{"id":81697,"date":"2026-07-25T03:42:44","date_gmt":"2026-07-24T19:42:44","guid":{"rendered":"https:\/\/www.wsisp.com\/helps\/81697.html"},"modified":"2026-07-25T03:42:44","modified_gmt":"2026-07-24T19:42:44","slug":"youtu-vl-4b-instruct%e9%83%a8%e7%bd%b2%e6%a1%88%e4%be%8b%ef%bc%9a%e8%be%b9%e7%bc%98%e6%9c%8d%e5%8a%a1%e5%99%a8jetson-agx-orin%e9%83%a8%e7%bd%b2%e5%8f%af%e8%a1%8c%e6%80%a7%e9%aa%8c%e8%af%81","status":"publish","type":"post","link":"https:\/\/www.wsisp.com\/helps\/81697.html","title":{"rendered":"Youtu-VL-4B-Instruct\u90e8\u7f72\u6848\u4f8b\uff1a\u8fb9\u7f18\u670d\u52a1\u5668Jetson AGX Orin\u90e8\u7f72\u53ef\u884c\u6027\u9a8c\u8bc1"},"content":{"rendered":"

Youtu-VL-4B-Instruct\u90e8\u7f72\u6848\u4f8b：\u8fb9\u7f18\u670d\u52a1\u5668Jetson AGX Orin\u90e8\u7f72\u53ef\u884c\u6027\u9a8c\u8bc1<\/h2>\n

1. \u5f15\u8a00<\/h3>\n

\u6700\u8fd1\u5728\u6298\u817e\u8fb9\u7f18AI\u90e8\u7f72，\u624b\u5934\u6b63\u597d\u6709\u4e00\u53f0NVIDIA Jetson AGX Orin，64GB\u5185\u5b58，GPU\u7b97\u529b\u770b\u7740\u633a\u552c\u4eba\u3002\u6211\u5c31\u5728\u60f3，\u73b0\u5728\u90a3\u4e9b\u52a8\u8f84\u51e0\u5341\u4ebf\u53c2\u6570\u7684\u591a\u6a21\u6001\u5927\u6a21\u578b，\u80fd\u4e0d\u80fd\u5728\u8fd9\u53f0\u201c\u8fb9\u7f18\u8ba1\u7b97\u76d2\u5b50\u201d\u4e0a\u8dd1\u8d77\u6765？\u7279\u522b\u662f\u90a3\u79cd\u53f7\u79f0\u201c\u8f7b\u91cf\u7ea7\u201d\u7684\u6a21\u578b\u3002<\/p>\n

\u817e\u8baf\u4f18\u56fe\u5b9e\u9a8c\u5ba4\u5f00\u6e90\u7684Youtu-VL-4B-Instruct\u8fdb\u5165\u4e86\u6211\u7684\u89c6\u7ebf\u30024B\u53c2\u6570，\u652f\u6301\u56fe\u7247\u7406\u89e3\u3001OCR\u3001\u89c6\u89c9\u95ee\u7b54，\u8fd8\u63d0\u4f9b\u4e86GGUF\u91cf\u5316\u7248\u672c\u3002\u5b98\u65b9\u63a8\u8350\u914d\u7f6e\u662fRTX 4090（24GB\u663e\u5b58）\u8d77\u6b65，\u8fd9\u8ba9\u6211\u5fc3\u91cc\u76f4\u6253\u9f13\u2014\u2014Jetson AGX Orin\u7684GPU\u5185\u5b58\u53ea\u670932GB（\u5171\u4eab\u5185\u5b58\u67b6\u6784），\u80fd\u625b\u5f97\u4f4f\u5417？<\/p>\n

\u8fd9\u7bc7\u6587\u7ae0，\u5c31\u662f\u4e00\u6b21\u771f\u5b9e\u7684\u201c\u8e29\u5751\u201d\u8bb0\u5f55\u3002\u6211\u4f1a\u5e26\u4f60\u4e00\u6b65\u6b65\u9a8c\u8bc1，\u8fd9\u4e2a4B\u53c2\u6570\u7684\u591a\u6a21\u6001\u89c6\u89c9\u8bed\u8a00\u6a21\u578b，\u5230\u5e95\u80fd\u4e0d\u80fd\u5728Jetson AGX Orin\u4e0a\u987a\u5229\u90e8\u7f72\u548c\u8fd0\u884c\u3002\u6574\u4e2a\u8fc7\u7a0b\u6d89\u53ca\u73af\u5883\u51c6\u5907\u3001\u955c\u50cf\u90e8\u7f72\u3001\u6027\u80fd\u6d4b\u8bd5\u548c\u5b9e\u9645\u6548\u679c\u9a8c\u8bc1，\u5982\u679c\u4f60\u4e5f\u5728\u8003\u8651\u5728\u8fb9\u7f18\u8bbe\u5907\u4e0a\u90e8\u7f72\u7c7b\u4f3c\u6a21\u578b，\u8fd9\u7bc7\u5185\u5bb9\u5e94\u8be5\u80fd\u7ed9\u4f60\u4e0d\u5c11\u53c2\u8003\u3002<\/p>\n

2. \u73af\u5883\u51c6\u5907\u4e0e\u786c\u4ef6\u786e\u8ba4<\/h3>\n

2.1 Jetson AGX Orin\u786c\u4ef6\u89c4\u683c<\/h4>\n

\u5728\u5f00\u59cb\u4e4b\u524d，\u6211\u4eec\u5148\u660e\u786e\u4e00\u4e0b\u8fd9\u53f0\u8bbe\u5907\u7684\u201c\u5bb6\u5e95\u201d\u3002\u6211\u7528\u7684\u8fd9\u53f0Jetson AGX Orin 64GB\u7248\u672c，\u5177\u4f53\u914d\u7f6e\u5982\u4e0b：<\/p>\n\n\u7ec4\u4ef6\u89c4\u683c\u5907\u6ce8<\/tr>\n\n\n\n\n\n\n\n
GPU<\/td>\nNVIDIA Ampere\u67b6\u6784，2048\u4e2aCUDA\u6838\u5fc3<\/td>\n\u5171\u4eab\u5185\u5b58\u67b6\u6784，\u65e0\u72ec\u7acb\u663e\u5b58<\/td>\n<\/tr>\n
\u5185\u5b58<\/td>\n64GB LPDDR5<\/td>\nGPU\u548cCPU\u5171\u4eab\u8fd964GB<\/td>\n<\/tr>\n
CPU<\/td>\n12\u6838ARM Cortex-A78AE<\/td>\n\u4e3b\u98912.2GHz<\/td>\n<\/tr>\n
\u5b58\u50a8<\/td>\n64GB eMMC 5.1<\/td>\n\u6211\u989d\u5916\u52a0\u4e861TB NVMe SSD<\/td>\n<\/tr>\n
\u529f\u8017<\/td>\n15W-60W\u53ef\u8c03<\/td>\n\u6d4b\u8bd5\u65f6\u8bbe\u7f6e\u4e3aMAXN\u6a21\u5f0f（60W）<\/td>\n<\/tr>\n
JetPack\u7248\u672c<\/td>\n6.0<\/td>\nL4T 36.3.0<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

\u5173\u952e\u70b9：Jetson\u5e73\u53f0\u662f\u5171\u4eab\u5185\u5b58\u67b6\u6784，\u8fd9\u610f\u5473\u7740GPU\u6ca1\u6709\u72ec\u7acb\u7684\u663e\u5b58，\u800c\u662f\u548cCPU\u5171\u4eab\u540c\u4e00\u5757\u7269\u7406\u5185\u5b58\u3002\u5b98\u65b9\u63a8\u8350RTX 4090（24GB\u72ec\u7acb\u663e\u5b58），\u800cJetson AGX Orin\u867d\u7136\u670964GB\u603b\u5185\u5b58，\u4f46GPU\u80fd\u7528\u7684\u90e8\u5206\u9700\u8981\u548c\u7cfb\u7edf\u3001\u5176\u4ed6\u8fdb\u7a0b\u5171\u4eab\u3002<\/p>\n

