feat: add one-line install script for Linux/macOS

.github/workflows/install-script.yaml

@@ -0,0 +1,52 @@
+name: Install Script CI
+
+on:
+  push:
+    branches:
+      - '*'
+    paths:
+      - 'scripts/install.sh'
+      - '.github/workflows/install-script.yaml'
+  pull_request:
+    types: ['opened', 'reopened', 'synchronize']
+    paths:
+      - 'scripts/install.sh'
+      - '.github/workflows/install-script.yaml'
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  shellcheck:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Check out code
+        uses: actions/checkout@v4
+      - name: Run shellcheck
+        run: shellcheck scripts/install.sh
+
+  install:
+    name: install (${{ matrix.os }}, cpu)
+    runs-on: ${{ matrix.os }}
+    strategy:
+      fail-fast: false
+      matrix:
+        os: [ "ubuntu-latest", "macos-latest" ]
+    steps:
+      - name: Check out code
+        uses: actions/checkout@v4
+
+      # Exercise the installer end-to-end. CPU backend keeps the download small
+      # and works on runners without a GPU (ignored on macOS).
+      - name: Run installer (CPU backend)
+        env:
+          XINFERENCE_BACKEND: cpu
+          XINFERENCE_HOME_DIR: ${{ runner.temp }}/xinf-venv
+        run: sh scripts/install.sh
+
+      - name: Verify the command is installed and runnable
+        run: |
+          export PATH="$HOME/.local/bin:$PATH"
+          xinference-local --help
+          xinference --version

README.md

@@ -161,6 +161,12 @@ Install Xinference by using pip as follows. (For more options, see [Installation
 pip install "xinference[all]"
 ```
 
+On Linux/macOS, a [one-line install script](https://inference.readthedocs.io/en/latest/getting_started/installation.html#one-line-install-script-linux-macos) can bootstrap the **base framework** into an isolated [uv](https://docs.astral.sh/uv/)-managed environment (no model backends installed; not a replacement for the `pip install` above):
+
+```bash
+curl -fsSL https://raw.githubusercontent.com/xorbitsai/inference/main/scripts/install.sh | sh
+```
+
 To start a local instance of Xinference, run the following command:
 
 ```bash

doc/source/getting_started/installation.rst

@@ -13,6 +13,58 @@ If you aim to serve all supported models, you can install all the necessary depe
 
    Due to irreconcilable package dependency conflicts between vLLM and sglang, we have removed sglang from the all extra. If you want to use sglang, please install it separately via ``pip install 'xinference[sglang]'``.
 
+.. _one_line_install:
+
+One-line install script (Linux/macOS)
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+As a convenience, Linux and macOS users can bootstrap Xinference with a single command::
+
+   curl -fsSL https://raw.githubusercontent.com/xorbitsai/inference/main/scripts/install.sh | sh
+
+The script installs the base framework into a dedicated, isolated virtualenv managed by
+`uv <https://docs.astral.sh/uv/>`_ (so it does not touch your system Python or hit PEP 668),
+selects a hardware-appropriate PyTorch build, and links the ``xinference`` commands into
+``~/.local/bin``. It installs ``uv`` first if it is not already present.
+
+.. important::
+
+   This installs **only the base framework** (CLI, server, and Web UI). It does **not**
+   install any model-serving backend (transformers, vLLM, sglang, MLX, llama.cpp, image,
+   audio, video). It is a quick way to get the ``xinference`` commands, **not** a replacement
+   for the functional ``pip install "xinference[all]"`` install above. To serve models, add
+   an extra afterwards (e.g. ``XINFERENCE_EXTRAS=transformers``) or use the pip install.
+
+The script's behavior can be tuned with environment variables:
+
+.. list-table::
+   :header-rows: 1
+   :widths: 25 75
+
+   * - Variable
+     - Meaning
+   * - ``XINFERENCE_BACKEND``
+     - PyTorch backend: ``auto`` (default, detects GPU/driver), ``cpu``, a specific CUDA
+       build such as ``cu128``, ``rocm`` (AMD, Linux), or ``xpu`` (Intel GPU, Linux).
+       Ignored on macOS, which ships a single universal wheel.
+   * - ``XINFERENCE_EXTRAS``
+     - Optional extras to include, e.g. ``all`` or ``vllm,transformers``. Some extras
+       (e.g. vLLM) require Linux + CUDA.
+   * - ``XINFERENCE_VERSION``
+     - Pin a specific version, e.g. ``1.8.1``.
+   * - ``XINFERENCE_PYTHON``
+     - Python version for the virtualenv (default ``3.12``).
+   * - ``XINFERENCE_HOME_DIR``
+     - Install location (default ``~/.xinference/venv``).
+
+For example, to install a CPU-only build with the transformers backend::
+
+   curl -fsSL https://raw.githubusercontent.com/xorbitsai/inference/main/scripts/install.sh | XINFERENCE_BACKEND=cpu XINFERENCE_EXTRAS=transformers sh
+
+.. note::
+
+   Windows is not covered by the one-line installer. Install with ``pip`` in a virtualenv
+   instead (see above).
+
 
