{"id":130,"date":"2026-06-27T06:41:03","date_gmt":"2026-06-27T06:41:03","guid":{"rendered":"https:\/\/leetap.com\/blog\/2026\/06\/27\/8g%e6%98%be%e5%ad%98%e9%80%86%e5%a4%a9%e4%ba%86%ef%bc%81rtx3070%e6%88%90%e5%8a%9f%e8%b7%91qwen3-6-35b%e5%a4%9a%e6%a8%a1%e6%80%81ai%e5%a4%a7%e6%a8%a1%e5%9e%8b%ef%bc%8c%e9%99%84%e5%ae%8c%e6%95%b4\/"},"modified":"2026-06-30T12:15:18","modified_gmt":"2026-06-30T12:15:18","slug":"8g%e6%98%be%e5%ad%98%e9%80%86%e5%a4%a9%e4%ba%86%ef%bc%81rtx3070%e6%88%90%e5%8a%9f%e8%b7%91qwen3-6-35b%e5%a4%9a%e6%a8%a1%e6%80%81ai%e5%a4%a7%e6%a8%a1%e5%9e%8b%ef%bc%8c%e9%99%84%e5%ae%8c%e6%95%b4","status":"publish","type":"post","link":"https:\/\/leetap.com\/blog\/2026\/06\/27\/8g%e6%98%be%e5%ad%98%e9%80%86%e5%a4%a9%e4%ba%86%ef%bc%81rtx3070%e6%88%90%e5%8a%9f%e8%b7%91qwen3-6-35b%e5%a4%9a%e6%a8%a1%e6%80%81ai%e5%a4%a7%e6%a8%a1%e5%9e%8b%ef%bc%8c%e9%99%84%e5%ae%8c%e6%95%b4\/","title":{"rendered":"8G\u663e\u5b58\u9006\u5929\u4e86\uff01RTX3070\u6210\u529f\u8dd1Qwen3.6-35B\u591a\u6a21\u6001AI\u5927\u6a21\u578b\uff0c\u9644\u5b8c\u6574\u90e8\u7f72\u6559\u7a0b"},"content":{"rendered":"\r\n

8G\u663e\u5b58\u9006\u5929\u4e86\uff01RTX3070\u6210\u529f\u8dd1Qwen3.6-35B\u591a\u6a21\u6001AI\u5927\u6a21\u578b\uff0c\u9644\u5b8c\u6574\u90e8\u7f72\u6559\u7a0b<\/h1>\r\n\r\n\r\n

\u5144\u5f1f\u4eec\uff0c\u4f60\u4eec\u6709\u6ca1\u6709\u8fd9\u79cd\u7ecf\u5386\u2014\u2014\u770b\u5230\u7f51\u4e0a\u5439\u7206\u7684AI\u5927\u6a21\u578b\uff0c\u4e00\u770b\u914d\u7f6e\u8981\u6c42\uff1a“\u6700\u4f4e\u663e\u5b58 24GB”<\/strong>\uff0c\u518d\u770b\u770b\u81ea\u5df1\u90a3\u5f20 RTX 3070\uff088GB\uff09\uff0c\u9ed8\u9ed8\u5173\u6389\u4e86\u7f51\u9875\uff1f<\/p>\r\n\r\n

\u6211\u4ee5\u524d\u4e5f\u8fd9\u6837\u3002\u76f4\u5230\u6211\u53d1\u73b0\u901a\u4e49\u5343\u95ee\u65b0\u51fa\u7684 Qwen3.6-35B<\/strong> \u5c45\u7136\u80fd\u7528 4-bit \u91cf\u5316\u585e\u8fdb 8GB \u663e\u5b58\u91cc\u8dd1\u3002\u6ca1\u9519\uff0c35B \u53c2\u6570\u7684\u5927\u6a21\u578b\uff0c8G \u5361\u80fd\u8dd1\uff0c\u800c\u4e14\u901f\u5ea6\u8fd8\u4e0d\u6162\u3002<\/p>\r\n\r\n