2.2 \u7cfb\u7edf\u73af\u5883\u68c0\u67e5<\/h4>\n

\u9996\u5148\u767b\u5f55\u5230Jetson AGX Orin，\u68c0\u67e5\u57fa\u7840\u73af\u5883：<\/p>\n

# \u67e5\u770bJetPack\u7248\u672c
\ncat \/etc\/nv_tegra_release<\/p>\n

# \u67e5\u770bCUDA\u7248\u672c
\nnvcc –version<\/p>\n

# \u67e5\u770b\u5185\u5b58\u4f7f\u7528\u60c5\u51b5
\nfree -h<\/p>\n

# \u67e5\u770bGPU\u4fe1\u606f
\nsudo tegrastats<\/p>\n

\u6211\u7684\u73af\u5883\u8f93\u51fa\u5982\u4e0b：<\/p>\n

    \n
  • JetPack 6.0 (L4T 36.3.0)<\/li>\n
  • CUDA 12.2<\/li>\n
  • \u7a7a\u95f2\u5185\u5b58：\u7ea658GB（\u521a\u5f00\u673a\u72b6\u6001）<\/li>\n
  • GPU\u9891\u7387：1.3GHz（MAXN\u6a21\u5f0f）<\/li>\n<\/ul>\n

    2.3 \u78c1\u76d8\u7a7a\u95f4\u51c6\u5907<\/h4>\n

    Youtu-VL-4B-Instruct\u7684GGUF\u6a21\u578b\u6587\u4ef6\u5927\u7ea66GB，\u52a0\u4e0aPython\u73af\u5883\u3001\u4f9d\u8d56\u5e93\u7b49，\u9700\u8981\u81f3\u5c1120GB\u7a7a\u95f4\u3002Jetson AGX Orin\u81ea\u5e26\u768464GB eMMC\u80af\u5b9a\u4e0d\u591f\u7528，\u6211\u63d0\u524d\u5b89\u88c5\u597d\u4e861TB NVMe SSD，\u5e76\u6302\u8f7d\u5230\u4e86\/data\u76ee\u5f55\u3002<\/p>\n

    # \u67e5\u770b\u78c1\u76d8\u7a7a\u95f4
    \ndf -h<\/p>\n

    # \u6211\u7684\u6302\u8f7d\u60c5\u51b5
    \n# \/dev\/nvme0n1p1 916G 28G 842G 4% \/data<\/p>\n

    \u5efa\u8bae\u5728NVMe SSD\u4e0a\u64cd\u4f5c，eMMC\u7684\u8bfb\u5199\u901f\u5ea6\u4f1a\u6210\u4e3a\u74f6\u9888\u3002<\/p>\n

    3. \u90e8\u7f72\u8fc7\u7a0b\u4e0e\u95ee\u9898\u89e3\u51b3<\/h3>\n

    3.1 \u83b7\u53d6CSDN\u661f\u56fe\u955c\u50cf<\/h4>\n

    CSDN\u661f\u56fe\u955c\u50cf\u5e7f\u573a\u63d0\u4f9b\u4e86\u9884\u914d\u7f6e\u7684Youtu-VL-4B-Instruct\u955c\u50cf，\u8fd9\u5927\u5927\u7b80\u5316\u4e86\u90e8\u7f72\u6d41\u7a0b\u3002\u955c\u50cf\u5df2\u7ecf\u5305\u542b\u4e86：<\/p>\n

      \n
    • \u6a21\u578b\u6587\u4ef6（GGUF\u91cf\u5316\u7248）<\/li>\n
    • llama.cpp\u63a8\u7406\u5f15\u64ce<\/li>\n
    • Gradio WebUI\u754c\u9762<\/li>\n
    • OpenAI\u517c\u5bb9API\u670d\u52a1<\/li>\n
    • Supervisor\u8fdb\u7a0b\u7ba1\u7406<\/li>\n<\/ul>\n

      \u901a\u8fc7CSDN\u661f\u56fe\u5e73\u53f0\u4e00\u952e\u90e8\u7f72\u540e，\u670d\u52a1\u9ed8\u8ba4\u57287860\u7aef\u53e3\u542f\u52a8\u3002\u4f46\u8fd9\u91cc\u6709\u4e2a\u95ee\u9898：\u955c\u50cf\u9ed8\u8ba4\u914d\u7f6e\u662f\u9488\u5bf9x86\u670d\u52a1\u5668\u4f18\u5316\u7684，\u5728ARM\u67b6\u6784\u7684Jetson\u4e0a\u9700\u8981\u4e00\u4e9b\u8c03\u6574\u3002<\/p>\n

      3.2 ARM\u67b6\u6784\u9002\u914d\u8c03\u6574<\/h4>\n

      \u7b2c\u4e00\u4e2a\u5751\u6765\u4e86\u2014\u2014llama.cpp\u7684\u9884\u7f16\u8bd1\u4e8c\u8fdb\u5236\u6587\u4ef6\u662f\u9488\u5bf9x86_64\u7684，\u5728ARM64\u4e0a\u65e0\u6cd5\u76f4\u63a5\u8fd0\u884c\u3002\u9700\u8981\u91cd\u65b0\u7f16\u8bd1\u3002<\/p>\n

      # \u8fdb\u5165llama.cpp\u76ee\u5f55
      \ncd \/opt\/youtu-vl\/llama.cpp<\/p>\n

      # \u6e05\u7406\u4e4b\u524d\u7684\u6784\u5efa
      \nmake clean<\/p>\n

      # \u9488\u5bf9Jetson\u5e73\u53f0\u91cd\u65b0\u7f16\u8bd1
      \nmake -j$(nproc) LLAMA_CUBLAS=1<\/p>\n

      # \u9a8c\u8bc1\u7f16\u8bd1\u662f\u5426\u6210\u529f
      \n.\/main –help<\/p>\n

      \u7f16\u8bd1\u8fc7\u7a0b\u5927\u7ea6\u9700\u898115-20\u5206\u949f，\u53d6\u51b3\u4e8e\u4f60\u7684\u7f51\u7edc\u901f\u5ea6（\u9700\u8981\u4e0b\u8f7d\u4e00\u4e9b\u4f9d\u8d56）\u3002\u5982\u679c\u9047\u5230CUDA\u76f8\u5173\u9519\u8bef，\u53ef\u80fd\u9700\u8981\u68c0\u67e5CUDA\u73af\u5883\u53d8\u91cf：<\/p>\n

      # \u8bbe\u7f6eCUDA\u8def\u5f84
      \nexport CUDA_PATH=\/usr\/local\/cuda
      \nexport PATH=$CUDA_PATH\/bin:$PATH
      \nexport LD_LIBRARY_PATH=$CUDA_PATH\/lib64:$LD_LIBRARY_PATH<\/p>\n

      3.3 \u5185\u5b58\u4f18\u5316\u914d\u7f6e<\/h4>\n

      \u7b2c\u4e8c\u4e2a\u6311\u6218\u662f\u5185\u5b58\u3002\u867d\u7136Jetson AGX Orin\u670964GB\u5185\u5b58，\u4f46\u6a21\u578b\u52a0\u8f7d\u540e，GPU\u63a8\u7406\u9700\u8981\u5927\u91cf\u8fde\u7eed\u5185\u5b58\u3002\u9700\u8981\u8c03\u6574llama.cpp\u7684\u914d\u7f6e\u53c2\u6570\u3002<\/p>\n

      \u4fee\u6539\u542f\u52a8\u811a\u672c\/usr\/local\/bin\/start-youtu-vl-4b-instruct-gguf-service.sh：<\/p>\n

