llama-swap

llama.cpp's server can't swap models on demand. So let's swap the server on demand instead!

llama-swap is a proxy server that sits in front of llama-server. When a request for /v1/chat/completions comes in it will extract the model requested and change the underlying llama-server automatically.

• ✅ easy to deploy: single binary with no dependencies
• ✅ full control over llama-server's startup settings
• ✅ ❤️ for users who are rely on llama.cpp for LLM inference

config.yaml

llama-swap's configuration purposefully simple.

# Seconds to wait for llama.cpp to load and be ready to serve requests
# Default (and minimum) is 15 seconds
healthCheckTimeout: 60

# define valid model values and the upstream server start
models:
  "llama":
    cmd: llama-server --port 8999 -m Llama-3.2-1B-Instruct-Q4_K_M.gguf

    # where to reach the server started by cmd
    proxy: http://127.0.0.1:8999

    # aliases model names to use this configuration for
    aliases:
    - "gpt-4o-mini"
    - "gpt-3.5-turbo"

    # wait for this path to return an HTTP 200 before serving requests
    # defaults to /health to match llama.cpp
    #
    # use "none" to skip endpoint checking. This may cause requests to fail
    # until the server is ready
    checkEndpoint: /custom-endpoint

  "qwen":
    # environment variables to pass to the command
    env:
      - "CUDA_VISIBLE_DEVICES=0"

    # multiline for readability
    cmd: >
      llama-server --port 8999
      --model path/to/Qwen2.5-1.5B-Instruct-Q4_K_M.gguf

    proxy: http://127.0.0.1:8999

Installation

1. Create a configuration file, see config.example.yaml
2. Download a release appropriate for your OS and architecture.
   • Note: Windows currently untested.
3. Run the binary with llama-swap --config path/to/config.yaml

Monitoring Logs

The /logs endpoint is available to monitor what llama-swap is doing. It will send the last 10KB of logs. Useful for monitoring the output of llama-server. It also supports streaming of logs.

Usage:

# basic, sends up to the last 10KB of logs
curl http://host/logs'

# add `stream` to stream new logs as they come in
curl -Ns 'http://host/logs?stream'

# add `skip` to skip history (only useful if used with stream)
curl -Ns 'http://host/logs?stream&skip'

# will output nothing :)
curl -Ns 'http://host/logs?skip'

Systemd Unit Files

Use this unit file to start llama-swap on boot. This is only tested on Ubuntu.

/etc/systemd/system/llama-swap.service

[Unit]
Description=llama-swap
After=network.target

[Service]
User=nobody

# set this to match your environment
ExecStart=/path/to/llama-swap --config /path/to/llama-swap.config.yml

Restart=on-failure
RestartSec=3
StartLimitBurst=3
StartLimitInterval=30

[Install]
WantedBy=multi-user.target

Building from Source

1. Install golang for your system
2. run make clean all
3. binaries will be built into build/ directory