\u4eca\u5929\u8fd9\u7bc7\u6559\u7a0b\uff0c\u5c31\u5e26\u4f60\u7684 RTX 3070 \/ 4060 \/ 3060 12G \u628a\u8fd9\u5934”\u5927\u6a21\u578b\u602a\u517d”\u8dd1\u8d77\u6765\u3002\u5168\u7a0b\u4e0d\u7528\u4e70\u65b0\u663e\u5361\uff0c\u4e0d\u7528\u79df\u4e91\u670d\u52a1\u5668\u3002<\/p>\r\n\r\n\r\n

\u4e00\u3001\u4f60\u7684\u663e\u5361\u591f\u4e0d\u591f\uff1f<\/h2>\r\n\r\n\r\n

\u5148\u770b\u770b\u4f60\u7684\u914d\u7f6e\u591f\u4e0d\u591f\u683c\uff1a<\/p>\r\n\r\n\r\n

\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n
\u786c\u4ef6<\/th>\r\n\u6700\u4f4e\u914d\u7f6e<\/th>\r\n\u63a8\u8350\u914d\u7f6e<\/th>\r\n<\/tr>\r\n<\/thead>\r\n
\u663e\u5361<\/strong><\/td>\r\nNVIDIA 20\u7cfb\u5217+\uff088GB\u663e\u5b58\uff09<\/td>\r\nRTX 3070 \/ 4060 \/ 3060 12G<\/td>\r\n<\/tr>\r\n
\u7cfb\u7edf\u5185\u5b58<\/strong><\/td>\r\n16GB<\/td>\r\n32GB<\/td>\r\n<\/tr>\r\n
\u786c\u76d8<\/strong><\/td>\r\n20GB \u7a7a\u95f2<\/td>\r\nSSD\uff0c50GB+\u66f4\u7a33<\/td>\r\n<\/tr>\r\n
\u9a71\u52a8<\/strong><\/td>\r\nNVIDIA \u9a71\u52a8\u5df2\u88c5\u597d<\/td>\r\n\u8dd1\u4e00\u4e0b nvidia-smi \u80fd\u8ba4\u51fa\u663e\u5361<\/td>\r\n<\/tr>\r\n<\/tbody>\r\n<\/table><\/figure>\r\n\r\n\r\n

\ud83d\udc41\ufe0f \u600e\u4e48\u770b\u81ea\u5df1\u7684\u663e\u5b58\uff1f<\/strong> \u6309 Win + R<\/code> \u2192 \u8f93\u5165 cmd<\/code> \u56de\u8f66 \u2192 \u5728\u9ed1\u6846\u91cc\u8f93\u5165 nvidia-smi<\/code> \u2192 \u770b\u6700\u4e0a\u9762\u90a3\u884c\u663e\u793a “NVIDIA GeForce RTX 3070” \u540e\u9762\u7684\u6570\u5b57\uff08\u6bd4\u5982 8192MiB \u5c31\u662f 8GB\uff09\u3002<\/p>\r\n\r\n

\r\n\u26a0\ufe0f \u6ce8\u610f\uff1a<\/strong>\u5982\u679c\u4f60\u662f\u7b14\u8bb0\u672c\u7684 RTX 3050 \/ 2050\uff084GB \u663e\u5b58\uff09\uff0c\u8fd9\u7bc7\u4e0d\u9002\u5408\u4f60\u3002\u5efa\u8bae\u6362\u5c0f\u6a21\u578b Qwen3.6:7b \u6216 qwen3.6:1.5b\u3002\u522b\u5f3a\u884c\u8dd1\uff0c\u4f1a\u5361\u6b7b\u3002\r\n<\/p>\r\n\r\n\r\n

\u4e8c\u3001\u65b9\u6848A\uff1aOllama \u90e8\u7f72\uff08\u63a8\u8350\u65b0\u624b\uff0c5\u5206\u949f\u641e\u5b9a\uff09<\/h2>\r\n\r\n\r\n

\u5982\u679c\u4f60\u4e0d\u60f3\u6298\u817e\u4ee3\u7801\uff0c\u9009\u8fd9\u4e2a\u65b9\u6848\u3002\u4e00\u884c\u547d\u4ee4\u5c31\u5b8c\u4e8b\u3002<\/p>\r\n\r\n