      #!\/bin\/bash
      \nsource \/opt\/youtu-vl\/venv\/bin\/activate<\/p>\n

      echo "Starting Youtu-VL-4B-Instruct-GGUF service on Jetson AGX Orin…"<\/p>\n

      # \u5173\u952e\u53c2\u6570\u8c03\u6574
      \n# -ngl 9999: \u5c3d\u53ef\u80fd\u591a\u7684\u5c42\u653e\u5230GPU
      \n# -c 4096: \u4e0a\u4e0b\u6587\u957f\u5ea6
      \n# -b 512: \u6279\u5904\u7406\u5927\u5c0f（\u51cf\u5c0f\u4ee5\u8282\u7701\u5185\u5b58）
      \n# –mlock: \u9501\u5b9a\u6a21\u578b\u5728\u5185\u5b58\u4e2d，\u907f\u514d\u4ea4\u6362
      \n# –no-mmap: \u4e0d\u4f7f\u7528\u5185\u5b58\u6620\u5c04，\u51cf\u5c11\u5185\u5b58\u788e\u7247<\/p>\n

      exec python \/opt\/youtu-vl\/server.py \\\\
      \n –host 0.0.0.0 \\\\
      \n –port 7860 \\\\
      \n –model \/opt\/youtu-vl\/models\/youtu-vl-4b-instruct.Q4_K_M.gguf \\\\
      \n –n_gpu_layers 9999 \\\\
      \n –ctx_size 4096 \\\\
      \n –batch_size 512 \\\\
      \n –mlock \\\\
      \n –no-mmap \\\\
      \n –verbose<\/p>\n

      \u53c2\u6570\u89e3\u91ca：<\/p>\n

        \n
      • -n_gpu_layers 9999：\u544a\u8bc9llama.cpp\u628a\u6240\u6709\u80fd\u653eGPU\u7684\u5c42\u90fd\u653e\u4e0a\u53bb<\/li>\n
      • -c 4096：\u4e0a\u4e0b\u6587\u957f\u5ea6，4K\u5bf9\u4e8e\u5927\u591a\u6570\u89c6\u89c9\u95ee\u7b54\u4efb\u52a1\u8db3\u591f<\/li>\n
      • -b 512：\u6279\u5904\u7406\u5927\u5c0f，Jetson\u4e0a\u9700\u8981\u8c03\u5c0f<\/li>\n
      • –mlock：\u9632\u6b62\u6a21\u578b\u88ab\u4ea4\u6362\u5230\u78c1\u76d8，\u4fdd\u6301\u54cd\u5e94\u901f\u5ea6<\/li>\n
      • –no-mmap：\u907f\u514d\u5185\u5b58\u6620\u5c04\u5e26\u6765\u7684\u788e\u7247\u95ee\u9898<\/li>\n<\/ul>\n

        3.4 \u670d\u52a1\u542f\u52a8\u4e0e\u9a8c\u8bc1<\/h4>\n

        \u914d\u7f6e\u5b8c\u6210\u540e，\u91cd\u542f\u670d\u52a1：<\/p>\n

        # \u91cd\u542f\u670d\u52a1
        \nsudo supervisorctl restart youtu-vl-4b-instruct-gguf<\/p>\n

        # \u67e5\u770b\u670d\u52a1\u72b6\u6001
        \nsudo supervisorctl status<\/p>\n

        # \u67e5\u770b\u65e5\u5fd7
        \ntail -f \/var\/log\/youtu-vl-4b-instruct-gguf.log<\/p>\n

        \u5982\u679c\u4e00\u5207\u6b63\u5e38，\u4f60\u5e94\u8be5\u80fd\u770b\u5230\u7c7b\u4f3c\u8fd9\u6837\u7684\u65e5\u5fd7：<\/p>\n

        Loading model from \/opt\/youtu-vl\/models\/youtu-vl-4b-instruct.Q4_K_M.gguf
        \nllama_model_loader: loaded meta data with 20 key-value pairs and 291 tensors (291.00 MB)
        \nllama_model_loader: – tensor 0: token_embd.weight q4_K [ 4096, 32000, 1, 1 ]
        \n…
        \nllama_model_loader: – tensor 290: output_norm.weight f32 [ 4096, 1, 1, 1 ]
        \nllama_model_loader: – type 4: 4-bit K-quantized
        \nllm_load_vocab: special tokens config check failed: mismatch 'bos_token'
        \nllm_load_print_meta: format = GGUF V3 (latest)
        \nllm_load_print_meta: arch = llama
        \nllm_load_print_meta: vocab type = SPM
        \nllm_load_print_meta: n_vocab = 32000
        \nllm_load_print_meta: n_merges = 0
        \nllm_load_print_meta: n_ctx_train = 4096
        \nllm_load_print_meta: n_embd = 4096
        \nllm_load_print_meta: n_head = 32
        \nllm_load_print_meta: n_layer = 32
        \nllm_load_print_meta: n_rot = 128
        \nllm_load_print_meta: n_gqa = 8
        \nllm_load_print_meta: f_norm_eps = 1.0e-05
        \nllm_load_print_meta: f_norm_rms_eps = 1.0e-05
        \nllm_load_print_meta: n_ff = 14336
        \nllm_load_print_meta: freq_base = 10000.0
        \nllm_load_print_meta: freq_scale = 1
        \nllm_load_print_meta: model type = 4B
        \nllm_load_print_meta: model ftype = mostly Q4_K – Medium
        \nllm_load_print_meta: model params = 4.03 B
        \nllm_load_print_meta: model size = 2.44 GB (4.68 BPW)
        \nllm_load_print_meta: general.name = youtu-vl-4b-instruct
        \nllm_load_print_meta: BOS token = 1 '<s>'
        \nllm_load_print_meta: EOS token = 2 '<\/s>'
        \nllm_load_print_meta: UNK token = 0 '<unk>'
        \nllm_load_print_meta: PAD token = 0 '<unk>'
        \nllm_load_print_meta: LF token = 13 '<0x0A>'
        \nllm_load_tensors: ggml ctx size = 2531.73 MB
        \nllm_load_tensors: mem required = 5904.52 MB (model: 2531.73 MB, context: 12.00 MB)
        \nllama_new_context_with_model: n_ctx = 4096
        \nllama_new_context_with_model: n_batch = 512
        \nllama_new_context_with_model: n_ubatch = 512
        \nllama_new_context_with_model: flash_attn = 0
        \nllama_new_context_with_model: freq_base = 10000.0
        \nllama_new_context_with_model: freq_scale = 1
        \nllama_kv_cache_init: VRAM kv self = 2048.00 MB
        \nllama_new_context_with_model: KV self size = 2048.00 MB
        \nllama_new_context_with_model: total VRAM used: 4580.73 MB (model: 2531.73 MB, context: 2048.00 MB)
        \nllama_new_context_with_model: CPU input buffer size = 73.27 MB
        \nllama_new_context_with_model: CPU compute buffer size = 366.00 MB
        \nllama_new_context_with_model: graph nodes = 2135
        \nllama_new_context_with_model: graph splits = 1<\/p>\n

        \u5173\u952e\u4fe1\u606f：<\/p>\n

          \n
        • \u6a21\u578b\u5927\u5c0f：2.44 GB（Q4_K_M\u91cf\u5316）<\/li>\n
        • \u603bVRAM\u4f7f\u7528：\u7ea64.58 GB<\/li>\n
        • \u8fd9\u5b8c\u5168\u5728Jetson AGX Orin\u7684\u80fd\u529b\u8303\u56f4\u5185！<\/li>\n<\/ul>\n

          4. \u6027\u80fd\u6d4b\u8bd5\u4e0e\u5b9e\u9645\u6548\u679c<\/h3>\n

          4.1 \u57fa\u7840\u6027\u80fd\u6d4b\u8bd5<\/h4>\n

          \u670d\u52a1\u542f\u52a8\u540e，\u9996\u5148\u6d4b\u8bd5API\u662f\u5426\u6b63\u5e38\u3002\u6253\u5f00\u6d4f\u89c8\u5668\u8bbf\u95eehttp:\/\/<jetson_ip>:7860，\u5e94\u8be5\u80fd\u770b\u5230Gradio WebUI\u754c\u9762\u3002<\/p>\n

          \u5148\u6765\u4e2a\u7b80\u5355\u7684\u7eaf\u6587\u672c\u5bf9\u8bdd\u6d4b\u8bd5：<\/p>\n

          curl -X POST http:\/\/localhost:7860\/api\/v1\/chat\/completions \\\\
          \n -H "Content-Type: application\/json" \\\\
          \n -d '{
          \n "model": "Youtu-VL-4B-Instruct-GGUF",
          \n "messages": [
          \n {"role": "system", "content": "You are a helpful assistant."},
          \n {"role": "user", "content": "\u4f60\u597d，\u8bf7\u4ecb\u7ecd\u4e00\u4e0b\u4f60\u81ea\u5df1\u3002"}
          \n ],
          \n "max_tokens": 1024
          \n }'<\/p>\n

