### First Run Example

Source: https://github.com/kaden-schutt/hipfire/blob/master/docs/GETTING_STARTED.md

Pulls a model and runs an inference query to demonstrate the first use of Hipfire.

```bash
hipfire pull qwen3.5:4b                         # ~2.6 GB download
hipfire run  qwen3.5:4b "Explain FFT in one line"
```

--------------------------------

### Quick Setup Example

Source: https://github.com/kaden-schutt/hipfire/blob/master/docs/CLI.md

Example demonstrating the workflow of quantizing a model, generating its sidecar, and configuring CASK.

```bash
hipfire quantize ./my-model/ --format mq4 -o my-finetune.mq4
hipfire sidecar-gen my-finetune.mq4 --corpus /path/to/corpus.txt
hipfire config cask-profile balanced
```

--------------------------------

### Windows Installation

Source: https://github.com/kaden-schutt/hipfire/blob/master/docs/GETTING_STARTED.md

Installs Hipfire on Windows using PowerShell, downloading the daemon and setting up the CLI.

```powershell
irm https://raw.githubusercontent.com/Kaden-Schutt/hipfire/master/scripts/install.ps1 | iex
```

--------------------------------

### Developer workflow: Setup

Source: https://github.com/kaden-schutt/hipfire/blob/master/CONTRIBUTING.md

Commands to clone the repository and build the project with specific features and examples.

```bash
git clone https://github.com/Kaden-Schutt/hipfire
cd hipfire
cargo build --release --features deltanet --example daemon -p hipfire-runtime
cargo build --release --features deltanet --example test_kernels -p hipfire-runtime
cargo build --release -p hipfire-quantize
```

--------------------------------

### Source Build

Source: https://github.com/kaden-schutt/hipfire/blob/master/docs/GETTING_STARTED.md

Builds Hipfire from source after cloning the repository.

```bash
git clone https://github.com/Kaden-Schutt/hipfire
cd hipfire
cargo build --release --features deltanet --example daemon -p hipfire-runtime
cargo build --release -p hipfire-quantize
```

--------------------------------

### README.md Addition - Quick Start

Source: https://github.com/kaden-schutt/hipfire/blob/master/docs/superpowers/specs/2026-05-07-nixos-packaging-design.md

Quick start commands for using hipfire with Nix.

```bash
nix develop github:Kaden-Schutt/hipfire  # dev shell
nix build github:Kaden-Schutt/hipfire    # build package

```

--------------------------------

### Quantization command example

Source: https://github.com/kaden-schutt/hipfire/blob/master/docs/plans/mq-lloyd-batched-prefill-followup.md

Example command to produce MQ3-Lloyd quantized models.

```bash
hipfire-quantize --format mq3-lloyd
```

--------------------------------

### Verify Installation

Source: https://github.com/kaden-schutt/hipfire/blob/master/docs/GETTING_STARTED.md

Runs a diagnostic command to confirm ROCm version, HIP runtime, GPU arch, VRAM, and kernel blob compatibility.

```bash
hipfire diag
```

--------------------------------

### Background Daemon Control

Source: https://github.com/kaden-schutt/hipfire/blob/master/docs/GETTING_STARTED.md

Commands to start, run inference with, and stop the background Hipfire daemon.

```bash
hipfire serve -d                                 # detaches, pre-warms default_model
hipfire run qwen3.5:4b "..."                     # auto-routes through HTTP, skips cold-start
hipfire stop                                     # graceful shutdown
```

--------------------------------

### Build Command

Source: https://github.com/kaden-schutt/hipfire/blob/master/docs/plans/gfx10_mq3_prefill.md

Standard build command for the release daemon example with deltanet features.

```bash
cargo build --release --example daemon --features deltanet
```

--------------------------------

### Configuration

Source: https://github.com/kaden-schutt/hipfire/blob/master/docs/GETTING_STARTED.md

Commands to access the interactive configuration TUI and set per-model overrides.

```bash
hipfire config                                   # interactive TUI for global keys
hipfire config qwen3.5:9b                        # per-model overlay
```