\u7b2c1\u6b65\uff1a\u88c5 Ollama<\/h3>\r\n

\u505a\u4ec0\u4e48\uff1a<\/strong>\u5b89\u88c5\u4e00\u4e2a\u53eb Ollama \u7684\u5de5\u5177\uff0c\u5b83\u662f\u4e13\u95e8\u5e2e\u4f60\u5728\u672c\u5730\u8dd1\u5927\u6a21\u578b\u7684”\u542f\u52a8\u5668”\u3002<\/p>\r\n\r\n

\u600e\u4e48\u505a\uff1a<\/strong><\/p>\r\n

    \r\n
  1. \u6253\u5f00\u6d4f\u89c8\u5668\uff0c\u8bbf\u95ee ollama.com<\/a><\/li>\r\n
  2. \u70b9\u5927\u5927\u7684 Download \u6309\u94ae\uff0c\u9009\u62e9 Windows \u7248\u4e0b\u8f7d<\/li>\r\n
  3. \u53cc\u51fb\u5b89\u88c5\u5305 \u2192 \u4e00\u8def\u70b9\u300cInstall\u300d\u7b49\u7740\u88c5\u5b8c<\/li>\r\n<\/ol>\r\n\r\n

    \ud83d\udc41\ufe0f \u9a8c\u8bc1\uff1a<\/strong>\u6309 Win + R<\/code> \u2192 \u8f93 cmd<\/code> \u56de\u8f66 \u2192 \u5728\u9ed1\u6846\u91cc\u8f93\u5165\uff1a<\/p>\r\n

