p>&nbsp;<\/p>\n<h2>Ollama \u6846\u67b6\u7b80\u4ecb<\/h2>\n<p>Ollama \u81f4\u529b\u4e8e\u7b80\u5316\u5927\u578b\u8bed\u8a00\u6a21\u578b (LLM) \u5728\u672c\u5730\u73af\u5883\u4e2d\u7684\u90e8\u7f72\u548c\u8fd0\u884c\u3002\u5b83\u5e7f\u6cdb\u652f\u6301\u5404\u7c7b\u4e3b\u6d41\u5927\u6a21\u578b\uff0c\u4f8b\u5982 Llama 2\u3001Code Llama\u3001Mistral\u3001Gemma \u7b49\uff0c\u5e76\u5141\u8bb8\u7528\u6237\u6839\u636e\u81ea\u8eab\u9700\u6c42\u5b9a\u5236\u548c\u521b\u5efa\u4e13\u5c5e\u6a21\u578b\u3002Ollama \u63d0\u4f9b\u4e86\u7b80\u6d01\u7684 API \u4ee5\u53ca\u7c7b\u4f3c <a href=\"https:\/\/www.kdjingpai.com\/en\/chatgpt-6\/\">ChatGPT<\/a> \u7684\u804a\u5929\u754c\u9762\uff0c\u7528\u6237\u65e0\u9700\u989d\u5916\u5f00\u53d1\uff0c\u5373\u53ef\u76f4\u63a5\u4e0e\u6a21\u578b\u4e92\u52a8\u3002\u800c\u4e14\uff0cOllama \u4ee3\u7801\u7b80\u6d01\uff0c\u8fd0\u884c\u65f6\u8d44\u6e90\u5360\u7528\u5c11\uff0c\u975e\u5e38\u9002\u5408\u5728\u672c\u5730\u8ba1\u7b97\u673a\u4e0a\u8fd0\u884c\u3002<\/p>\n<p><img decoding=\"async\" title=\"Ollama\uff1a\u672c\u5730\u4e00\u952e\u90e8\u7f72\u5f00\u6e90\u5927\u8bed\u8a00\u6a21\u578b-1\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/02\/7bdc646dfbd340f.png\" alt=\"Ollama\uff1a\u672c\u5730\u4e00\u952e\u90e8\u7f72\u5f00\u6e90\u5927\u8bed\u8a00\u6a21\u578b-1\" \/><\/p>\n<p>&nbsp;<\/p>\n<h2>\u4e0e Ollama \u7c7b\u4f3c\u7684\u5de5\u5177<\/h2>\n<h3>vLLM \u7b80\u4ecb<\/h3>\n<p><a href=\"https:\/\/www.kdjingpai.com\/vllm\/\">vLLM<\/a> (Vectorized Large Language Model Serving System) \u662f\u4e00\u6b3e\u9ad8\u6548\u7684\u5927\u6a21\u578b\u63a8\u7406\u4e0e\u670d\u52a1\u5f15\u64ce\uff0c\u4e13\u4e3a LLM \u91cf\u8eab\u6253\u9020\u3002vLLM \u901a\u8fc7\u521b\u65b0\u7684 PagedAttention \u6280\u672f\u3001\u8fde\u7eed\u6279\u5904\u7406\u3001CUDA \u6838\u5fc3\u4f18\u5316\u548c\u5206\u5e03\u5f0f\u63a8\u7406\u652f\u6301\u7b49\u6280\u672f\uff0c\u663e\u8457\u63d0\u5347\u4e86 LLM \u7684\u63a8\u7406\u6027\u80fd\u3002<\/p>\n<p><strong>\u7279\u70b9<\/strong><\/p>\n<p><a href=\"https:\/\/www.kdjingpai.com\/en\/vllm\/\">vLLM<\/a> \u652f\u6301\u591a\u79cd\u6a21\u578b\u683c\u5f0f\uff0c\u5305\u62ec PyTorch\u3001TensorFlow \u7b49\uff0c\u65b9\u4fbf\u7528\u6237\u6839\u636e\u9700\u6c42\u7075\u6d3b\u9009\u62e9\u3002\u540c\u65f6\uff0cvLLM \u8fd8\u5177\u5907\u9ad8\u6027\u80fd\u7684\u63a8\u7406\u5f15\u64ce\uff0c\u652f\u6301\u5728\u7ebf\u63a8\u7406\u548c\u6279\u91cf\u63a8\u7406\uff0c\u80fd\u591f\u8fc5\u901f\u54cd\u5e94\u5927\u91cf\u5e76\u53d1\u8bf7\u6c42\uff0c\u5728\u9ad8\u8d1f\u8f7d\u573a\u666f\u4e0b\u4f9d\u7136\u8868\u73b0\u51fa\u8272\u3002<\/p>\n<p><img