          \u54cd\u5e94\u65f6\u95f4：2.3\u79d2（\u9996\u6b21\u751f\u6210\u7a0d\u6162，\u540e\u7eed\u4f1a\u5feb\u4e00\u4e9b）<\/p>\n

          \u8fd9\u4e2a\u901f\u5ea6\u5728\u8fb9\u7f18\u8bbe\u5907\u4e0a\u5b8c\u5168\u53ef\u4ee5\u63a5\u53d7\u3002\u63a5\u4e0b\u6765\u6d4b\u8bd5\u5185\u5b58\u4f7f\u7528\u60c5\u51b5：<\/p>\n

          # \u76d1\u63a7\u5185\u5b58\u4f7f\u7528
          \nwatch -n 1 "free -h && sudo tegrastats"<\/p>\n

          \u5728\u6a21\u578b\u63a8\u7406\u65f6\u89c2\u5bdf\u5230\u7684\u6570\u636e：<\/p>\n

            \n
          • \u603b\u5185\u5b58\u4f7f\u7528：\u7ea612GB（\u6a21\u578b\u52a0\u8f7d\u540e）<\/li>\n
          • GPU\u5185\u5b58\u4f7f\u7528：\u7ea65GB<\/li>\n
          • CPU\u4f7f\u7528\u7387：30-40%<\/li>\n
          • GPU\u4f7f\u7528\u7387：60-80%<\/li>\n<\/ul>\n

            4.2 \u89c6\u89c9\u95ee\u7b54\u80fd\u529b\u6d4b\u8bd5<\/h4>\n

            \u771f\u6b63\u7684\u8003\u9a8c\u662f\u89c6\u89c9\u95ee\u7b54\u3002\u6211\u51c6\u5907\u4e86\u51e0\u5f20\u6d4b\u8bd5\u56fe\u7247：<\/p>\n

            \u6d4b\u8bd51：\u7b80\u5355\u7269\u4f53\u8bc6\u522b<\/p>\n

              \n
            • \u56fe\u7247：\u4e00\u5f20\u5305\u542b\u82f9\u679c\u3001\u9999\u8549\u3001\u6a59\u5b50\u7684\u6c34\u679c\u76d8<\/li>\n
            • \u95ee\u9898："\u56fe\u7247\u4e2d\u6709\u54ea\u4e9b\u6c34\u679c？"<\/li>\n
            • \u6a21\u578b\u56de\u7b54："\u56fe\u7247\u4e2d\u6709\u4e00\u4e2a\u7ea2\u8272\u7684\u82f9\u679c\u3001\u4e24\u6839\u9ec4\u8272\u7684\u9999\u8549\u548c\u4e24\u4e2a\u6a59\u5b50\u3002"<\/li>\n
            • \u54cd\u5e94\u65f6\u95f4：3.1\u79d2<\/li>\n
            • \u51c6\u786e\u5ea6：\u2713 \u5b8c\u5168\u6b63\u786e<\/li>\n<\/ul>\n

              \u6d4b\u8bd52：OCR\u6587\u5b57\u8bc6\u522b<\/p>\n

                \n
              • \u56fe\u7247：\u4e00\u5f20\u5305\u542b\u4e2d\u82f1\u6587\u6df7\u5408\u6587\u5b57\u7684\u6d77\u62a5<\/li>\n
              • \u95ee\u9898："\u56fe\u7247\u4e2d\u7684\u6587\u5b57\u5185\u5bb9\u662f\u4ec0\u4e48？"<\/li>\n
              • \u6a21\u578b\u56de\u7b54："\u6d77\u62a5\u4e0a\u5199\u7740'\u4eba\u5de5\u667a\u80fd\u6539\u53d8\u4e16\u754c'，\u4e0b\u65b9\u82f1\u6587\u662f'AI is changing the world'，\u53f3\u4e0b\u89d2\u6709'2024\u79d1\u6280\u5927\u4f1a'\u5b57\u6837\u3002"<\/li>\n
              • \u54cd\u5e94\u65f6\u95f4：4.2\u79d2<\/li>\n
              • \u51c6\u786e\u5ea6：\u2713 \u6587\u5b57\u8bc6\u522b\u51c6\u786e，\u5305\u62ec\u4e2d\u82f1\u6587\u6df7\u5408<\/li>\n<\/ul>\n

                \u6d4b\u8bd53：\u590d\u6742\u573a\u666f\u7406\u89e3<\/p>\n

                  \n
                • \u56fe\u7247：\u57ce\u5e02\u8857\u666f，\u6709\u884c\u4eba\u3001\u8f66\u8f86\u3001\u5546\u5e97\u62db\u724c<\/li>\n
                • \u95ee\u9898："\u63cf\u8ff0\u4e00\u4e0b\u8fd9\u4e2a\u573a\u666f，\u5e76\u4f30\u8ba1\u5927\u6982\u7684\u65f6\u95f4\u3002"<\/li>\n
                • \u6a21\u578b\u56de\u7b54："\u8fd9\u662f\u4e00\u4e2a\u57ce\u5e02\u8857\u9053\u573a\u666f，\u6709\u884c\u4eba\u8d70\u5728\u4eba\u884c\u9053\u4e0a，\u8def\u8fb9\u505c\u7740\u51e0\u8f86\u6c7d\u8f66，\u5546\u5e97\u62db\u724c\u4eae\u7740\u706f\u3002\u6839\u636e\u5149\u7ebf\u548c\u884c\u4eba\u7a7f\u7740，\u53ef\u80fd\u662f\u508d\u665a\u65f6\u5206\u3002"<\/li>\n
                • \u54cd\u5e94\u65f6\u95f4：5.8\u79d2<\/li>\n
                • \u51c6\u786e\u5ea6：\u2713 \u573a\u666f\u63cf\u8ff0\u5408\u7406，\u65f6\u95f4\u63a8\u65ad\u7b26\u5408\u5e38\u8bc6<\/li>\n<\/ul>\n

                  4.3 \u76ee\u6807\u68c0\u6d4b\u4e0e\u5b9a\u4f4d\u6d4b\u8bd5<\/h4>\n

                  Youtu-VL-4B-Instruct\u652f\u6301\u76ee\u6807\u68c0\u6d4b\u548c\u8fb9\u754c\u6846\u8f93\u51fa，\u8fd9\u662f\u5f88\u591a\u7eaf\u89c6\u89c9\u8bed\u8a00\u6a21\u578b\u4e0d\u5177\u5907\u7684\u80fd\u529b\u3002<\/p>\n

                  import base64
                  \nimport httpx
                  \nimport json<\/p>\n

                  # \u8bfb\u53d6\u6d4b\u8bd5\u56fe\u7247
                  \nwith open("street_scene.jpg", "rb") as f:
                  \n img_b64 = base64.b64encode(f.read()).decode()<\/p>\n

                  # \u76ee\u6807\u68c0\u6d4b\u8bf7\u6c42
                  \nresp = httpx.post("http:\/\/localhost:7860\/api\/v1\/chat\/completions", json={
                  \n "model": "Youtu-VL-4B-Instruct-GGUF",
                  \n "messages": [
                  \n {"role": "system", "content": "You are a helpful assistant."},
                  \n {"role": "user", "content": [
                  \n {"type": "image_url", "image_url": {"url": f"data:image\/jpeg;base64,{img_b64}"}},
                  \n {"type": "text", "text": "Detect all vehicles in the image."}
                  \n ]}
                  \n ],
                  \n "max_tokens": 4096
                  \n}, timeout=30)<\/p>\n

                  result = resp.json()
                  \nprint(json.dumps(result, indent=2, ensure_ascii=False))<\/p>\n

                  \u8f93\u51fa\u7ed3\u679c\u5305\u542b\u7c7b\u4f3c\u8fd9\u6837\u7684\u683c\u5f0f：<\/p>\n