    \r\nollama --version\r\n<\/pre>\r\n
    \u770b\u5230\u7c7b\u4f3c ollama version 0.5.4<\/code> \u8fd9\u6837\u7684\u8f93\u51fa\uff0c\u8bf4\u660e\u88c5\u597d\u4e86\uff01<\/p>\r\n\r\n
    \u7b2c2\u6b65\uff1a\u4e0b\u8f7d\u6a21\u578b<\/h3>\r\n
    \u505a\u4ec0\u4e48\uff1a<\/strong>\u628a Qwen3.6-35B \u6a21\u578b\u62c9\u5230\u4f60\u7684\u7535\u8111\u4e0a\u3002\u8fd9\u91cc\u7684\u5173\u952e\u662f\u9009\u91cf\u5316\u7248<\/strong>\u2014\u2014\u4f60\u53ef\u4ee5\u7406\u89e3\u4e3a”\u538b\u7f29\u5305”\uff0c\u5728\u51e0\u4e4e\u4e0d\u964d\u8d28\u91cf\u7684\u524d\u63d0\u4e0b\u628a\u6a21\u578b\u4f53\u79ef\u538b\u5230 8GB \u80fd\u8dd1\u7684\u6c34\u5e73\u3002<\/p>\r\n\r\n
    \r\n# 4-bit \u91cf\u5316\u7248\uff08\u63a8\u8350\uff0c8G\u663e\u5b58\u8dd1\u8fd9\u4e2a\uff09\r\nollama pull qwen3.6:35b-instruct-q4_K_M\r\n\r\n# \u6216\u8005 3-bit \u7248\uff08\u66f4\u7701\u663e\u5b58\uff0c\u8d28\u91cf\u7565\u964d\u4e00\u70b9\uff09\r\nollama pull qwen3.6:35b-instruct-q3_K_M\r\n<\/pre>\r\n\r\n
    \ud83d\udc41\ufe0f \u4e0b\u8f7d\u8981\u591a\u4e45\uff1f<\/strong> \u6a21\u578b\u5927\u6982 20GB\uff0c\u770b\u4f60\u7f51\u901f\u3002\u4e00\u822c 10-30 \u5206\u949f\u3002\u53bb\u559d\u676f\u5496\u5561\u5148\u3002<\/p>\r\n\r\n
    \r\n\u26a0\ufe0f \u6ce8\u610f\uff1a<\/strong>\u4e0b\u8f7d\u5f88\u5403\u786c\u76d8\u7a7a\u95f4\u3002\u786e\u4fdd C \u76d8\u81f3\u5c11\u6709 30GB \u7a7a\u95f2\u3002\u5982\u679c C \u76d8\u4e0d\u591f\uff0c\u53ef\u4ee5\u8bbe\u7f6e Ollama \u7684\u6a21\u578b\u5b58\u5230\u5176\u4ed6\u76d8\u2014\u2014\u53f3\u952e\u6211\u7684\u7535\u8111\u2192\u5c5e\u6027\u2192\u73af\u5883\u53d8\u91cf\u2192\u65b0\u5efa\u53d8\u91cf OLLAMA_MODELS \u2192 \u8bbe\u6210 D:\\ollama_models \u4e4b\u7c7b\u7684\u8def\u5f84\uff0c\u7136\u540e\u91cd\u542f\u7535\u8111\u3002\r\n<\/p>\r\n\r\n
    \u7b2c3\u6b65\uff1a\u8fd0\u884c\u6a21\u578b<\/h3>\r\n
    \u505a\u4ec0\u4e48\uff1a<\/strong>\u542f\u52a8\u6a21\u578b\uff0c\u5f00\u59cb\u8ddf\u5b83\u804a\u5929\u3002<\/p>\r\n\r\n
    \r\nollama run qwen3.6:35b-instruct-q4_K_M\r\n<\/pre>\r\n\r\n
    \u7b49\u51e0\u79d2\u949f\u52a0\u8f7d\u5b8c\u6210\u540e\uff0c\u4f60\u5c31\u53ef\u4ee5\u76f4\u63a5\u6253\u5b57\u8ddf\u5b83\u804a\u5929\u4e86\u3002<\/p>\r\n\r\n
    \ud83d\udc41\ufe0f \u9a8c\u8bc1\uff1a<\/strong>\u8f93\u5165”\u4f60\u597d\uff0c\u4ecb\u7ecd\u4e00\u4e0b\u4f60\u81ea\u5df1”\uff0c\u770b\u5b83\u80fd\u4e0d\u80fd\u6b63\u5e38\u56de\u590d\u3002\u540c\u65f6\u6253\u5f00\u4efb\u52a1\u7ba1\u7406\u5668 \u2192 \u6027\u80fd \u2192 GPU\uff0c\u770b\u770b\u663e\u5b58\u5360\u7528\u662f\u4e0d\u662f\u5728 6-7GB \u5de6\u53f3\u3002<\/p>\r\n\r\n