decoding=\"async\" title=\"vLLM\uff1a\u9ad8\u6548\u5185\u5b58\u5229\u7528\u7684LLM\u63a8\u7406\u548c\u670d\u52a1\u5f15\u64ce-1\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/02\/4b1a173278cef82.png\" alt=\"vLLM\uff1a\u9ad8\u6548\u5185\u5b58\u5229\u7528\u7684LLM\u63a8\u7406\u548c\u670d\u52a1\u5f15\u64ce-1\" \/><\/p>\n<p>&nbsp;<\/p>\n<h3>LightLLM \u7b80\u4ecb<\/h3>\n<p><a href=\"https:\/\/www.kdjingpai.com\/lightllm\/\">LightLLM<\/a> \u662f\u4e00\u6b3e\u57fa\u4e8e Python \u7684\u8f7b\u91cf\u7ea7\u9ad8\u6027\u80fd LLM \u63a8\u7406\u548c\u670d\u52a1\u6846\u67b6\u3002\u5b83\u96c6\u5404\u5bb6\u4e4b\u6240\u957f\uff0c\u501f\u9274\u5e76\u6574\u5408\u4e86 FasterTransformer\u3001TGI\u3001vLLM \u548c FlashAttention \u7b49\u5f00\u6e90\u5b9e\u73b0\u7684\u4f18\u70b9\uff0c\u4e3a\u7528\u6237\u63d0\u4f9b\u4e86\u4e00\u79cd\u5168\u65b0\u7684 LLM \u670d\u52a1\u6a21\u5f0f\u3002<\/p>\n<p><strong>\u7279\u70b9<\/strong><\/p>\n<p><a href=\"https:\/\/www.kdjingpai.com\/en\/litellm\/\">LightLLM<\/a> \u91c7\u7528\u4e86\u72ec\u7279\u7684\u4e09\u8fdb\u7a0b\u67b6\u6784\uff0c\u5c06\u5206\u8bcd (tokenize)\u3001\u6a21\u578b\u63a8\u65ad\u548c\u9006\u5206\u8bcd (detokenize) \u8fd9\u4e09\u5927\u6b65\u9aa4\u89e3\u8026\uff0c\u5e76\u901a\u8fc7\u5f02\u6b65\u534f\u4f5c\u673a\u5236\u5e76\u884c\u8fd0\u884c\u3002\u8fd9\u6837\u7684\u8bbe\u8ba1\u5927\u5e45\u63d0\u5347\u4e86 GPU \u7684\u5229\u7528\u7387\uff0c\u5e76\u51cf\u5c11\u4e86\u6570\u636e\u4f20\u8f93\u9020\u6210\u7684\u5ef6\u8fdf\uff0c\u6709\u6548\u63d0\u5347\u4e86\u63a8\u7406\u6548\u7387\u3002\u6b64\u5916\uff0cLightLLM \u8fd8\u652f\u6301 Nopad \u65e0\u586b\u5145\u64cd\u4f5c\uff0c\u80fd\u66f4\u6709\u6548\u5730\u5904\u7406\u957f\u5ea6\u5dee\u5f02\u5927\u7684\u8bf7\u6c42\uff0c\u907f\u514d\u65e0\u6548\u586b\u5145\uff0c\u8fdb\u800c\u63d0\u5347\u8d44\u6e90\u5229\u7528\u7387\u3002<\/p>\n<p><img decoding=\"async\" title=\"LightLLM\uff1a\u9ad8\u6548\u7684\u8f7b\u91cf\u7ea7\u5927\u8bed\u8a00\u6a21\u578b\u63a8\u7406\u548c\u670d\u52a1\u6846\u67b6-1\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/02\/58bee253dc22cef.png\" alt=\"LightLLM\uff1a\u9ad8\u6548\u7684\u8f7b\u91cf\u7ea7\u5927\u8bed\u8a00\u6a21\u578b\u63a8\u7406\u548c\u670d\u52a1\u6846\u67b6-1\" \/><\/p>\n<p>&nbsp;<\/p>\n<h3>llama.cpp \u7b80\u4ecb<\/h3>\n<p><a href=\"https:\/\/www.kdjingpai.com\/en\/llamacpp\/\">llama.cpp<\/a> \u662f\u4e00\u6b3e\u57fa\u4e8e C \u548c C++ \u5f00\u53d1\u7684 LLM \u63a8\u7406\u5f15\u64ce\u3002\u5b83\u9488\u5bf9\u82f9\u679c\u7845\u82af\u7247\u8fdb\u884c\u4e86\u6df1\u5ea6\u6027\u80fd\u4f18\u5316\uff0c\u80fd\u591f\u5728\u82f9\u679c\u8bbe\u5907\u4e0a\u9ad8\u6548\u8fd0\u884c Meta \u516c\u53f8\u63a8\u51fa\u7684 Llama 2 \u6a21\u578b\u3002<\/p>\n<p><strong>\u7279\u70b9<\/strong><\/p>\n<p>llama.cpp \u7684\u4e3b\u8981\u76ee\u6807\u662f\u5728\u5404\u79cd\u786c\u4ef6\u5e73\u53f0\u4e0a\u5b9e\u73b0 LLM \u63a8\u7406\uff0c\u53ea\u9700\u6781\u5c11\u7684\u914d\u7f6e\uff0c\u5c31\u80fd\u63d0\u4f9b\u9876\u5c16\u7684\u6027\u80fd\u3002\u4e3a\u4e86\u8fdb\u4e00\u6b65\u63d0\u5347\u6027\u80fd\uff0cllama.cpp \u63d0\u4f9b\u4e86 1.5 \u4f4d\u30012 \u4f4d\u30013 \u4f4d\u30014 \u4f4d\u30015 \u4f4d\u30016 \u4f4d\u548c 8 \u4f4d\u6574\u6570\u91cf\u5316\u7b49\u591a\u79cd\u91cf\u5316\u9009\u9879\uff0c\u65e8\u5728\u52a0\u901f\u63a8\u7406\u901f\u5ea6\u5e76\u964d\u4f4e\u5185\u5b58\u5360\u7528\u3002\u5e76\u4e14\uff0cllama.cpp \u8fd8\u652f\u6301 CPU\/GPU \u6df7\u5408\u63a8\u7406\uff0c\u8fdb\u4e00\u6b65\u589e\u5f3a\u4e86\u63a8\u7406\u7684\u7075\u6d3b\u6027\u548c\u6548\u7387\u3002<\/p>\n<p><img decoding=\"async\" title=\"llama.cpp\uff1a\u9ad8\u6548\u63a8\u7406\u5de5\u5177\uff0c\u652f\u6301\u591a\u79cd\u786c\u4ef6\uff0c\u8f7b\u677e\u5b9e\u73b0LLM\u63a8\u7406-1\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/02\/5d4e9f1bd340110.jpg\" alt=\"llama.cpp\uff1a\u9ad8\u6548\u63a8\u7406\u5de5\u5177\uff0c\u652f\u6301\u591a\u79cd\u786c\u4ef6\uff0c\u8f7b\u677e\u5b9e\u73b0LLM\u63a8\u7406-1\" \/><\/p>\n<p>&nbsp;<\/p>\n<h3>Xinference \u7b80\u4ecb<\/h3>\n<p><a href=\"https:\/\/www.kdjingpai.com\/inference\/\">Xinference<\/a> \u662f\u4e00\u6b3e\u65b0\u5174\u7684\u901a\u7528\u63a8\u7406\u6846\u67b6\uff0c\u4e0d\u4ec5\u652f\u6301 LLM\uff0c\u8fd8\u80fd\u652f\u6301\u5305\u62ec\u56fe\u50cf\u3001\u8bed\u97f3\u7b49\u591a\u79cd\u6a21\u6001\u7684\u6a21\u578b\u3002\u5b83\u65e8\u5728\u964d\u4f4e\u6a21\u578b\u90e8\u7f72\u548c\u4f7f\u7528\u7684\u95e8\u69db\uff0c\u63d0\u4f9b\u7edf\u4e00\u7684\u63a5\u53e3\u548c\u53cb\u597d\u7684\u7528\u6237\u4f53\u9a8c\u3002<\/p>\n<p><strong>\u7279\u70b9<\/strong><\/p>\n<p><a href=\"https:\/\/www.kdjingpai.com\/en\/xorbits-inference\/\">Xinference<\/a> \u7684\u4eae\u70b9\u5728\u4e8e\u5176\u5bf9\u591a\u6a21\u6001\u6a21\u578b\u7684\u5e7f\u6cdb\u652f\u6301\uff0c\u7528\u6237\u53ef\u4ee5\u5728\u540c\u4e00\u5e73\u53f0\u4e0a\u8f7b\u677e\u90e8\u7f72\u548c\u4f53\u9a8c\u4e0d\u540c\u7c7b\u578b\u7684\u6a21\u578b\u3002\u540c\u65f6\uff0cXinference \u6ce8\u91cd\u6613\u7528\u6027\uff0c\u63d0\u4f9b\u4e86 Web UI\u3001Python \u5ba2\u6237\u7aef\u7b49\u591a\u79cd\u4ea4\u4e92\u65b9\u5f0f\uff0c\u5e76\u652f\u6301\u6a21\u578b\u7684\u7edf\u4e00\u7ba1\u7406\u548c\u7075\u6d3b\u6269\u5c55\u3002<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" class=\"aligncenter size-full