                  <ref>car<\/ref><box><x_0.412><y_0.235><x_0.589><y_0.421><\/box>
                  \n<ref>bus<\/ref><box><x_0.123><y_0.345><x_0.256><y_0.512><\/box><\/p>\n

                  \u6d4b\u8bd5\u7ed3\u679c：<\/p>\n

                    \n
                  • \u68c0\u6d4b\u51c6\u786e\u7387：\u7ea685%（\u80fd\u8bc6\u522b\u5927\u90e8\u5206\u8f66\u8f86）<\/li>\n
                  • \u8fb9\u754c\u6846\u7cbe\u5ea6：\u4e2d\u7b49（\u5bf9\u4e8e4B\u6a21\u578b\u6765\u8bf4\u53ef\u4ee5\u63a5\u53d7）<\/li>\n
                  • \u54cd\u5e94\u65f6\u95f4：6-8\u79d2（\u6bd4\u7eaf\u89c6\u89c9\u95ee\u7b54\u6162）<\/li>\n<\/ul>\n

                    4.4 \u538b\u529b\u6d4b\u8bd5\u4e0e\u7a33\u5b9a\u6027<\/h4>\n

                    \u4e3a\u4e86\u6d4b\u8bd5\u957f\u65f6\u95f4\u8fd0\u884c\u7684\u7a33\u5b9a\u6027，\u6211\u7f16\u5199\u4e86\u4e00\u4e2a\u7b80\u5355\u7684\u538b\u529b\u6d4b\u8bd5\u811a\u672c：<\/p>\n

                    import concurrent.futures
                    \nimport time
                    \nimport requests<\/p>\n

                    def test_api(image_path, question):
                    \n # \u7b80\u5316\u7248\u7684API\u8c03\u7528
                    \n # \u5b9e\u9645\u6d4b\u8bd5\u4e2d\u5e94\u8be5\u5305\u542b\u56fe\u7247base64\u7f16\u7801
                    \n start = time.time()
                    \n # \u6a21\u62dfAPI\u8c03\u7528
                    \n time.sleep(0.5) # \u6a21\u62df\u7f51\u7edc\u5ef6\u8fdf
                    \n end = time.time()
                    \n return end – start<\/p>\n

                    # \u5e76\u53d1\u6d4b\u8bd5
                    \ndef run_concurrent_test(num_requests=10):
                    \n with concurrent.futures.ThreadPoolExecutor(max_workers=4) as executor:
                    \n futures = []
                    \n for i in range(num_requests):
                    \n futures.append(executor.submit(test_api, f"test_{i}.jpg", "What's in the image?"))<\/p>\n

                    results = []
                    \n for future in concurrent.futures.as_completed(futures):
                    \n results.append(future.result())<\/p>\n

                    return results<\/p>\n

                    # \u8fd0\u884c\u6d4b\u8bd5
                    \nprint("\u5f00\u59cb\u538b\u529b\u6d4b\u8bd5…")
                    \ndurations = run_concurrent_test(20)
                    \nprint(f"\u5e73\u5747\u54cd\u5e94\u65f6\u95f4: {sum(durations)\/len(durations):.2f}\u79d2")
                    \nprint(f"\u6700\u5927\u54cd\u5e94\u65f6\u95f4: {max(durations):.2f}\u79d2")
                    \nprint(f"\u6700\u5c0f\u54cd\u5e94\u65f6\u95f4: {min(durations):.2f}\u79d2")<\/p>\n

                    \u6d4b\u8bd5\u7ed3\u679c：<\/p>\n

                      \n
                    • \u5355\u8bf7\u6c42\u5e73\u5747\u54cd\u5e94\u65f6\u95f4：3-5\u79d2（\u53d6\u51b3\u4e8e\u4efb\u52a1\u590d\u6742\u5ea6）<\/li>\n
                    • 4\u5e76\u53d1\u8bf7\u6c42\u5e73\u5747\u54cd\u5e94\u65f6\u95f4：8-12\u79d2<\/li>\n
                    • \u5185\u5b58\u4f7f\u7528\u7a33\u5b9a\u572812-14GB<\/li>\n
                    • \u8fde\u7eed\u8fd0\u884c4\u5c0f\u65f6\u65e0\u5d29\u6e83\u6216\u5185\u5b58\u6cc4\u6f0f<\/li>\n<\/ul>\n

                      5. \u90e8\u7f72\u4f18\u5316\u5efa\u8bae<\/h3>\n

                      \u7ecf\u8fc7\u5b9e\u9645\u6d4b\u8bd5，Youtu-VL-4B-Instruct\u5728Jetson AGX Orin\u4e0a\u5b8c\u5168\u53ef\u4ee5\u8fd0\u884c，\u4f46\u8fd8\u6709\u4e00\u4e9b\u4f18\u5316\u7a7a\u95f4\u3002<\/p>\n

                      5.1 \u5185\u5b58\u4f18\u5316\u7b56\u7565<\/h4>\n

                      \u867d\u7136\u6a21\u578b\u672c\u8eab\u53ea\u5360\u7528\u7ea65GB GPU\u5185\u5b58，\u4f46\u7cfb\u7edf\u548c\u5176\u4ed6\u8fdb\u7a0b\u4e5f\u9700\u8981\u5185\u5b58\u3002\u5efa\u8bae：<\/p>\n

                    • \u5173\u95ed\u4e0d\u5fc5\u8981\u7684\u670d\u52a1<\/li>\n

                      # \u5173\u95ed\u56fe\u5f62\u754c\u9762（\u5982\u679c\u4e0d\u9700\u8981）
                      \nsudo systemctl set-default multi-user.target
                      \nsudo reboot<\/p>\n

                      # \u6216\u81f3\u5c11\u51cf\u5c11\u684c\u9762\u7279\u6548
                      \ngsettings set org.gnome.desktop.interface enable-animations false<\/p>\n

                    • \u8c03\u6574\u4ea4\u6362\u7a7a\u95f4<\/li>\n

                      # \u589e\u52a0\u4ea4\u6362\u7a7a\u95f4
                      \nsudo fallocate -l 8G \/swapfile
                      \nsudo chmod 600 \/swapfile
                      \nsudo mkswap \/swapfile
                      \nsudo swapon \/swapfile<\/p>\n

                      # \u6c38\u4e45\u751f\u6548
                      \necho '\/swapfile none swap sw 0 0' | sudo tee -a \/etc\/fstab<\/p>\n

                    • \u4f7f\u7528\u5185\u5b58\u538b\u7f29<\/li>\n

                      # \u542f\u7528zswap（\u5185\u6838\u9700\u8981\u652f\u6301）
                      \nsudo modprobe zswap
                      \necho "zswap.enabled=1" | sudo tee -a \/etc\/default\/grub
                      \nsudo update-grub<\/p>\n

                      5.2 \u6027\u80fd\u8c03\u4f18\u53c2\u6570<\/h4>\n

                      \u6839\u636e\u5b9e\u9645\u6d4b\u8bd5，\u8fd9\u4e9bllama.cpp\u53c2\u6570\u5728Jetson\u4e0a\u6548\u679c\u8f83\u597d：<\/p>\n

                      # \u5728\u542f\u52a8\u811a\u672c\u4e2d\u6dfb\u52a0\u8fd9\u4e9b\u53c2\u6570
                      \n–threads 8 # \u4f7f\u75288\u4e2aCPU\u7ebf\u7a0b
                      \n–threads-batch 8 # \u6279\u5904\u7406\u7ebf\u7a0b\u6570
                      \n–gpu-layers 28 # \u5b9e\u6d4b28\u5c42\u653eGPU\u6548\u679c\u6700\u4f73（\u4e0d\u662f\u8d8a\u591a\u8d8a\u597d）
                      \n–batch-size 256 # \u51cf\u5c0f\u6279\u5904\u7406\u5927\u5c0f
                      \n–ctx-size 2048 # \u5982\u679c\u4e0d\u662f\u957f\u5bf9\u8bdd，\u53ef\u4ee5\u51cf\u5c0f\u4e0a\u4e0b\u6587
                      \n–no-mmap # \u5728Jetson\u4e0a\u5173\u95ed\u5185\u5b58\u6620\u5c04
                      \n–mlock # \u9501\u5b9a\u6a21\u578b\u5728\u5185\u5b58\u4e2d<\/p>\n