    \u591a\u6a21\u6001\uff08\u770b\u56fe\u529f\u80fd\uff09\uff1a<\/strong>\u8fd9\u6a21\u578b\u80fd\u770b\u56fe\u3002\u968f\u4fbf\u62d6\u4e00\u5f20\u56fe\u7247\u5230\u7ec8\u7aef\u7a97\u53e3\u91cc\uff0c\u518d\u8f93\u5165\uff1a<\/p>\r\n
    \r\n\u63cf\u8ff0\u8fd9\u5f20\u56fe\u7247\u91cc\u6709\u4ec0\u4e48\r\n<\/pre>\r\n\r\n
    AI \u4f1a\u5206\u6790\u56fe\u7247\u5185\u5bb9\u56de\u7b54\u4f60\u3002<\/p>\r\n\r\n\"RTX3070\u6210\u529f\u8fd0\u884cQwen3.6-35B\u7684\u622a\u56fe\"\r\n\r\n
    \u7b2c4\u6b65\uff1a\u901a\u8fc7 API \u8c03\u7528\uff08\u8fdb\u9636\uff09<\/h3>\r\n
    \u5982\u679c\u4f60\u60f3\u628a Qwen \u96c6\u6210\u5230\u5176\u4ed6\u5de5\u5177\u91cc\uff08\u6bd4\u5982\u63a5\u5165 Hermes Agent\uff09\uff0c\u53ef\u4ee5\u7528 API \u8c03\uff1a<\/p>\r\n
    \r\ncurl http:\/\/localhost:11434\/api\/generate -d '{\r\n  \"model\": \"qwen3.6:35b-instruct-q4_K_M\",\r\n  \"prompt\": \"\u7528100\u5b57\u89e3\u91ca\u5927\u8bed\u8a00\u6a21\u578b\u7684\u5de5\u4f5c\u539f\u7406\",\r\n  \"stream\": false\r\n}'\r\n<\/pre>\r\n\r\n
    Ollama \u9ed8\u8ba4\u5728 localhost:11434<\/code> \u4e0a\u8dd1\uff0c\u652f\u6301 OpenAI \u517c\u5bb9\u7684 API \u683c\u5f0f\uff0c\u5f88\u591a\u7b2c\u4e09\u65b9\u5de5\u5177\u90fd\u80fd\u76f4\u63a5\u63a5\u3002<\/p>\r\n\r\n\r\n
    \u4e09\u3001\u65b9\u6848B\uff1allama.cpp \u90e8\u7f72\uff08\u8fdb\u9636\u6298\u817e\u515a\uff09<\/h2>\r\n\r\n\r\n
    \u5982\u679c\u4f60\u7684\u663e\u5361\u4e0d\u662f NVIDIA\uff08\u6bd4\u5982 AMD \/ Intel\uff09\uff0c\u6216\u8005\u4f60\u60f3\u628a\u663e\u5b58\u5229\u7528\u69a8\u5230\u6781\u81f4\uff0c\u7528\u8fd9\u4e2a\u65b9\u6848\u3002<\/p>\r\n\r\n
    \u7b2c1\u6b65\uff1a\u7f16\u7ece llama.cpp<\/h3>\r\n
    \r\ngit clone https:\/\/github.com\/ggerganov\/llama.cpp\r\ncd llama.cpp\r\ncmake -B build -DGGML_CUDA=ON\r\ncmake --build build --config Release -j$(nproc)\r\n<\/pre>\r\n\r\n
    \u7b2c2\u6b65\uff1a\u4e0b\u8f7d GGUF \u6a21\u578b<\/h3>\r\n
    \r\npip install huggingface-hub\r\nhuggingface-cli download Qwen\/Qwen3.6-35B-Instruct-GGUF \\\r\n  qwen3.6-35b-instruct-q4_k_m.gguf --local-dir .\/models\/\r\n<\/pre>\r\n\r\n
    \u7b2c3\u6b65\uff1a\u8fd0\u884c\uff08\u7cbe\u8c03 -ngl\uff09<\/h3>\r\n
    -ngl<\/strong> \u7684\u610f\u601d\u662f”\u628a\u591a\u5c11\u5c42\u6a21\u578b\u653e\u5230 GPU \u4e0a\u8dd1”\u3002\u6570\u5b57\u8d8a\u5927\uff0cGPU \u8dd1\u5f97\u8d8a\u591a\u8d8a\u5feb\uff0c\u4f46\u663e\u5b58\u5360\u7528\u4e5f\u8d8a\u9ad8\u30028G \u663e\u5b58\u4e00\u822c\u8bbe 20-24\uff1a<\/p>\r\n
