Serve LLMs with a single binary.

A Rust-native LLM serving engine. One binary, no Python runtime.

中文说明

See it in action

One local model. Three Orchestral terminals inspecting code, fixing bugs, and running tests concurrently.

Watch the English demo · 50 seconds · 8× speed.

curl -fsSL https://ferrum.pandaailabs.com/install.sh | sh
curl -fsSL https://orch.pandaailabs.com/install.sh | sh

irm https://ferrum.pandaailabs.com/install.ps1 | iex
irm https://orch.pandaailabs.com/install.ps1 | iex

ferrum serve --model unsloth/Qwen3.5-9B-GGUF

orchestral --base-url http://127.0.0.1:8000/v1 --no-auth

curl -fsSL https://ferrum.pandaailabs.com/install.sh | sh -s -- --version 0.10.0 --backend metal
curl -fsSL https://orch.pandaailabs.com/install.sh | sh -s -- --version 0.3.1
export PATH="$HOME/.local/bin:$PATH"
ferrum --version
orchestral --version

ferrum serve \
  --model unsloth/Qwen3.5-9B-GGUF@3885219b6810b007914f3a7950a8d1b469d598a5 \
  --gguf-file Qwen3.5-9B-Q4_K_M.gguf \
  --served-model-name Qwen3.5-9B \
  --backend metal \
  --numerical-profile qwen3_5.f32-master \
  --host 127.0.0.1 --port 8001 \
  --max-model-len 24576 \
  --max-num-seqs 3 \
  --max-num-batched-tokens 3072 \
  --scheduler-prefill-step-chunk 1024 \
  --scheduler-active-decode-prefill-chunk 256 \
  --enable-prefix-cache \
  --runtime-memory-budget-bytes 21474836480 \
  --prefix-rendezvous-max-wait-ms 180000

orchestral --base-url http://127.0.0.1:8001/v1 --no-auth doctor --check-connection

cd /path/to/project-a
orchestral --base-url http://127.0.0.1:8001/v1 --no-auth

cd /path/to/project-b
orchestral --base-url http://127.0.0.1:8001/v1 --no-auth

cd /path/to/project-c
orchestral --base-url http://127.0.0.1:8001/v1 --no-auth

Vision

Make high-performance LLM serving simple to deploy and operate.

Quick Start

Install the latest stable Ferrum on macOS Apple Silicon or Linux x86_64:

curl -fsSL https://ferrum.pandaailabs.com/install.sh | sh

The installer verifies release checksums and adds ~/.local/bin to your shell's PATH. Open a new terminal afterward. Homebrew and manual installation are also available.

Windows x64 supports CPU inference and compatible NVIDIA sm89 GPUs. Install from PowerShell:

irm https://ferrum.pandaailabs.com/install.ps1 | iex

The script verifies the setup checksum, installs for the current user, and adds Ferrum to PATH, including the current PowerShell session.

Installers select CPU when a supported GPU is unavailable. Package downloads use Cloudflare CDN, retain SHA256 verification, and fall back to GitHub if needed. Running the same command again installs the latest formal release.

Inspect the installed binary before downloading weights:

ferrum --version
ferrum --help
ferrum doctor

Run a model

With Ferrum 0.9.0 or later, use the same GGUF model on macOS, Linux, and Windows. Ferrum automatically selects the available backend; both Metal and CUDA support this Q4_K_M example.

ferrum run qwen3.5:4b-q4_k_m --disable-thinking

The first run downloads about 2.55 GiB. Download time depends on your route to Hugging Face; the CLI displays download progress. On a 6 GB GPU, append --max-model-len 2048 --max-num-seqs 1 to either run or serve to limit the context and active sequences.

Serve an API

The server command is also the same on all three platforms:

ferrum serve --model qwen3.5:4b-q4_k_m --served-model-name ferrum --disable-thinking --port 8000

Send a request from another terminal. On macOS or Linux:

curl http://localhost:8000/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model":"ferrum","messages":[{"role":"user","content":"Reply with a short hello from Ferrum."}],"max_tokens":32}'

In Windows PowerShell:

$body = @{ model = 'ferrum'; messages = @(@{ role = 'user'; content = 'Reply with a short hello from Ferrum.' }); max_tokens = 32 } | ConvertTo-Json -Depth 4
Invoke-RestMethod http://localhost:8000/v1/chat/completions -Method Post -ContentType 'application/json' -Body $body

Ferrum does not silently select a model. run requires MODEL, and serve requires either --model or an intentional default_model in ferrum.toml.