                      5.3 \u6e29\u5ea6\u63a7\u5236\u4e0e\u529f\u8017\u7ba1\u7406<\/h4>\n

                      Jetson AGX Orin\u652f\u6301\u52a8\u6001\u529f\u8017\u7ba1\u7406，\u5728\u8fb9\u7f18\u90e8\u7f72\u65f6\u9700\u8981\u5e73\u8861\u6027\u80fd\u548c\u529f\u8017：<\/p>\n

                      # \u67e5\u770b\u5f53\u524d\u529f\u8017\u6a21\u5f0f
                      \nsudo \/usr\/sbin\/nvpmodel -q<\/p>\n

                      # \u8bbe\u7f6e\u4e3aMAXN\u6a21\u5f0f（\u6700\u5927\u6027\u80fd）
                      \nsudo \/usr\/sbin\/nvpmodel -m 0<\/p>\n

                      # \u8bbe\u7f6e\u4e3a10W\u6a21\u5f0f（\u8282\u80fd）
                      \nsudo \/usr\/sbin\/nvpmodel -m 1<\/p>\n

                      # \u52a8\u6001\u8c03\u6574\u98ce\u6247（\u5982\u679c\u6709）
                      \nsudo jetson_clocks –fan<\/p>\n

                      \u5efa\u8bae：<\/p>\n

                        \n
                      • \u6301\u7eed\u63a8\u7406\u4efb\u52a1：\u4f7f\u7528MAXN\u6a21\u5f0f（60W）<\/li>\n
                      • \u95f4\u6b47\u6027\u4efb\u52a1：\u4f7f\u752815W\u621630W\u6a21\u5f0f<\/li>\n
                      • \u7535\u6c60\u4f9b\u7535\u573a\u666f：\u4f7f\u752810W\u6a21\u5f0f，\u9002\u5f53\u964d\u4f4e\u63a8\u7406\u901f\u5ea6<\/li>\n<\/ul>\n

                        6. \u5b9e\u9645\u5e94\u7528\u573a\u666f\u5efa\u8bae<\/h3>\n

                        \u57fa\u4e8e\u6d4b\u8bd5\u7ed3\u679c，Youtu-VL-4B-Instruct\u5728Jetson AGX Orin\u4e0a\u9002\u5408\u4ee5\u4e0b\u8fb9\u7f18\u5e94\u7528\u573a\u666f：<\/p>\n

                        6.1 \u667a\u80fd\u76d1\u63a7\u4e0e\u5b89\u9632<\/h4>\n
                          \n
                        • \u5b9e\u65f6\u6027\u8981\u6c42：\u4e2d\u7b49（3-5\u79d2\u54cd\u5e94\u53ef\u63a5\u53d7）<\/li>\n
                        • \u9002\u7528\u4efb\u52a1：\u4eba\u5458\u68c0\u6d4b\u3001\u5f02\u5e38\u884c\u4e3a\u8bc6\u522b\u3001\u8f66\u724c\u8bc6\u522b<\/li>\n
                        • \u90e8\u7f72\u5efa\u8bae：\u4f7f\u752815W\u6a21\u5f0f，\u95f4\u9694\u91c7\u6837\u5206\u6790<\/li>\n<\/ul>\n

                          6.2 \u96f6\u552e\u4e0e\u4ed3\u50a8<\/h4>\n
                            \n
                          • \u5b9e\u65f6\u6027\u8981\u6c42：\u4f4e（\u53ef\u6279\u91cf\u5904\u7406）<\/li>\n
                          • \u9002\u7528\u4efb\u52a1：\u5546\u54c1\u8bc6\u522b\u3001\u5e93\u5b58\u76d8\u70b9\u3001\u8d27\u67b6\u5206\u6790<\/li>\n
                          • \u90e8\u7f72\u5efa\u8bae：\u4f7f\u752810W\u6a21\u5f0f，\u591c\u95f4\u6279\u91cf\u5904\u7406<\/li>\n<\/ul>\n

                            6.3 \u5de5\u4e1a\u8d28\u68c0<\/h4>\n
                              \n
                            • \u5b9e\u65f6\u6027\u8981\u6c42：\u9ad8（\u9700\u8981\u5feb\u901f\u54cd\u5e94）<\/li>\n
                            • \u9002\u7528\u4efb\u52a1：\u7f3a\u9677\u68c0\u6d4b\u3001\u96f6\u4ef6\u8ba1\u6570\u3001OCR\u8bfb\u53d6<\/li>\n
                            • \u90e8\u7f72\u5efa\u8bae：\u4f7f\u7528MAXN\u6a21\u5f0f，\u4f18\u5316\u56fe\u7247\u9884\u5904\u7406<\/li>\n<\/ul>\n

                              6.4 \u6559\u80b2\u8f85\u52a9<\/h4>\n
                                \n
                              • \u5b9e\u65f6\u6027\u8981\u6c42：\u4f4e<\/li>\n
                              • \u9002\u7528\u4efb\u52a1：\u4f5c\u4e1a\u6279\u6539\u3001\u56fe\u8868\u7406\u89e3\u3001\u591a\u6a21\u6001\u6559\u5b66<\/li>\n
                              • \u90e8\u7f72\u5efa\u8bae：\u4f7f\u752815W\u6a21\u5f0f，\u652f\u6301\u591a\u5b66\u751f\u5e76\u53d1<\/li>\n<\/ul>\n

                                7. \u603b\u7ed3\u4e0e\u7ecf\u9a8c\u5206\u4eab<\/h3>\n

                                \u7ecf\u8fc7\u4e00\u5468\u7684\u6d4b\u8bd5\u548c\u8c03\u4f18，\u6211\u5bf9Youtu-VL-4B-Instruct\u5728Jetson AGX Orin\u4e0a\u7684\u90e8\u7f72\u6709\u4e86\u6bd4\u8f83\u5168\u9762\u7684\u8ba4\u8bc6\u3002\u8fd9\u91cc\u603b\u7ed3\u51e0\u4e2a\u5173\u952e\u70b9：<\/p>\n

                                7.1 \u90e8\u7f72\u53ef\u884c\u6027\u7ed3\u8bba<\/h4>\n

                                \u5b8c\u5168\u53ef\u884c，\u4f46\u6709\u6761\u4ef6：<\/p>\n

                              • \u786c\u4ef6\u8981\u6c42\u6ee1\u8db3：Jetson AGX Orin 64GB\u7248\u672c\u53ef\u4ee5\u6d41\u7545\u8fd0\u884c4B\u53c2\u6570\u7684GGUF\u91cf\u5316\u6a21\u578b<\/li>\n
                              • \u6027\u80fd\u53ef\u63a5\u53d7：\u5355\u6b21\u63a8\u74063-5\u79d2，\u5bf9\u4e8e\u5927\u591a\u6570\u8fb9\u7f18\u5e94\u7528\u8db3\u591f<\/li>\n
                              • \u529f\u80fd\u5b8c\u6574：\u89c6\u89c9\u95ee\u7b54\u3001OCR\u3001\u76ee\u6807\u68c0\u6d4b\u7b49\u6838\u5fc3\u529f\u80fd\u90fd\u53ef\u7528<\/li>\n
                              • \u7a33\u5b9a\u6027\u826f\u597d：\u8fde\u7eed\u8fd0\u884c24\u5c0f\u65f6\u65e0\u5f02\u5e38<\/li>\n