    \r\n# -ngl 20 \u8868\u793a\u628a\u524d20\u5c42\u653eGPU\uff0c\u5269\u4e0b\u653e\u5185\u5b58\r\n.\/build\/bin\/llama-cli \\\r\n  -m .\/models\/qwen3.6-35b-instruct-q4_k_m.gguf \\\r\n  -ngl 22 -c 4096 --temp 0.7 \\\r\n  --prompt \"\u4f60\u597d\uff0c\u8bf7\u4ecb\u7ecd\u4e00\u4e0b\u4f60\u81ea\u5df1\"\r\n<\/pre>\r\n\r\n
    \ud83d\udc41\ufe0f \u8c03\u4f18\u6280\u5de7\uff1a<\/strong>\u8dd1\u7684\u65f6\u5019\u5f00\u7740\u4efb\u52a1\u7ba1\u7406\u5668\u770b\u663e\u5b58\u3002\u6ca1\u5230 7.5GB \u2192 \u628a -ngl \u52a0 2\uff1b\u8d85\u8fc7\u4e86 7.8GB \u2192 \u628a -ngl \u51cf 2\u3002\u63a7\u5236\u5728 7.5GB \u4ee5\u5185\u6700\u7a33\u3002<\/p>\r\n\r\n\"\u663e\u5b58\u5360\u7528\u60c5\u51b5\u622a\u56fe\"\r\n\r\n\r\n
    \u56db\u30018G \u663e\u5b58\u4f18\u5316\u4e09\u677f\u65a7<\/h2>\r\n\r\n\r\n
    8GB \u663e\u5b58\u8dd1 35B \u6a21\u578b\u5c5e\u4e8e”\u6781\u9650\u64cd\u4f5c”\u3002\u4e0b\u9762\u4e09\u62db\u5e2e\u4f60\u8dd1\u5f97\u66f4\u987a\uff1a<\/p>\r\n\r\n
    1. \u9009\u5bf9\u91cf\u5316\u7b49\u7ea7<\/h3>\r\n\r\n
    \r\n\r\n\r\n\r\n\r\n\r\n\r\n
    \u91cf\u5316\u7b49\u7ea7<\/th>\r\n\u663e\u5b58\u5360\u7528<\/th>\r\n\u8d28\u91cf<\/th>\r\n\u63a8\u8350\uff1f<\/th>\r\n<\/tr>\r\n<\/thead>\r\n
    Q4_K_M<\/td>\r\n~6.5GB<\/td>\r\n\u4f18\u79c0\uff0c\u51e0\u4e4e\u65e0\u635f<\/td>\r\n\u2705 \u9996\u9009<\/td>\r\n<\/tr>\r\n
    Q3_K_M<\/td>\r\n~5.0GB<\/td>\r\n\u826f\u597d\uff0c\u7a0d\u6709\u635f\u5931<\/td>\r\n\ud83d\udfe1 \u5907\u9009<\/td>\r\n<\/tr>\r\n
    Q2_K<\/td>\r\n~3.8GB<\/td>\r\n\u53ca\u683c\uff0c\u80fd\u8dd1\u4f46\u53d8\u7b28<\/td>\r\n\u274c \u4e0d\u63a8\u8350<\/td>\r\n<\/tr>\r\n<\/tbody>\r\n<\/table><\/figure>\r\n\r\n\r\n
    2. \u7cbe\u8c03 -ngl \u53c2\u6570<\/h3>\r\n
    \u7528 llama.cpp \u7684\u60c5\u51b5\u4e0b\uff0c-ngl 20 \u8d77\u6b65\uff0c\u6162\u6162\u5f80\u4e0a\u52a0<\/strong>\uff0c\u627e\u5230\u4f60\u7684\u663e\u5361\u6781\u9650\uff1a<\/p>\r\n
    \r\n# \u5148\u8bd5 20\uff0c\u663e\u5b58\u6709\u4f59\u52a0\u5230 24\uff0c\u6ea2\u51fa\u4e86\u964d\u5230 16\r\n-ngl 20\r\n\r\n# \u7528 nvidia-smi \u76ef\u7740\u663e\u5b58\uff0c\u522b\u8d85 7.5GB\r\n<\/pre>\r\n\r\n
    3. \u7cfb\u7edf\u5185\u5b58\u8981\u591f<\/h3>\r\n
    \u90e8\u5206\u5c42\u4f1a\u5378\u8f7d\u5230\u7cfb\u7edf\u5185\u5b58\u4e0a\u8dd1\uff0c\u6240\u4ee5\u7cfb\u7edf\u5185\u5b58\u5efa\u8bae 16GB \u8d77\u6b65\uff0c32GB \u66f4\u7a33<\/strong>\u3002\u5982\u679c\u4f60\u53ea\u6709 8GB \u7cfb\u7edf\u5185\u5b58\uff0c\u8dd1\u5927\u6a21\u578b\u65f6\u5173\u6389\u6d4f\u89c8\u5668\u548c\u5fae\u4fe1\u3002<\/p>\r\n\r\n\"\u4f18\u5316\u524d\u540e\u5bf9\u6bd4\u56fe\"\r\n\r\n\r\n
    \u4e94\u3001\u5e38\u89c1\u95ee\u9898<\/h2>\r\n\r\n\r\n