--------------------------------

### Full NixOS Module Example

Source: https://github.com/kaden-schutt/hipfire/blob/master/docs/superpowers/plans/2026-05-07-nixos-packaging.md

A comprehensive NixOS configuration for hipfire, including global settings, per-model overrides, and environment variables.

```nix
services.hipfire = {
  enable = true;
  gpuTargets = [ "gfx1100" ];

  # Global config (written to config.json)
  settings = {
    temperature = 0.3;
    top_p = 0.8;
    max_tokens = 512;
    kv_cache = "asym3";
    dflash_mode = "auto";
    port = 11435;
    idle_timeout = 300;
    default_model = "qwen3.5:9b";
  };

  # Per-model overrides (written to per_model_config.json)
  perModelSettings = {
    "qwen3.5:27b" = {
      max_seq = 16384;
      kv_cache = "q8";
      dflash_mode = "on";
    };
  };

  # Extra env vars (highest precedence)
  environment = {
    HIPFIRE_NORMALIZE_PROMPT = "1";
  };

  modelDir = "/var/lib/hipfire/models";
};
```

--------------------------------

### Minimal NixOS Module Example

Source: https://github.com/kaden-schutt/hipfire/blob/master/docs/superpowers/plans/2026-05-07-nixos-packaging.md

A minimal NixOS configuration to enable hipfire as a service.

```nix
# flake.nix
{
  inputs = {
    nixpkgs.url = "github:NixOS/nixpkgs/nixos-unstable";
    hipfire.url = "github:Kaden-Schutt/hipfire";
  };

  outputs = { nixpkgs, hipfire, ... }:
    {
    nixosConfigurations.myhost = nixpkgs.lib.nixosSystem {
      system = "x86_64-linux";
      modules = [
        hipfire.nixosModules.default
        {
          nixpkgs.overlays = [ hipfire.overlays.default ];
          services.hipfire.enable = true;
          services.hipfire.gpuTargets = [ "gfx1100" ];
        }
      ];
    };
  };
}
```

--------------------------------

### Generate KV Cache Sidecar

Source: https://github.com/kaden-schutt/hipfire/blob/master/docs/GETTING_STARTED.md

Generates a CASK sidecar for custom or quantized models before configuring eviction.

```bash
hipfire sidecar-gen ~/models/my-finetune.mq4 --corpus corpus.txt
hipfire config cask-profile balanced
```

--------------------------------

### Enable KV Cache Eviction

Source: https://github.com/kaden-schutt/hipfire/blob/master/docs/GETTING_STARTED.md

Enables CASK-based KV cache eviction for long-context prompts by setting a profile.

```bash
hipfire config cask-profile balanced   # or conservative / aggressive-vram
```

--------------------------------

### Force Kernel Recompile (Windows)

Source: https://github.com/kaden-schutt/hipfire/blob/master/docs/GETTING_STARTED.md

Forces a fresh compile of the full kernel set for a specific GPU architecture on Windows.

```powershell
.\scripts\compile-kernels.ps1 gfx1100   # or your arch
```

--------------------------------

### imatrix collection example

Source: https://github.com/kaden-schutt/hipfire/blob/master/docs/plans/qwen35-mq4-quality-gap.md

Example command for collecting imatrix data.

```bash
imatrix_collect --model path/to/bf16/model --corpus path/to/corpus.txt --output path/to/output.imatrix
```

--------------------------------

### Install Dependencies

Source: https://github.com/kaden-schutt/hipfire/blob/master/cli/README.md

Installs the project dependencies using Bun.

```bash
bun install
```

--------------------------------

### llama.cpp benchmark command

Source: https://github.com/kaden-schutt/hipfire/blob/master/docs/perf-checkpoints/2026-04-27-mmq-prefill-experiment-report.md

The reference comparison was llama.cpp on the same Strix Halo machine.

```bash
llama-bench \
  -m /home/kotdath/omp/personal/amd-strix-halo-toolboxes/models/Qwen3.5-9B-Q4_K_M.gguf \
  -ngl 99 -p 2048 -n 1 -r 3 -ctk q8_0 -ctv q8_0 -fa 1
```