wp-image-21426\" title=\"Inference\uff1a\u8f7b\u677e\u5b9e\u73b0\u5206\u5e03\u5f0fAI\u6a21\u578b\u90e8\u7f72\u548c\u670d\u52a1-1\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/02\/78f145b3140f8a3.jpg\" alt=\"Inference\uff1a\u8f7b\u677e\u5b9e\u73b0\u5206\u5e03\u5f0fAI\u6a21\u578b\u90e8\u7f72\u548c\u670d\u52a1-1\" width=\"1726\" height=\"1026\" srcset=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/02\/78f145b3140f8a3.jpg 1726w, https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/02\/78f145b3140f8a3-768x457.jpg 768w\" sizes=\"auto, (max-width: 1726px) 100vw, 1726px\" \/><\/p>\n<p>&nbsp;<\/p>\n<h2>\u5de5\u5177\u5bf9\u6bd4\u4e0e\u5206\u6790<\/h2>\n<h3>\u5b89\u88c5\u4e0e\u90e8\u7f72<\/h3>\n<p><strong>Ollama:<\/strong>\u00a0\u5b89\u88c5\u8fc7\u7a0b\u6781\u5176\u7b80\u4fbf\uff0cOllama \u63d0\u4f9b\u4e86\u7b80\u6d01\u660e\u4e86\u7684\u5b89\u88c5\u6307\u5357\uff0c\u5e76\u652f\u6301\u591a\u79cd\u4e3b\u6d41\u64cd\u4f5c\u7cfb\u7edf\u3002\u7528\u6237\u53ea\u9700\u51e0\u6761\u7b80\u5355\u7684\u547d\u4ee4\u884c\u64cd\u4f5c\uff0c\u5373\u53ef\u8f7b\u677e\u542f\u52a8\u548c\u8fd0\u884c\u5927\u578b\u8bed\u8a00\u6a21\u578b\u3002<\/p>\n<p><strong>vLLM:<\/strong>\u00a0\u76f8\u6bd4\u4e4b\u4e0b\uff0cvLLM \u7684\u5b89\u88c5\u8fc7\u7a0b\u7565\u663e\u590d\u6742\uff0c\u9700\u8981\u7528\u6237<strong>\u9884\u5148<\/strong>\u914d\u7f6e Python \u73af\u5883\u4ee5\u53ca\u5b89\u88c5\u4e00\u7cfb\u5217\u4f9d\u8d56\u5e93\u3002\u4f46\u4e00\u65e6\u5b8c\u6210\u90e8\u7f72\uff0cvLLM \u7684\u9ad8\u6548\u63a8\u7406\u6027\u80fd\u5c06\u4e3a\u7528\u6237\u5e26\u6765\u663e\u8457\u7684\u63d0\u5347\uff0c\u5c24\u5176\u662f\u5728\u5bf9\u6027\u80fd\u6709\u8f83\u9ad8\u8981\u6c42\u7684\u573a\u666f\u4e0b\u3002<\/p>\n<p><strong>LightLLM:<\/strong>\u00a0LightLLM \u7684\u5b89\u88c5\u8fc7\u7a0b\u76f8\u5bf9\u53cb\u597d\uff0c\u5b98\u65b9\u63d0\u4f9b\u4e86\u8be6\u7ec6\u7684\u5b89\u88c5\u548c\u914d\u7f6e\u6307\u5357\uff0c\u6307\u5f15\u7528\u6237\u5feb\u901f\u4e0a\u624b\u3002\u7528\u6237\u53ef\u4ee5\u6839\u636e\u81ea\u8eab\u9700\u6c42\uff0c\u7075\u6d3b\u9009\u62e9\u9002\u5408\u7684\u6a21\u578b\u683c\u5f0f\u8fdb\u884c\u90e8\u7f72\u3002<\/p>\n<p><strong>llama.cpp:<\/strong>\u00a0llama.cpp \u7684\u90e8\u7f72\u5219\u5bf9\u6280\u672f\u57fa\u7840\u6709\u4e00\u5b9a\u8981\u6c42\uff0c\u7528\u6237\u9700\u8981<strong>\u63d0\u524d<\/strong>\u914d\u7f6e C++ \u5f00\u53d1\u73af\u5883\u3002\u7136\u800c\uff0c\u5bf9\u4e8e\u6280\u672f\u7231\u597d\u8005\u548c\u5f00\u53d1\u8005\u6765\u8bf4\uff0c\u4e00\u65e6\u914d\u7f6e\u5b8c\u6210\uff0cllama.cpp \u7684\u9ad8\u6027\u80fd\u63a8\u7406\u5f15\u64ce\u5c06\u4e3a\u7528\u6237\u5e26\u6765\u6781\u81f4\u7684\u4f7f\u7528\u4f53\u9a8c\uff0c\u6df1\u5ea6\u638c\u63a7\u6a21\u578b\u63a8\u7406\u7684\u6bcf\u4e00\u4e2a\u73af\u8282\u3002<\/p>\n<p><strong>Xinference:<\/strong> Xinference \u7684\u5b89\u88c5\u90e8\u7f72\u4e5f\u76f8\u5bf9\u7b80\u5355\uff0c\u63d0\u4f9b\u4e86 Docker \u955c\u50cf\u548c Python \u5305\u7b49\u591a\u79cd\u5b89\u88c5\u65b9\u5f0f\uff0c\u65b9\u4fbf\u7528\u6237\u5feb\u901f\u642d\u5efa\u73af\u5883\u3002\u540c\u65f6\uff0cXinference \u4e5f\u63d0\u4f9b\u4e86\u8f83\u4e3a\u5b8c\u5584\u7684 Web UI \u754c\u9762\uff0c\u964d\u4f4e\u4e86\u4f7f\u7528\u95e8\u69db\u3002<\/p>\n<p>&nbsp;<\/p>\n<h3>\u6a21\u578b\u652f\u6301\u4e0e\u517c\u5bb9\u6027<\/h3>\n<p><strong>Ollama:<\/strong>\u00a0\u5728\u6a21\u578b\u652f\u6301\u65b9\u9762\uff0cOllama \u5c55\u73b0\u51fa\u4e86\u826f\u597d\u7684\u5f00\u653e\u6027\uff0c\u5b83\u652f\u6301\u5305\u62ec Llama 2\u3001Code Llama \u7b49\u5728\u5185\u7684\u591a\u79cd\u5f00\u6e90\u6a21\u578b\uff0c\u5e76\u5141\u8bb8\u7528\u6237\u4e0a\u4f20\u81ea\u5b9a\u4e49\u6a21\u578b\u3002\u6b64\u5916\uff0cOllama \u8fd8\u8d34\u5fc3\u5730\u63d0\u4f9b\u4e86\u9884\u6784\u5efa\u6a21\u578b\u5e93\uff0c\u65b9\u4fbf\u7528\u6237\u5feb\u901f\u4e0a\u624b\u4f53\u9a8c\u5404\u7c7b\u6a21\u578b\u3002<\/p>\n<p><strong>vLLM:<\/strong>\u00a0vLLM \u5728\u6a21\u578b\u683c\u5f0f\u517c\u5bb9\u6027\u65b9\u9762\u8868\u73b0\u51fa\u8272\uff0c\u5b83\u652f\u6301 PyTorch\u3001TensorFlow \u7b49\u591a\u79cd\u4e3b\u6d41\u6a21\u578b\u683c\u5f0f\uff0c<strong>\u4f7f\u5f97<\/strong>\u00a0\u7528\u6237 \u043c\u0430\u043a\u0441\u0438\u043c\u0430\u043b\u044c\u043d\u043e \u7075\u6d3b\u5730\u9009\u62e9\u6a21\u578b\u3002\u503c\u5f97\u4e00\u63d0\u7684\u662f\uff0cvLLM \u8fd8\u652f\u6301\u5206\u5e03\u5f0f\u63a8\u7406\uff0c\u80fd\u591f\u5145\u5206\u5229\u7528\u591a\u53f0 GPU \u7684\u7b97\u529b\uff0c\u5e76\u884c\u8fd0\u884c\u6a21\u578b\uff0c\u8fdb\u4e00\u6b65\u63d0\u5347\u63a8\u7406\u6548\u7387\u3002<\/p>\n<p><strong>LightLLM:<\/strong>\u00a0LightLLM \u540c\u6837\u5177\u5907\u4f18\u79c0\u7684\u6a21\u578b\u683c\u5f0f\u517c\u5bb9\u6027\uff0c\u652f\u6301\u591a\u79cd\u5e38\u7528\u6a21\u578b\u683c\u5f0f\u3002\u540c\u65f6\uff0cLightLLM \u8fd8\u63d0\u4f9b\u4e86\u4e30\u5bcc\u7684 API \u548c\u5de5\u5177\uff0c\u65b9\u4fbf\u7528\u6237\u8fdb\u884c\u6df1\u5ea6\u5b9a\u5236\u5316\u5f00\u53d1\uff0c\u6ee1\u8db3\u66f4<strong>\u7279\u5b9a<\/strong>\u5e94\u7528\u573a\u666f\u7684\u9700\u6c42\u3002<\/p>\n<p><strong>llama.cpp:<\/strong>\u00a0llama.cpp \u4e3b\u8981\u4e13\u6ce8\u4e8e\u5bf9 Llama \u6a21\u578b\u7684\u6781\u81f4\u63a8\u7406\u4f18\u5316\uff0c\u4f46\u4e5f\u517c\u987e\u652f\u6301\u5176\u4ed6\u6a21\u578b\u3002\u5176\u9ad8\u6548\u7684\u63a8\u7406\u5f15\u64ce\uff0c\u4f7f\u5f97 llama.cpp \u5728\u5904\u7406\u5927\u89c4\u6a21\u6570\u636e\u96c6\u65f6\u4f9d\u7136\u80fd\u591f\u4fdd\u6301\u51fa\u8272\u7684\u6027\u80fd\u8868\u73b0\u3002<\/p>\n<p><strong>Xinference:<\/strong> Xinference \u7684\u6a21\u578b\u652f\u6301\u662f\u5176\u4e00\u5927\u4eae\u70b9\uff0c\u5b83\u4e0d\u4ec5\u652f\u6301\u5404\u79cd LLM\uff0c\u8fd8\u652f\u6301 Stable Diffusion \u7b49\u56fe\u50cf\u751f\u6210\u6a21\u578b\uff0c\u4ee5\u53ca <a href=\"https:\/\/www.kdjingpai.com\/en\/crisperwhisper\/\">Whisper<\/a> \u7b49\u8bed\u97f3\u6a21\u578b\uff0c\u5c55\u73b0\u51fa\u5f3a\u5927\u7684\u591a\u6a21\u6001\u6a21\u578b\u517c\u5bb9\u6027\uff0c\u4e3a\u7528\u6237\u63d0\u4f9b\u4e86\u66f4\u5e7f\u9614\u7684\u5e94\u7528\u7a7a\u95f4\u3002<\/p>\n<h3>\u6027\u80fd\u4e0e\u4f18\u5316<\/h3>\n<p><strong>Ollama:<\/strong>\u00a0Ollama \u4ee5\u5176\u7b80\u6d01\u7684 API \u548c\u9ad8\u6548\u7684\u63a8\u7406\u6027\u80fd\u8457\u79f0\uff0c\u4f46\u5728\u5e94\u5bf9\u5927\u91cf\u5e76\u53d1\u8bf7\u6c42\u7684\u573a\u666f\u4e0b\uff0c\u53ef\u80fd\u5b58\u5728\u6027\u80fd\u74f6\u9888\u3002\u5bf9\u4e8e\u8ffd\u6c42\u6613\u7528\u6027\u548c\u5feb\u901f\u90e8\u7f72\u7684\u7528\u6237\u6765\u8bf4\uff0cOllama \u4f9d\u7136\u662f\u4e0d\u9519\u7684\u9009\u62e9\u3002<\/p>\n<p><strong>vLLM:<\/strong>\u00a0vLLM \u4e3a\u4e86\u6781\u81f4\u7684\u6027\u80fd\u800c\u751f\u3002\u5b83\u901a\u8fc7 PagedAttention \u7b49\u6280\u672f\u663e\u8457\u63d0\u5347\u4e86 LLM \u63a8\u7406\u6027\u80fd\uff0c\u5c24\u5176\u5728\u5904\u7406\u5927\u91cf\u5e76\u53d1\u8bf7\u6c42\u65f6\u8868\u73b0\u7a81\u51fa\uff0c\u80fd\u591f\u6ee1\u8db3\u5bf9\u670d\u52a1\u6027\u80fd\u6709\u4e25\u82db\u8981\u6c42\u7684\u5e94\u7528\u573a\u666f\u3002<\/p>\n<p><strong>LightLLM:<\/strong>\u00a0LightLLM \u5728\u6027\u80fd\u4f18\u5316\u65b9\u9762\u4e5f\u4e0b\u8db3\u4e86\u529f\u592b\u3002\u5f97\u76ca\u4e8e\u5176\u72ec\u7279\u7684\u4e09\u8fdb\u7a0b\u67b6\u6784\u548c\u5f02\u6b65\u534f\u4f5c\u673a\u5236\uff0cLightLLM \u6709\u6548\u63d0\u5347\u4e86 GPU \u5229\u7528\u7387\u548c\u63a8\u7406\u901f\u5ea6\u3002 \u6b64\u5916\uff0cLightLLM \u8fd8\u652f\u6301 Nopad \u65e0\u586b\u5145\u64cd\u4f5c\uff0c\u8fdb\u4e00\u6b65\u63d0\u5347\u4e86\u8d44\u6e90\u5229\u7528\u7387\uff0c\u00a0<strong>\u4e3a\u6574\u4f53\u6027\u80fd\u505a\u51fa\u4e86\u8d21\u732e<\/strong>\u3002<\/p>\n<p><strong>llama.cpp:<\/strong>\u00a0llama.cpp \u5728\u6027\u80fd\u65b9\u9762\u540c\u6837\u8868\u73b0\u51fa\u8272\u3002\u5b83\u4e0d\u4ec5\u5177\u5907\u9ad8\u6548\u7684\u63a8\u7406\u5f15\u64ce\uff0c\u8fd8\u63d0\u4f9b\u4e86\u591a\u79cd\u91cf\u5316\u9009\u9879\uff0c\u53ef\u4ee5\u7075\u6d3b\u5730\u5728\u63a8\u7406\u901f\u5ea6\u548c\u5185\u5b58\u5360\u7528\u4e4b\u95f4\u8fdb\u884c<strong>\u5e73\u8861<\/strong>\u3002\u800c\u4e14\uff0cllama.cpp \u8fd8\u652f\u6301 CPU\/GPU \u6df7\u5408\u63a8\u7406\uff0c\u8fdb\u4e00\u6b65\u589e\u5f3a\u4e86\u63a8\u7406\u7684\u7075\u6d3b\u6027\u548c\u6548\u7387\uff0c\u4f7f\u5f97\u7528\u6237\u53ef\u4ee5\u6839\u636e\u5b9e\u9645\u786c\u4ef6\u73af\u5883\u8fdb\u884c<strong>\u6700\u4f73<\/strong>\u914d\u7f6e\u3002<\/p>\n<p><strong>Xinference:<\/strong> Xinference \u5728\u6027\u80fd\u65b9\u9762\u4e5f\u8fdb\u884c\u4e86\u4f18\u5316\uff0c\u652f\u6301\u6a21\u578b\u7684\u91cf\u5316\u548c\u526a\u679d\u7b49\u6280\u672f\uff0c\u4ee5\u63d0\u5347\u63a8\u7406\u6548\u7387\u548c\u964d\u4f4e\u8d44\u6e90\u6d88\u8017\u3002 \u6b64\u5916\uff0cXinference \u4e5f\u5728\u4e0d\u65ad\u8fed\u4ee3\u4f18\u5316\uff0c\u63d0\u5347\u591a\u6a21\u6001\u6a21\u578b\u63a8\u7406\u7684\u6027\u80fd\u3002<\/p>\n<h2>\u603b\u7ed3\u4e0e\u5c55\u671b<\/h2>\n<p>\u4e0e Ollama \u7c7b\u4f3c\u7684\u5de5\u5177\u5404\u6709\u7279\u70b9\u548c\u4f18\u52bf\uff0c\u7528\u6237\u53ef\u4ee5\u6839\u636e\u5e94\u7528\u573a\u666f\u548c\u5b9e\u9645\u9700\u6c42\u9009\u62e9\u5408\u9002\u7684\u6846\u67b6\u3002 \u968f\u7740\u4eba\u5de5\u667a\u80fd\u548c\u5927\u578b\u8bed\u8a00\u6a21\u578b\u6280\u672f\u7684\u6301\u7eed\u53d1\u5c55\uff0c\u8fd9\u4e9b\u5de5\u5177\u4e5f\u5c06\u4e0d\u65ad\u5b8c\u5584\u5347\u7ea7\uff0c\u4e3a\u7528\u6237\u63d0\u4f9b\u66f4\u9ad8\u6548\u4fbf\u6377\u7684\u89e3\u51b3\u65b9\u6848\u3002 \u5c55\u671b\u672a\u6765\uff0c\u8fd9\u4e9b\u5de5\u5177\u5c06\u5728\u81ea\u7136\u8bed\u8a00\u5904\u7406\u3001\u667a\u80fd\u5ba2\u670d\u3001\u6587\u672c\u751f\u6210\u7b49\u9886\u57df\u53d1\u6325\u66f4\u5927\u4f5c\u7528\uff0c\u8fdb\u4e00\u6b65\u63a8\u52a8\u4eba\u5de5\u667a\u80fd\u6280\u672f\u53d1\u5c55\u548c\u5e94\u7528\uff0c \u6700\u7ec8<strong>\u5b9e\u73b0<\/strong>\u00a0\u4eba\u5de5\u667a\u80fd\u666e\u60e0\u5927\u4f17\u7684\u7f8e\u597d\u613f\u666f\u3002<\/p>\n<h2>\u63a8\u8350\u4e0e\u5efa\u8bae<\/h2>\n<p>\u5bf9\u4e8e\u65b0\u624b\u548c\u4e2a\u4eba\u7528\u6237\u800c\u8a00\uff0cOllama \u6216\u8bb8\u662f\u4e0d\u9519\u7684\u9009\u62e9\u3002 Ollama \u63d0\u4f9b\u4e86\u7b80\u6d01\u7684 API \u548c\u53cb\u597d\u7684\u7528\u6237\u754c\u9762\uff0c\u7528\u6237\u65e0\u9700\u989d\u5916\u5f00\u53d1\u5373\u53ef\u76f4\u63a5\u4e0e\u6a21\u578b\u4e92\u52a8\u3002 \u800c\u4e14\uff0cOllama \u8fd8\u652f\u6301\u591a\u79cd\u5f00\u6e90\u6a21\u578b\uff0c\u65b9\u4fbf\u7528\u6237\u5feb\u901f\u4e0a\u624b\u5e76\u5c1d\u8bd5\u5404\u79cd\u5e94\u7528\u573a\u666f\uff0c \u5feb\u901f\u4f53\u9a8c\u672c\u5730\u90e8\u7f72 LLM \u7684\u4e50\u8da3\u3002<\/p>\n<p>\u5bf9\u4e8e\u9700\u8981\u5904\u7406\u9ad8\u5e76\u53d1\u8bf7\u6c42\u7684\u5546\u4e1a\u7528\u6237\uff0cvLLM \u53ef\u80fd\u662f\u66f4\u4f18\u9009\u62e9\u3002 vLLM \u7684\u9ad8\u6548\u63a8\u7406\u6027\u80fd\u548c\u5206\u5e03\u5f0f\u63a8\u7406\u652f\u6301\uff0c\u80fd\u591f\u6ee1\u8db3\u9ad8\u5e76\u53d1\u573a\u666f\u9700\u6c42\uff0c\u4e3a\u7528\u6237\u63d0\u4f9b\u66f4\u7a33\u5b9a\u53ef\u9760\u7684\u89e3\u51b3\u65b9\u6848\uff0c \u4fdd\u969c\u4e1a\u52a1\u7684<strong>\u7a33\u5b9a\u6027\u4e0e\u8fde\u7eed\u6027<\/strong>\u3002<\/p>\n<p>LightLLM \u548c llama.cpp \u5219\u66f4\u9002\u5408\u5177\u5907\u4e00\u5b9a\u6280\u672f\u57fa\u7840\u7684\u7528\u6237\uff0c\u8fdb\u884c\u5b9a\u5236\u5316\u5f00\u53d1\u548c\u6df1\u5ea6\u4f18\u5316\u3002 \u8fd9\u4e24\u6b3e\u5de5\u5177\u4e3a\u5f00\u53d1\u8005\u63d0\u4f9b\u4e86\u66f4\u5927\u7684\u7075\u6d3b\u6027\u548c\u66f4\u5e95\u5c42\u7684\u63a7\u5236\u6743\uff0c\u00a0<strong>\u5141\u8bb8<\/strong>\u00a0\u7528\u6237\u6839\u636e<strong>\u5177\u4f53\u8981\u6c42<\/strong>\u00a0\u6253\u9020\u66f4\u5177\u4e2a\u6027\u5316\u548c\u9ad8\u6027\u80fd\u7684 LLM \u5e94\u7528\u3002<\/p>\n<p>\u5bf9\u4e8e\u5e0c\u671b\u5c1d\u8bd5\u591a\u6a21\u6001\u6a21\u578b\uff0c\u5e76\u8ffd\u6c42\u5f00\u7bb1\u5373\u7528\u4f53\u9a8c\u7684\u7528\u6237\uff0c Xinference \u4e5f\u662f\u4e00\u4e2a\u503c\u5f97\u8003\u8651\u7684\u9009\u62e9\u3002\u00a0Xinference \u964d\u4f4e\u4e86\u591a\u6a21\u6001\u6a21\u578b\u7684\u4f7f\u7528\u95e8\u69db\uff0c\u8ba9\u7528\u6237\u5728\u4e00\u4e2a\u5e73\u53f0\u4e0a\u5373\u53ef\u4f53\u9a8c\u00a0\u5404\u79cd\u5404\u6837\u7684\u00a0AI \u6a21\u578b\u7684\u80fd\u529b\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u5728\u4eba\u5de5\u667a\u80fd\u548c\u5927\u578b\u8bed\u8a00\u6a21\u578b (LLMs) \u9886\u57df\uff0cOllama \u6846\u67b6\u7684\u51fa\u73b0\u65e0\u7591\u5438\u5f15\u4e86\u4f17\u591a\u76ee\u5149\u3002\u8fd9\u6b3e\u5f00\u6e90\u6846\u67b6\u4e13\u6ce8\u4e8e\u7b80\u5316\u5927\u578b\u8bed\u8a00\u6a21\u578b\u5728\u672c\u5730\u7684\u90e8\u7f72\u548c\u8fd0\u884c\uff0c\u8ba9\u66f4\u591a\u5f00\u53d1\u8005\u80fd\u591f\u8f7b\u677e\u4f53\u9a8c LLM \u7684\u9b45\u529b\u3002\u7136\u800c\uff0c\u653e\u773c\u5e02\u573a\uff0cOllama \u5e76\u975e\u5b64\u519b\u594b\u6218\uff0c\u8fd8\u6709\u4e0d\u5c11\u540c&#8230;<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[34],"tags":[],"class_list":["post-21416","post","type-post","status-publish","format-standard","hentry","category-knowledge"],"_links":{"self":[{"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/posts\/21416","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=21416"}],"version-history":[{"count":0,"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/posts\/21416\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/media?parent=21416"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/categories?post=21416"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/tags?post=21416"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}