    Q\uff1a\u8dd1\u5230\u4e00\u534a\u663e\u5b58\u6ea2\u51fa\u4e86\u600e\u4e48\u529e\uff1f<\/h3>\r\n
    \u6362\u66f4\u4f4e\u91cf\u5316\u7684\u7248\u672c\u3002\u4ece Q4_K_M \u964d\u5230 Q3_K_M\uff0c\u663e\u5b58\u5360\u7528\u80fd\u5c11 1.5GB\u3002\u6216\u8005\u8c03\u5c0f -ngl \u7684\u503c\uff0c\u8ba9 CPU \u591a\u5206\u62c5\u4e00\u4e9b\u3002<\/p>\r\n\r\n
    Q\uff1a\u8dd1\u5f97\u7279\u522b\u6162\uff0c\u4e00\u79d2\u4e00\u4e2a\u5b57<\/h3>\r\n
    \u68c0\u67e5\u4e09\u4ef6\u4e8b\uff1a\u2460 \u786e\u8ba4 CUDA \u88c5\u597d\u4e86\uff08\u8dd1 nvidia-smi \u770b\u6709\u6ca1\u6709 CUDA \u7248\u672c\u53f7\uff09\u2461 \u52a0\u5927 -ngl \u503c \u2462 \u628a\u4e0a\u4e0b\u6587\u957f\u5ea6 -c \u4ece 4096 \u964d\u5230 2048\u3002\u5982\u679c\u8fd8\u6162\uff0c\u90a3\u5c31\u662f\u4f60\u663e\u5361\u672c\u8eab\u7684\u7b97\u529b\u74f6\u9888\uff0c\u53ea\u80fd\u6362\u5361\u3002<\/p>\r\n\r\n
    Q\uff1a\u4e0b\u8f7d\u6a21\u578b\u592a\u6162\u4e86\uff0c\u51e0\u5341 KB\/s<\/h3>\r\n
    HuggingFace \u56fd\u5185\u4e0b\u8f7d\u786e\u5b9e\u6162\u3002\u8bbe\u955c\u50cf\u7ad9\uff1a<\/p>\r\n
    \r\n# Linux\/Mac \u4e34\u65f6\u8bbe\u955c\u50cf\r\nexport HF_ENDPOINT=https:\/\/hf-mirror.com\r\n\r\n# \u7136\u540e\u91cd\u65b0\u6267\u884c\u4e0b\u8f7d\u547d\u4ee4\r\n<\/pre>\r\n
    \u6216\u8005\u51cc\u6668\u4e0b\u8f7d\uff0c\u901f\u5ea6\u4f1a\u5feb\u5f88\u591a\u3002<\/p>\r\n\r\n
    Q\uff1aOllama \u4e0b\u8f7d\u5b8c\u8dd1\u4e0d\u8d77\u6765\u600e\u4e48\u529e\uff1f<\/h3>\r\n
    \u5148\u786e\u8ba4 Ollama \u5728\u8fd0\u884c\u2014\u2014\u53f3\u4e0b\u89d2\u4efb\u52a1\u680f\u6709\u6ca1\u6709\u7f8a\u9a7c\u56fe\u6807\uff1f\u6ca1\u6709\u7684\u8bdd\u53bb\u5f00\u59cb\u83dc\u5355\u641c Ollama \u624b\u52a8\u6253\u5f00\u3002\u8fd8\u4e0d\u884c\u5c31\u91cd\u542f\u7535\u8111\u4e00\u6b21\u3002<\/p>\r\n\r\n
    Q\uff1a\u6211\u7684\u663e\u5361\u662f AMD \u7684\u80fd\u8dd1\u5417\uff1f<\/h3>\r\n
    \u7528 Ollama \u7684\u8bdd AMD \u663e\u5361\u4e5f\u80fd\u8dd1\uff08Ollama \u6700\u65b0\u7248\u652f\u6301 ROCm\uff09\u3002\u7528 llama.cpp \u7684\u8bdd\u9700\u8981\u7f16\u8bd1\u7684\u65f6\u5019\u52a0 -DGGML_HIPBLAS=ON<\/code>\u3002\u4f46\u4f53\u9a8c\u4e0d\u5982 NVIDIA \u7a33\u5b9a\uff0c\u5efa\u8bae\u5148\u7528\u65b9\u6848A\u7684 Ollama\u3002<\/p>\r\n\r\n\r\n
    \u516d\u3001\u603b\u7ed3<\/h2>\r\n\r\n\r\n
    RTX3070\uff088GB\uff09\u8dd1 Qwen3.6-35B \u5b8c\u5168\u53ef\u884c\u3002<\/strong>\u8bb0\u4f4f\u4e09\u6761\uff1a<\/p>\r\n
      \r\n
    1. \u9009\u5bf9\u91cf\u5316<\/strong> \u2192 Q4_K_M\uff0c\u663e\u5b58\u53ea\u5403 6.5GB<\/li>\r\n