 Several usage scenarios require special attention.
 

scripts/install.sh

@@ -0,0 +1,168 @@
+#!/bin/sh
+# Xinference installer
+#
+#   curl -fsSL https://raw.githubusercontent.com/xorbitsai/inference/main/scripts/install.sh | sh
+#
+# Installs the *base* Xinference framework (CLI + server + Web UI) into a
+# dedicated, isolated virtualenv managed by uv, and links its commands into
+# ~/.local/bin. Because Xinference depends on PyTorch, the installer selects a
+# hardware-appropriate PyTorch build (CPU / CUDA / ROCm / Intel XPU) via uv's
+# --torch-backend, which requires the `uv pip` interface (and therefore a
+# managed venv rather than `uv tool install`).
+#
+# IMPORTANT: this installs only the base framework. Model-serving backends
+# (transformers, vLLM, sglang, MLX, llama.cpp, image/audio/video, ...) are NOT
+# installed. To serve models, install the relevant extra afterwards, or use the
+# functional Quick Start: pip install "xinference[all]".
+#
+# Options (environment variables):
+#   XINFERENCE_BACKEND   PyTorch backend to install. Default: auto.
+#                        auto  - detect GPU/driver and pick the best build
+#                        cpu   - CPU-only PyTorch
+#                        cu128, cu126, ... - a specific CUDA build
+#                        rocm  - AMD ROCm build (Linux only)
+#                        xpu   - Intel GPU build (Linux only)
+#                        On macOS this is ignored (a single universal wheel).
+#   XINFERENCE_EXTRAS    optional extras, e.g. XINFERENCE_EXTRAS=all or
+#                        XINFERENCE_EXTRAS=vllm,transformers (default: none).
+#                        Some extras (e.g. vllm) require Linux + CUDA.
+#   XINFERENCE_VERSION   pin a version, e.g. XINFERENCE_VERSION=1.8.1
+#   XINFERENCE_PYTHON    Python version for the venv (default: 3.12)
+#   XINFERENCE_HOME_DIR  install location (default: ~/.xinference/venv)
+#
+# Prefer to manage the environment yourself? The equivalent manual install is:
+#   uv venv --python 3.12 .venv && . .venv/bin/activate
+#   uv pip install --torch-backend auto xinference
+set -eu
+
+ORIG_PATH="$PATH"
+
+APP="xinference"
+SERVE_CMD="xinference-local"
+PORT="9997"
+BIN_DIR="$HOME/.local/bin"
+VENV_DIR="${XINFERENCE_HOME_DIR:-$HOME/.xinference/venv}"
+PY_VERSION="${XINFERENCE_PYTHON:-3.12}"
+# Commands exposed by the package; symlinked from the venv into BIN_DIR.
+CMDS="xinference xinference-local xinference-supervisor xinference-worker"
+
+info() { printf '\033[1;34m==>\033[0m %s\n' "$1"; }
+warn() { printf '\033[1;33mwarning:\033[0m %s\n' "$1" >&2; }
+err() {
+  printf '\033[1;31merror:\033[0m %s\n' "$1" >&2
+  exit 1
+}
+
+# ---------------------------------------------------------------------------
+# Platform detection
+# ---------------------------------------------------------------------------
+os="$(uname -s)"
+arch="$(uname -m)"
+case "$os" in
+  Linux) platform="linux" ;;
+  Darwin)
+    platform="macos"
+    case "$arch" in
+      arm64) ;;  # Apple silicon: fully supported
+      x86_64)
+        warn "Intel-based macOS detected. Recent PyTorch releases no longer ship"
+        warn "Intel-mac wheels, so installation may fail or pin an old torch."
+        ;;
+    esac
+    ;;
+  *) err "Unsupported OS '$os'. On Windows, install with: pip install \"${APP}[all]\" (in a virtualenv)." ;;
+esac
+
+# ---------------------------------------------------------------------------
+# Preflight: warn about an existing installation / environment conflicts
+# ---------------------------------------------------------------------------
+# shellcheck disable=SC2030  # the PATH override is deliberately subshell-local
+existing="$( ( PATH="$ORIG_PATH"; command -v "$SERVE_CMD" ) 2>/dev/null || true )"
+if [ -n "$existing" ]; then
+  warn "An existing '$SERVE_CMD' was found at: $existing"