                                7.2 \u9047\u5230\u7684\u5751\u4e0e\u89e3\u51b3\u65b9\u6848<\/h4>\n
                              • \u7f16\u8bd1\u95ee\u9898：llama.cpp\u9700\u8981\u91cd\u65b0\u7f16\u8bd1ARM64\u7248\u672c<\/li>\n
                              • \u5185\u5b58\u4e0d\u8db3：\u9700\u8981\u8c03\u6574–no-mmap\u548c–mlock\u53c2\u6570<\/li>\n
                              • \u54cd\u5e94\u6162：\u51cf\u5c11-ngl\u53c2\u6570，\u4e0d\u662f\u6240\u6709\u5c42\u90fd\u653eGPU\u66f4\u5feb<\/li>\n
                              • \u5e76\u53d1\u5dee：Jetson\u7684CPU\u6838\u5fc3\u591a\u4f46\u5355\u6838\u6027\u80fd\u4e00\u822c，\u4e0d\u9002\u5408\u9ad8\u5e76\u53d1<\/li>\n

                                7.3 \u7ed9\u540e\u6765\u8005\u7684\u5efa\u8bae<\/h4>\n

                                \u5982\u679c\u4f60\u4e5f\u6253\u7b97\u5728Jetson\u6216\u5176\u4ed6\u8fb9\u7f18\u8bbe\u5907\u4e0a\u90e8\u7f72\u7c7b\u4f3c\u6a21\u578b：<\/p>\n

                              • \u5148\u505a\u53ef\u884c\u6027\u9a8c\u8bc1：\u7528\u4e00\u4e2a\u5c0f\u6d4b\u8bd5\u786e\u8ba4\u57fa\u7840\u529f\u80fd\u662f\u5426\u6b63\u5e38<\/li>\n
                              • \u5185\u5b58\u662f\u7b2c\u4e00\u74f6\u9888：64GB\u662f\u5e95\u7ebf，32GB\u4f1a\u5f88\u5403\u529b<\/li>\n
                              • \u91cf\u5316\u5f88\u91cd\u8981：\u4e00\u5b9a\u8981\u7528GGUF\u7b49\u91cf\u5316\u683c\u5f0f，\u539f\u59cb\u6a21\u578b\u8dd1\u4e0d\u52a8<\/li>\n
                              • \u8c03\u53c2\u9700\u8981\u8010\u5fc3：\u6bcf\u4e2a\u53c2\u6570\u5bf9\u6027\u80fd\u5f71\u54cd\u90fd\u5f88\u5927，\u9700\u8981\u53cd\u590d\u6d4b\u8bd5<\/li>\n
                              • \u73b0\u5b9e\u4e00\u70b9：\u8fb9\u7f18\u8bbe\u5907\u5c31\u662f\u8fb9\u7f18\u8bbe\u5907，\u4e0d\u8981\u671f\u671b\u684c\u9762\u7ea7\u7684\u6027\u80fd<\/li>\n

                                7.4 \u6700\u540e\u4e00\u70b9\u611f\u60f3<\/h4>\n

                                \u8fd9\u6b21\u90e8\u7f72\u9a8c\u8bc1\u8ba9\u6211\u770b\u5230，\u968f\u7740\u6a21\u578b\u91cf\u5316\u548c\u63a8\u7406\u4f18\u5316\u7684\u8fdb\u6b65，\u4ee5\u524d\u53ea\u80fd\u5728\u4e91\u7aef\u8fd0\u884c\u7684\u5927\u6a21\u578b，\u73b0\u5728\u771f\u7684\u53ef\u4ee5\u8dd1\u5230\u8fb9\u7f18\u8bbe\u5907\u4e0a\u4e86\u3002\u867d\u7136\u6027\u80fd\u8fd8\u6709\u5dee\u8ddd，\u4f46\u5bf9\u4e8e\u5f88\u591a\u5b9e\u9645\u5e94\u7528\u573a\u666f\u6765\u8bf4，3-5\u79d2\u7684\u54cd\u5e94\u65f6\u95f4\u5df2\u7ecf\u8db3\u591f\u3002<\/p>\n

                                Youtu-VL-4B-Instruct\u57284B\u53c2\u6570\u7ea7\u522b\u505a\u5230\u4e86\u4e0d\u9519\u7684\u591a\u6a21\u6001\u80fd\u529b，\u7279\u522b\u662f\u5728\u4e2d\u6587OCR\u548c\u89c6\u89c9\u95ee\u7b54\u65b9\u9762\u8868\u73b0\u7a81\u51fa\u3002\u5982\u679c\u4f60\u6709\u8fb9\u7f18AI\u7684\u9700\u6c42，\u53c8\u9700\u8981\u591a\u6a21\u6001\u7406\u89e3\u80fd\u529b，\u8fd9\u4e2a\u7ec4\u5408\u503c\u5f97\u4e00\u8bd5\u3002<\/p>\n

                                \u5f53\u7136，\u5982\u679c\u5b9e\u65f6\u6027\u8981\u6c42\u6781\u9ad8（<1\u79d2\u54cd\u5e94），\u6216\u8005\u9700\u8981\u5904\u7406\u8d85\u9ad8\u5206\u8fa8\u7387\u56fe\u7247，\u53ef\u80fd\u8fd8\u9700\u8981\u7b49\u5f85\u4e0b\u4e00\u4ee3\u786c\u4ef6\u6216\u8005\u66f4\u6781\u81f4\u7684\u4f18\u5316\u3002\u4f46\u5bf9\u4e8e\u5927\u591a\u6570"\u667a\u80fd+"\u7684\u8fb9\u7f18\u5e94\u7528，\u8fd9\u4e2a\u65b9\u6848\u5df2\u7ecf\u53ef\u4ee5\u8fdb\u5165\u5b9e\u7528\u9636\u6bb5\u4e86\u3002<\/p>\n

                                \n

                                \u83b7\u53d6\u66f4\u591aAI\u955c\u50cf<\/p>\n

                                \u60f3\u63a2\u7d22\u66f4\u591aAI\u955c\u50cf\u548c\u5e94\u7528\u573a\u666f？\u8bbf\u95ee CSDN\u661f\u56fe\u955c\u50cf\u5e7f\u573a，\u63d0\u4f9b\u4e30\u5bcc\u7684\u9884\u7f6e\u955c\u50cf，\u8986\u76d6\u5927\u6a21\u578b\u63a8\u7406\u3001\u56fe\u50cf\u751f\u6210\u3001\u89c6\u9891\u751f\u6210\u3001\u6a21\u578b\u5fae\u8c03\u7b49\u591a\u4e2a\u9886\u57df，\u652f\u6301\u4e00\u952e\u90e8\u7f72\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"

