🍺 Heretek-AI Homebrew Tap

Zero-dependency, turnkey Homebrew distribution tap for high-performance LLM inference engines:
CachyLLama, ROCmFPX, and llama-ai.

---

⚡ Quick Start

1. Tap this Repository

brew tap Heretek-AI/tap

2. Install Packages

CachyLLama (Persistent KV Cache & MoE Residency)

# Universal Vulkan RADV (Runs on any Linux AMD APU, dGPU, Intel, or NVIDIA GPU):
brew install cachy-llama

# Or install dedicated ROCm 7 GPU acceleration for your hardware:
brew install cachy-llama --with-rocm-gfx1151  # AMD Strix Halo (Radeon 8060S / 128GB)
brew install cachy-llama --with-rocm-gfx1150  # AMD Strix Point (Radeon 890M / 880M)
brew install cachy-llama --with-rocm-gfx120X  # AMD RDNA4 (RX 9070 XT / 9070)
brew install cachy-llama --with-rocm-gfx110X  # AMD RDNA3 (RX 7900 / 7800, Radeon 780M)
brew install cachy-llama --with-rocm-gfx103X  # AMD RDNA2 / Steam Deck
brew install cachy-llama --with-rocm-gfx90a   # AMD Instinct MI210 / MI250X

# macOS (Apple Silicon M1/M2/M3/M4 Metal):
brew install cachy-llama

ROCmFPX (Low-bit Quantized ROCm 7 Inference Stack)

# Default: AMD Strix Halo (gfx1151):
brew install rocmfpx

# Or choose specific GPU architecture:
brew install rocmfpx --with-gfx1150  # AMD Strix Point (Radeon 890M)
brew install rocmfpx --with-gfx120X  # AMD RDNA4 (RX 9070 XT)
brew install rocmfpx --with-gfx110X  # AMD RDNA3 (RX 7900, Radeon 780M)
brew install rocmfpx --with-gfx103X  # AMD RDNA2 / Steam Deck
brew install rocmfpx --with-gfx90a   # AMD Instinct MI210 / MI250X
brew install rocmfpx --with-gfx908   # AMD Instinct MI100

llama-ai (Turnkey APU Runner & Optimistic Solver)

brew install llama-ai

---

🎯 Supported Hardware Matrix

GPU / APU Target	GFX Target	Target Hardware	cachy-llama Option	rocmfpx Option
RDNA3.5 (Strix Halo)	gfx1151	Ryzen AI MAX+ Pro 395, Radeon 8060S (128GB)	--with-rocm-gfx1151	(Default)
RDNA3.5 (Strix Point)	gfx1150	Ryzen AI 9 HX 370 / 365, Radeon 890M / 880M	--with-rocm-gfx1150	--with-gfx1150
RDNA4	gfx120X	AMD Radeon RX 9070 XT, RX 9070, RX 9060 XT	--with-rocm-gfx120X	--with-gfx120X
RDNA3	gfx110X	Radeon 780M / 760M, RX 7900 XTX / XT / 7800 XT	--with-rocm-gfx110X	--with-gfx110X
RDNA2	gfx103X	Steam Deck (Van Gogh), 680M, RX 6800 / 6700 XT	--with-rocm-gfx103X	--with-gfx103X
CDNA / CDNA2	gfx90a	AMD Instinct MI250X, MI210	--with-rocm-gfx90a	--with-gfx90a
CDNA1	gfx908	AMD Instinct MI100	--with-rocm-gfx908	--with-gfx908
Universal Vulkan	RADV	Universal Linux APU/dGPU Support	(Linux Default)	N/A
Apple Silicon	Metal	Apple M1 / M2 / M3 / M4 (Pro / Max / Ultra)	(macOS Default)	N/A

---

🚀 Running Inference & Background Services

Command-Line Inference

All commands are installed with both upstream aliases and namespaced identifiers:

# Chat in terminal with GPU offload:
llama-cli -m /path/to/model.gguf -ngl 99 -p "Hello world!"
# Or namespaced:
cachy-llama-cli -m /path/to/model.gguf -ngl 99
rocmfpx-cli -m /path/to/model.gguf -ngl 99

# Quantize models to ROCmFPX formats:
rocmfpx-quantize model-BF16.gguf model-Q7.gguf Q7_0_ROCMFPX

Background Daemon Service (brew services)

Run OpenAI-compatible local server as a background service:

# Start background server daemon on port 8080:
brew services start cachy-llama
# or
brew services start rocmfpx

# Check status:
brew services list

# Stop background server:
brew services stop cachy-llama

---

🛠️ Upstream Repositories & Automation

• Automated Builders:
  • Heretek-AI/CachyLLama-BUILDER
  • Heretek-AI/ROCmFPX-BUILDER
• Upstream Engines:
  • fewtarius/CachyLLama
  • ciru-ai/ROCmFPX
  • fewtarius/llama-ai
  • ggml-org/llama.cpp

---

📄 License

Licensed under the MIT License.