A working request returns HTTP 200 with a non-empty assistant response. Ferrum uses the model's context limit unless --max-model-len is set explicitly; any explicit limit must fit the rendered input plus the requested output budget.

The examples use --disable-thinking so the first response is short and direct. Omit the flag to preserve the model template's default reasoning behavior; an HTTP request can override the server default with chat_template_kwargs.enable_thinking, Chat reasoning_effort, or Responses reasoning.effort. See reasoning control behavior for model support and compatibility details.

GET /v1/models also exposes optional reasoning metadata. A supported thinking switch reports its effective default in thinking.default_enabled; explicitly declared effort levels appear in supported_efforts. A thinking switch alone does not imply low/medium/high levels. Omitted effort metadata means unknown support.

ferrum doctor <MODEL> resolves an alias and prints the next run and serve commands without downloading the model or starting an inference engine.

For vNext execution, run and serve share this optional ferrum.toml setting in the working directory:

[runtime]
reusable_execution_preparation = "auto" # auto, startup, on_demand

auto uses bounded on-demand preparation on runtimes that declare support (currently CUDA). A new shape first executes normally; later occurrences can prepare and reuse a device program. First-use latency can therefore be higher than steady-state latency. startup prepares the configured matrix before the server becomes ready. Other backends retain their existing behavior; explicitly requesting unsupported on_demand reports an error. Set reusable_execution = false to disable device-program preparation. These options do not change request admission, queuing, or the model's numerical profile.

KV cache precision

Ferrum v0.11.0 accepts --kv-dtype int8 in both run and serve. FP16 remains the default. INT8 requires supported vNext standard causal attention on Metal or portable CUDA; unsupported combinations report an error.

ferrum run unsloth/Qwen3.5-9B-GGUF --kv-dtype int8 --disable-thinking
ferrum serve --model unsloth/Qwen3.5-9B-GGUF --kv-dtype int8 --disable-thinking

This reduces attention KV storage, including its quantization scales. Model weights and fixed recurrent state retain their existing sizes. Inspect /health → kv_storage to confirm the selected format. Whole-model checkpoint restore requires support for every model state; resending conversation history recomputes the input.

Features

• ferrum run and ferrum serve in one Rust binary.
• OpenAI-compatible Chat Completions and stateless Responses APIs, streaming, tools, and structured output.
• Apple Silicon Metal and NVIDIA CUDA from the same runtime.
• Continuous batching, paged KV cache, prefix cache, and typed admission control.
• Optional 8-bit KV storage for supported vNext Metal and portable CUDA attention paths; FP16 remains the default.
• GGUF on Metal and CUDA; CUDA also supports GPTQ/safetensors.
• Ferrum covers language-model inference only. Supported models include Qwen3.5 4B, Qwen3.5 35B-A3B, Qwen3 30B-A3B, and Llama 3.1 8B dense.

Performance Snapshot

Latest R2 development ferrum serve checkpoint. The first three rows use 64-token input / 128-token output on Metal and 256 / 128 on CUDA. Values are mean tok/s with the 95% confidence-interval half-width across three repeats.

c is active server concurrency. The first three rows completed 100 requests × 3 repeats with zero errors.

OpenAI-Compatible API

Ferrum supports:

• chat completions and streaming usage
• stateless Responses text, reasoning replay, streaming, usage, and caller-owned function/namespace tool loops
• function tools with auto, none, required, or a named function
• json_object and strict json_schema structured output
• multi-turn sessions, prefix cache, and session cache
• typed concurrency, memory, and scheduler controls

See OpenAI API compatibility for the exact request contract and cache product controls for prefix and session caching.