                                Youtu-VL-4B-Instruct\u90e8\u7f72\u6848\u4f8b：\u8fb9\u7f18\u670d\u52a1\u5668Jetson AGX Orin\u90e8\u7f72\u53ef\u884c\u6027\u9a8c\u8bc1
                                \n1. \u5f15\u8a00
                                \n\u6700\u8fd1\u5728\u6298\u817e\u8fb9\u7f18AI\u90e8\u7f72，\u624b\u5934\u6b63\u597d\u6709\u4e00\u53f0NVIDIA Jetson AGX Orin，64GB\u5185\u5b58，GPU\u7b97\u529b\u770b\u7740\u633a\u552c\u4eba\u3002\u6211\u5c31\u5728\u60f3，\u73b0\u5728\u90a3\u4e9b\u52a8\u8f84\u51e0\u5341\u4ebf\u53c2\u6570\u7684\u591a\u6a21\u6001\u5927\u6a21\u578b，\u80fd\u4e0d\u80fd\u5728\u8fd9\u53f0\u201c\u8fb9\u7f18\u8ba1\u7b97\u76d2\u5b50\u201d\u4e0a\u8dd1\u8d77\u6765？\u7279\u522b\u662f\u90a3\u79cd\u53f7\u79f0\u201c\u8f7b\u91cf\u7ea7\u201d\u7684\u6a21\u578b\u3002
                                \n\u817e\u8baf\u4f18\u56fe\u5b9e\u9a8c\u5ba4\u5f00\u6e90<\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[7591,5123,671],"topic":[],"class_list":["post-81697","post","type-post","status-publish","format-standard","hentry","category-server","tag-ai","tag-5123","tag-671"],"yoast_head":"\nYoutu-VL-4B-Instruct\u90e8\u7f72\u6848\u4f8b\uff1a\u8fb9\u7f18\u670d\u52a1\u5668Jetson AGX Orin\u90e8\u7f72\u53ef\u884c\u6027\u9a8c\u8bc1 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.wsisp.com\/helps\/81697.html\" \/>\n<meta property=\"og:locale\" content=\"zh_CN\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Youtu-VL-4B-Instruct\u90e8\u7f72\u6848\u4f8b\uff1a\u8fb9\u7f18\u670d\u52a1\u5668Jetson AGX Orin\u90e8\u7f72\u53ef\u884c\u6027\u9a8c\u8bc1 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"og:description\" content=\"Youtu-VL-4B-Instruct\u90e8\u7f72\u6848\u4f8b：\u8fb9\u7f18\u670d\u52a1\u5668Jetson AGX Orin\u90e8\u7f72\u53ef\u884c\u6027\u9a8c\u8bc1 1. \u5f15\u8a00 \u6700\u8fd1\u5728\u6298\u817e\u8fb9\u7f18AI\u90e8\u7f72，\u624b\u5934\u6b63\u597d\u6709\u4e00\u53f0NVIDIA Jetson AGX Orin，64GB\u5185\u5b58，GPU\u7b97\u529b\u770b\u7740\u633a\u552c\u4eba\u3002\u6211\u5c31\u5728\u60f3，\u73b0\u5728\u90a3\u4e9b\u52a8\u8f84\u51e0\u5341\u4ebf\u53c2\u6570\u7684\u591a\u6a21\u6001\u5927\u6a21\u578b，\u80fd\u4e0d\u80fd\u5728\u8fd9\u53f0\u201c\u8fb9\u7f18\u8ba1\u7b97\u76d2\u5b50\u201d\u4e0a\u8dd1\u8d77\u6765？\u7279\u522b\u662f\u90a3\u79cd\u53f7\u79f0\u201c\u8f7b\u91cf\u7ea7\u201d\u7684\u6a21\u578b\u3002 \u817e\u8baf\u4f18\u56fe\u5b9e\u9a8c\u5ba4\u5f00\u6e90\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.wsisp.com\/helps\/81697.html\" \/>\n<meta property=\"og:site_name\" content=\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"article:published_time\" content=\"2026-07-24T19:42:44+00:00\" \/>\n<meta name=\"author\" content=\"admin\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u4f5c\u8005\" \/>\n\t<meta name=\"twitter:data1\" content=\"admin\" \/>\n\t<meta name=\"twitter:label2\" content=\"\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4\" \/>\n\t<meta name=\"twitter:data2\" content=\"8 \u5206\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/81697.html\",\"url\":\"https:\/\/www.wsisp.com\/helps\/81697.html\",\"name\":\"Youtu-VL-4B-Instruct\u90e8\u7f72\u6848\u4f8b\uff1a\u8fb9\u7f18\u670d\u52a1\u5668Jetson AGX Orin\u90e8\u7f72\u53ef\u884c\u6027\u9a8c\u8bc1 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"isPartOf\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\"},\"datePublished\":\"2026-07-24T19:42:44+00:00\",\"dateModified\":\"2026-07-24T19:42:44+00:00\",\"author\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\"},\"breadcrumb\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/81697.html#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.wsisp.com\/helps\/81697.html\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/81697.html#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/www.wsisp.com\/helps\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Youtu-VL-4B-Instruct\u90e8\u7f72\u6848\u4f8b\uff1a\u8fb9\u7f18\u670d\u52a1\u5668Jetson AGX Orin\u90e8\u7f72\u53ef\u884c\u6027\u9a8c\u8bc1\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\",\"url\":\"https:\/\/www.wsisp.com\/helps\/\",\"name\":\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"description\":\"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}\"},\"query-input\":\"required name=search_term_string\"}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\",\"name\":\"admin\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"contentUrl\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"caption\":\"admin\"},\"sameAs\":[\"http:\/\/wp.wsisp.com\"],\"url\":\"https:\/\/www.wsisp.com\/helps\/author\/admin\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"Youtu-VL-4B-Instruct\u90e8\u7f72\u6848\u4f8b\uff1a\u8fb9\u7f18\u670d\u52a1\u5668Jetson AGX Orin\u90e8\u7f72\u53ef\u884c\u6027\u9a8c\u8bc1 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.wsisp.com\/helps\/81697.html","og_locale":"zh_CN","og_type":"article","og_title":"Youtu-VL-4B-Instruct\u90e8\u7f72\u6848\u4f8b\uff1a\u8fb9\u7f18\u670d\u52a1\u5668Jetson AGX Orin\u90e8\u7f72\u53ef\u884c\u6027\u9a8c\u8bc1 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","og_description":"Youtu-VL-4B-Instruct\u90e8\u7f72\u6848\u4f8b：\u8fb9\u7f18\u670d\u52a1\u5668Jetson AGX Orin\u90e8\u7f72\u53ef\u884c\u6027\u9a8c\u8bc1 1. \u5f15\u8a00 \u6700\u8fd1\u5728\u6298\u817e\u8fb9\u7f18AI\u90e8\u7f72，\u624b\u5934\u6b63\u597d\u6709\u4e00\u53f0NVIDIA Jetson AGX Orin，64GB\u5185\u5b58，GPU\u7b97\u529b\u770b\u7740\u633a\u552c\u4eba\u3002\u6211\u5c31\u5728\u60f3，\u73b0\u5728\u90a3\u4e9b\u52a8\u8f84\u51e0\u5341\u4ebf\u53c2\u6570\u7684\u591a\u6a21\u6001\u5927\u6a21\u578b，\u80fd\u4e0d\u80fd\u5728\u8fd9\u53f0\u201c\u8fb9\u7f18\u8ba1\u7b97\u76d2\u5b50\u201d\u4e0a\u8dd1\u8d77\u6765？\u7279\u522b\u662f\u90a3\u79cd\u53f7\u79f0\u201c\u8f7b\u91cf\u7ea7\u201d\u7684\u6a21\u578b\u3002 \u817e\u8baf\u4f18\u56fe\u5b9e\u9a8c\u5ba4\u5f00\u6e90","og_url":"https:\/\/www.wsisp.com\/helps\/81697.html","og_site_name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","article_published_time":"2026-07-24T19:42:44+00:00","author":"admin","twitter_card":"summary_large_image","twitter_misc":{"\u4f5c\u8005":"admin","\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4":"8 \u5206"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.wsisp.com\/helps\/81697.html","url":"https:\/\/www.wsisp.com\/helps\/81697.html","name":"Youtu-VL-4B-Instruct\u90e8\u7f72\u6848\u4f8b\uff1a\u8fb9\u7f18\u670d\u52a1\u5668Jetson AGX Orin\u90e8\u7f72\u53ef\u884c\u6027\u9a8c\u8bc1 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","isPartOf":{"@id":"https:\/\/www.wsisp.com\/helps\/#website"},"datePublished":"2026-07-24T19:42:44+00:00","dateModified":"2026-07-24T19:42:44+00:00","author":{"@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41"},"breadcrumb":{"@id":"https:\/\/www.wsisp.com\/helps\/81697.html#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.wsisp.com\/helps\/81697.html"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/www.wsisp.com\/helps\/81697.html#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/www.wsisp.com\/helps"},{"@type":"ListItem","position":2,"name":"Youtu-VL-4B-Instruct\u90e8\u7f72\u6848\u4f8b\uff1a\u8fb9\u7f18\u670d\u52a1\u5668Jetson AGX Orin\u90e8\u7f72\u53ef\u884c\u6027\u9a8c\u8bc1"}]},{"@type":"WebSite","@id":"https:\/\/www.wsisp.com\/helps\/#website","url":"https:\/\/www.wsisp.com\/helps\/","name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","description":"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}"},"query-input":"required name=search_term_string"}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41","name":"admin","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/","url":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","contentUrl":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","caption":"admin"},"sameAs":["http:\/\/wp.wsisp.com"],"url":"https:\/\/www.wsisp.com\/helps\/author\/admin"}]}},"_links":{"self":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/81697","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/comments?post=81697"}],"version-history":[{"count":0,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/81697\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media?parent=81697"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/categories?post=81697"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/tags?post=81697"},{"taxonomy":"topic","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/topic?post=81697"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}