llama.cpp/examples/rpc
Radoslav Gerganov 5e31828d3e
ggml : add RPC backend (#6829)
* ggml : add RPC backend

The RPC backend proxies all operations to a remote server which runs a
regular backend (CPU, CUDA, Metal, etc).

* set TCP_NODELAY

* add CI workflows

* Address review comments

* fix warning

* implement llama_max_devices() for RPC

* Address review comments

* Address review comments

* wrap sockfd into a struct

* implement get_alignment and get_max_size

* add get_device_memory

* fix warning

* win32 support

* add README

* readme : trim trailing whitespace

* Address review comments

* win32 fix

* Address review comments

* fix compile warnings on macos
2024-05-14 14:27:19 +03:00
..
CMakeLists.txt ggml : add RPC backend (#6829) 2024-05-14 14:27:19 +03:00
README.md ggml : add RPC backend (#6829) 2024-05-14 14:27:19 +03:00
rpc-server.cpp ggml : add RPC backend (#6829) 2024-05-14 14:27:19 +03:00

Overview

The rpc-server allows running ggml backend on a remote host. The RPC backend communicates with one or several instances of rpc-server and offloads computations to them. This can be used for distributed LLM inference with llama.cpp in the following way:

flowchart TD
    rpcb---|TCP|srva
    rpcb---|TCP|srvb
    rpcb-.-|TCP|srvn
    subgraph hostn[Host N]
    srvn[rpc-server]-.-backend3["Backend (CUDA,Metal,etc.)"]
    end
    subgraph hostb[Host B]
    srvb[rpc-server]---backend2["Backend (CUDA,Metal,etc.)"]
    end
    subgraph hosta[Host A]
    srva[rpc-server]---backend["Backend (CUDA,Metal,etc.)"]
    end
    subgraph host[Main Host]
    ggml[llama.cpp]---rpcb[RPC backend]
    end
    style hostn stroke:#66,stroke-width:2px,stroke-dasharray: 5 5

Each host can run a different backend, e.g. one with CUDA and another with Metal. You can also run multiple rpc-server instances on the same host, each with a different backend.

Usage

On each host, build the corresponding backend with cmake and add -DLLAMA_RPC=ON to the build options. For example, to build the CUDA backend with RPC support:

mkdir build-rpc-cuda
cd build-rpc-cuda
cmake .. -DLLAMA_CUDA=ON -DLLAMA_RPC=ON
cmake --build . --config Release

Then, start the rpc-server with the backend:

$ bin/rpc-server 0.0.0.0 50052
create_backend: using CUDA backend
ggml_cuda_init: GGML_CUDA_FORCE_MMQ:   no
ggml_cuda_init: CUDA_USE_TENSOR_CORES: yes
ggml_cuda_init: found 1 CUDA devices:
  Device 0: NVIDIA T1200 Laptop GPU, compute capability 7.5, VMM: yes
Starting RPC server on 0.0.0.0:50052

When using the CUDA backend, you can specify the device with the CUDA_VISIBLE_DEVICES environment variable, e.g.:

$ CUDA_VISIBLE_DEVICES=0 bin/rpc-server 0.0.0.0 50052

This way you can run multiple rpc-server instances on the same host, each with a different CUDA device.

On the main host build llama.cpp only with -DLLAMA_RPC=ON:

mkdir build-rpc
cd build-rpc
cmake .. -DLLAMA_RPC=ON
cmake --build . --config Release

Finally, use the --rpc option to specify the host and port of each rpc-server:

$ bin/main -m ../models/tinyllama-1b/ggml-model-f16.gguf -p "Hello, my name is" --repeat-penalty 1.0 -n 64 --rpc 192.168.88.10:50052,192.168.88.11:50052 -ngl 99