Installation

Windows 0.8.9 and later can also be installed by downloading ferrum-<version>-windows-x86_64-cuda-sm89-setup.exe and its .sha256 file from Releases, verifying the checksum, and running setup. It installs under %LOCALAPPDATA%\Programs\Ferrum and adds the current-user PATH; open a new terminal after a manual setup install. The package includes CUDA and VC runtimes. It requires a compatible NVIDIA sm89 GPU and driver (551.78 or later); it does not install the system driver or include models. CUDA Toolkit, Rust, and build tools are not needed. Ferrum remains a command-line application with run and serve, without a GUI or background service.

To upgrade Windows, rerun the same PowerShell install command or the newer setup. Existing sessions keep running their original version; new launches use the updated version. Models, configuration, and existing version directories are preserved. Restart an existing server when you want it to use the update.

The macOS/Linux one-line installer selects Metal on Apple Silicon. On Linux it selects CUDA for compatible sm89 GPUs when the driver, CUDA 12.4 and NCCL runtimes can load, and otherwise selects CPU. You can require a backend or install a specific version:

curl -fsSL https://ferrum.pandaailabs.com/install.sh | sh -s -- --backend cuda
curl -fsSL https://ferrum.pandaailabs.com/install.sh | sh -s -- --version 0.11.0

To upgrade an installation made with the script, rerun the original install command. If the selected version and backend are already installed and verify successfully, the script checks the small release checksum files and skips the package download. It keeps existing version directories and switches the entry point to the verified new binary. Running sessions continue using their current version; new launches use the new version. Restart an existing server when you want it to use the update. Models and configuration are preserved.

For immediate PATH setup in the current terminal:

. "$HOME/.local/share/ferrum/installer/env"

For Homebrew installations, use brew upgrade for the installed formula. Homebrew 6 needs both formula definitions trusted for its conflict check. Review them before running the trust command; older Homebrew versions can skip it. See Homebrew's trust documentation.

# Homebrew 6: trust the reviewed formula definitions
brew trust --formula sizzlecar/ferrum/ferrum sizzlecar/ferrum/ferrum-cuda

# macOS Apple Silicon Metal
brew install sizzlecar/ferrum/ferrum

# Linux x86_64 CUDA sm89
brew install sizzlecar/ferrum/ferrum-cuda

Prebuilt tarballs from the latest stable release:

# Linux x86_64 CUDA sm89
curl --fail --location --remote-name https://github.com/sizzlecar/ferrum-infer-rs/releases/latest/download/ferrum-linux-x86_64-cuda-sm89.tar.gz
curl --fail --location --remote-name https://github.com/sizzlecar/ferrum-infer-rs/releases/latest/download/ferrum-linux-x86_64-cuda-sm89.tar.gz.sha256
sha256sum --check ferrum-linux-x86_64-cuda-sm89.tar.gz.sha256
tar -xzf ferrum-linux-x86_64-cuda-sm89.tar.gz
LD_LIBRARY_PATH=/usr/local/cuda/lib64:${LD_LIBRARY_PATH:-} ./ferrum --version

# macOS Apple Silicon Metal
curl --fail --location --remote-name https://github.com/sizzlecar/ferrum-infer-rs/releases/latest/download/ferrum-macos-aarch64.tar.gz
curl --fail --location --remote-name https://github.com/sizzlecar/ferrum-infer-rs/releases/latest/download/ferrum-macos-aarch64.tar.gz.sha256
shasum -a 256 --check ferrum-macos-aarch64.tar.gz.sha256
tar -xzf ferrum-macos-aarch64.tar.gz
./ferrum --version

Install the latest Metal build from crates.io:

# macOS Apple Silicon Metal
cargo install ferrum-cli --locked --features metal

The official prebuilt Linux CUDA asset targets sm89. Linux CUDA installation requires a compatible NVIDIA driver, CUDA runtime, and NCCL runtime on the target host. CUDA source builds also require Ferrum's matching native-operator set, so use the prebuilt CUDA tarball or Homebrew formula for the supported install path.

Architecture

• Contracts: ferrum-types, ferrum-interfaces
• Execution: ferrum-engine, ferrum-scheduler, ferrum-kv, ferrum-sampler
• Models and compute: ferrum-models, ferrum-kernels, ferrum-native-ops, ferrum-quantization
• Product surface: ferrum-cli, ferrum-server, ferrum-tokenizer
• Validation: ferrum-bench-core, ferrum-testkit

Development notes: numerical execution profiles (中文).

License

MIT