Jetson Orin 部署 OpenClaw 开启 Memory Search

1
apt-get update
2
apt-get install pciutils build-essential cmake curl libcurl4-openssl-dev -y
3
git clone https://github.com/ggml-org/llama.cpp
4

5
cmake -S llama.cpp -B llama.cpp/build \
6
    -DCMAKE_BUILD_TYPE=Release \
7
    -DBUILD_SHARED_LIBS=OFF \
8
    -DGGML_CUDA=ON \
9
    -DCMAKE_CUDA_ARCHITECTURES=87 \
10
    -DLLAMA_CURL=ON \
11
    -DLLAMA_SERVER=ON
12

13
cmake --build llama.cpp/build -j$(nproc) \
14
    --target llama-cli llama-gguf-split llama-server

1
mkdir -p llama-ws
2
cp llama.cpp/build/bin/llama-* llama-ws
3

4
wget -O llama-ws/embeddinggemma-300m-qat-Q8_0.gguf \
5
  "https://huggingface.co/ggml-org/embeddinggemma-300m-qat-q8_0-GGUF/resolve/main/embeddinggemma-300m-qat-Q8_0.gguf"

1
cd llama-ws
2
./llama-server --model /home/ros/llama.cpp/embeddinggemma-300m-qat-Q8_0.gguf --embedding --n-gpu-layers 99 --batch-size 1024 --ubatch-size 1024 --cache-ram 0 --host 127.0.0.1 --port 8080

1
"memorySearch": {
2
        "enabled": true,
3
        "sources": [
4
          "memory",
5
          "sessions"
6
        ],
7
        "provider": "openai-compatible",
8
        "fallback": "none",
9
        "remote": {
10
          "baseUrl": "http://127.0.0.1:8080/v1",
11
          "apiKey": "llama-local"
12
        },
13
        "model": "embeddinggemma-300m-qat-Q8_0.gguf"
14
      }