+  warn "This installer creates an isolated environment and will shadow it via $BIN_DIR."
+fi
+if [ -n "${VIRTUAL_ENV:-}" ] || [ -n "${CONDA_PREFIX:-}" ]; then
+  warn "You are inside an active virtualenv/conda environment."
+  warn "uv installs into a separate isolated env; packages from the active"
+  warn "environment are NOT reused. To install into the current environment"
+  warn "instead, run: pip install \"${APP}[all]\""
+fi
+
+# ---------------------------------------------------------------------------
+# Ensure uv is available (isolates the install; avoids system-Python/PEP 668).
+# ---------------------------------------------------------------------------
+if ! command -v uv >/dev/null 2>&1; then
+  info "Installing uv (Python package/tool manager)..."
+  curl -LsSf https://astral.sh/uv/install.sh | sh
+  for d in "$HOME/.local/bin" "$HOME/.cargo/bin"; do
+    # shellcheck disable=SC2031  # this runs in the main shell, not a subshell
+    [ -d "$d" ] && PATH="$d:$PATH"
+  done
+  export PATH
+fi
+command -v uv >/dev/null 2>&1 || err "uv not found on PATH after install; open a new shell and re-run."
+
+# ---------------------------------------------------------------------------
+# Resolve the PyTorch backend
+# ---------------------------------------------------------------------------
+backend="${XINFERENCE_BACKEND:-auto}"
+if [ "$platform" = "macos" ]; then
+  # macOS ships a single PyTorch wheel; --torch-backend does not apply.
+  torch_arg=""
+  [ "$backend" != "auto" ] && warn "XINFERENCE_BACKEND='$backend' ignored on macOS."
+else
+  torch_arg="--torch-backend=$backend"
+fi
+
+# Build the package spec: name[extras]==version, with each part optional.
+spec="$APP"
+if [ -n "${XINFERENCE_EXTRAS:-}" ]; then
+  spec="${spec}[${XINFERENCE_EXTRAS}]"
+fi
+if [ -n "${XINFERENCE_VERSION:-}" ]; then
+  version="${XINFERENCE_VERSION#v}"  # strip a leading 'v' (v1.8.1 -> 1.8.1)
+  [ -n "$version" ] || err "XINFERENCE_VERSION='$XINFERENCE_VERSION' is not a valid version."
+  spec="$spec==$version"
+fi
+
+# ---------------------------------------------------------------------------
+# Create the venv and install
+# ---------------------------------------------------------------------------
+info "Creating environment at $VENV_DIR (Python $PY_VERSION)..."
+uv venv --python "$PY_VERSION" "$VENV_DIR"
+
+info "Installing $spec (torch backend: ${backend})..."
+# shellcheck disable=SC2086  # torch_arg is intentionally unquoted (may be empty)
+VIRTUAL_ENV="$VENV_DIR" uv pip install --python "$VENV_DIR/bin/python" $torch_arg "$spec"
+
+# ---------------------------------------------------------------------------
+# Link commands into ~/.local/bin
+# ---------------------------------------------------------------------------
+mkdir -p "$BIN_DIR"
+for c in $CMDS; do
+  if [ -x "$VENV_DIR/bin/$c" ]; then
+    ln -sf "$VENV_DIR/bin/$c" "$BIN_DIR/$c"
+  fi
+done
+
+printf '\n'
+info "Installed the base Xinference framework. Next steps:"
+printf '\n'
+printf '  Start the server:   %s\n' "$SERVE_CMD"
+printf '  Open the Web UI:    http://127.0.0.1:%s\n' "$PORT"
+printf '\n'
+if [ -z "${XINFERENCE_EXTRAS:-}" ]; then
+  printf '  No model backend was installed. Add one when you need it, e.g.:\n'
+  printf '    uv pip install --python "%s/bin/python" "%s[transformers]"\n' "$VENV_DIR" "$APP"
+  printf '  (transformers / vllm [Linux+CUDA] / mlx [Apple silicon] / llama_cpp / all)\n'
+  printf '\n'
+fi
+
+# Warn if BIN_DIR is not on the parent shell's PATH. Use a subshell to scope the
+# PATH override: a leading `PATH=... command -v` assignment on a shell built-in
+# is not clearly specified by POSIX, so the subshell form is unambiguous.
+if ! ( PATH="$ORIG_PATH"; command -v "$SERVE_CMD" ) >/dev/null 2>&1; then
+  warn "'$BIN_DIR' does not appear to be on your PATH in this shell."
+  warn "Open a new terminal, or run: export PATH=\"$BIN_DIR:\$PATH\""
+fi