    2. \u8c03\u597d\u5206\u5c42<\/strong> \u2192 -ngl 20-24\uff0cGPU\/CPU \u6df7\u5408\u8dd1<\/li>\r\n
    3. \u5185\u5b58\u7ba1\u591f<\/strong> \u2192 \u7cfb\u7edf\u5185\u5b58\u81f3\u5c11 16GB<\/li>\r\n<\/ol>\r\n\r\n
      \u65b0\u624b\u7528\u65b9\u6848A\uff08Ollama\uff09\uff0c\u4e00\u884c\u547d\u4ee4\u641e\u5b9a\u3002\u6298\u817e\u515a\u7528\u65b9\u6848B\uff08llama.cpp\uff09\uff0c\u663e\u5b58\u5229\u7528\u66f4\u6781\u81f4\u3002<\/p>\r\n\r\n
      \r\n\ud83d\udca1 \u7ed9\u65b0\u624b\u7684\u8bdd\uff1a<\/strong>\u522b\u4e00\u4e0a\u6765\u5c31\u641e llama.cpp\u3002\u5148\u88c5 Ollama\uff0c\u8dd1\u8d77\u6765\u518d\u8bf4\u3002\u7b49\u4f60\u89c9\u5f97 Ollama \u4e0d\u591f\u6298\u817e\u4e86\uff0c\u518d\u6362\u65b9\u6848B\u4e5f\u4e0d\u8fdf\u3002\u4f60\u90a38G\u5361\u771f\u7684\u522b\u5403\u7070\u4e86\uff0c\u88c5\u4e0a\u8bd5\u8bd5\uff01<\/p>\r\n","protected":false},"excerpt":{"rendered":"
      8G\u663e\u5b58\u9006\u5929\u4e86\uff01RTX3070\u6210\u529f\u8dd1Qwen3.6-35B\u591a\u6a21\u6001AI\u5927\u6a21\u578b\uff0c\u9644\u5b8c\u6574\u90e8\u7f72\u6559\u7a0b \u5144\u5f1f\u4eec\uff0c\u4f60\u4eec\u6709\u6ca1\u6709… <\/p>\n","protected":false},"author":0,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"neve_meta_sidebar":"","neve_meta_container":"","neve_meta_enable_content_width":"","neve_meta_content_width":0,"neve_meta_title_alignment":"","neve_meta_author_avatar":"","neve_post_elements_order":"","neve_meta_disable_header":"","neve_meta_disable_footer":"","neve_meta_disable_title":"","footnotes":""},"categories":[39],"tags":[],"class_list":["post-130","post","type-post","status-publish","format-standard","hentry","category-qwen"],"_links":{"self":[{"href":"https:\/\/leetap.com\/blog\/wp-json\/wp\/v2\/posts\/130","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/leetap.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/leetap.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"replies":[{"embeddable":true,"href":"https:\/\/leetap.com\/blog\/wp-json\/wp\/v2\/comments?post=130"}],"version-history":[{"count":4,"href":"https:\/\/leetap.com\/blog\/wp-json\/wp\/v2\/posts\/130\/revisions"}],"predecessor-version":[{"id":153,"href":"https:\/\/leetap.com\/blog\/wp-json\/wp\/v2\/posts\/130\/revisions\/153"}],"wp:attachment":[{"href":"https:\/\/leetap.com\/blog\/wp-json\/wp\/v2\/media?parent=130"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/leetap.com\/blog\/wp-json\/wp\/v2\/categories?post=130"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/leetap.com\/blog\/wp-json\/wp\/v2\/tags?post=130"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}