PS5LM: Native GGUF Model Execution on Jailbroken PlayStation 5

PS5LM transforms a jailbroken PS5 into a powerful, private home AI server. By leveraging llama.cpp, this project allows the console to run any GGUF-formatted model natively on its hardware, starting with the Qwen 3.8 series.

🚀 Core Functionality

The application utilizes the console's RDNA 2 GPU via the llama.cpp Vulkan backend. It allocates approximately $11.4\text{ GB}$ of shared memory to handle both the model weights and the KV cache.

Key Features

• Universal Compatibility: Supports every quantization method and architecture recognized by llama.cpp.
• Industry Standard APIs:
  • OpenAI compatible: /v1/chat/completions
  • Anthropic compatible: /v1/messages (fully supports Claude Code)
• Web Interface: A built-in chat page accessible via any browser.
• Connectivity: Simply point your client to http://console:8081. Any loaded model will respond to requests.
• TV Integration: All requests are streamed directly to the TV screen, showing the requesting device, the prompt, and the real-time response.
• Local Control: Use the $\square$ button for demos and $\triangle$ to trigger the keyboard for couch-based interaction.
• Telemetry Dashboard: A dedicated "Console" page tracks $\text{CPU/GPU load}$, $\text{power consumption}$, and $\text{temperature}$ over 2-minute or 1-hour intervals.

---

🧠 Decision Models & Ollaya Integration

PS5LM introduces Decision Models (e.g., d1, Clef, OpenJev), powered by the logic of Ollaya. Unlike standard LLMs, these models perform a single forward pass to provide typed decisions with calibrated probabilities rather than generating text.

What is a Decision Model? While Ollama runs LLMs on PCs, Ollaya runs decision models. PS5LM brings this to the console, allowing for near-instantaneous "Yes/No" or categorical answers.

Example API Request

curl http://console:8081/v1/systemone -H 'Content-Type: application/json' -d '{
  "state": "I was charged twice this month. Refund me or I cancel.",
  "questions": {
    "intent": {
      "type": "choice", 
      "instructions": "What does the customer want?", 
      "criteria": {"refund": null, "information": null, "cancel": null}
    },
    "angry": {
      "type": "noul", 
      "instructions": "Is the customer angry?"
    }
  }
}'

---

📊 Performance Benchmarks

The following data was captured on a PS5 Slim:

---

🛠️ Technical Architecture

The system flow can be visualized as follows:

The Tech Stack

• Language: C++
• UI Framework: ps5-homebrew-ui
• Backend: llama.cpp with ggml-vulkan
• Drivers: PS5_Vulkan's RADV (built on Linux)

---

📦 Installation & Requirements

Prerequisites

Your PS5 must be jailbroken and running firmware between 7.00 and 13.60.

Required Payloads:

• kstuff-lite (Kernel patches for homebrew)
• etaHEN (Homebrew support)
• ShadowMountPlus (Adds PS5LM to the home screen via /data/homebrew)

Recommended Payloads:

• elfldr (Port 9021: for power/temp/space monitoring)
• ftpsrv (Port 2121: for easy file transfers)

Setup Steps

1. Download ps5lm-app-vX.Y.Z.zip from the latest release and extract it.
2. Manually move models via FTP $\rightarrow$ Use the built-in Suggested list to download directly from Hugging Face.
3. Alternatively, place .gguf files in /data/PS5LM/models or on a USB drive.
4. Once the Home screen displays "Ready", access the interface at http://console:8081.

---

🎮 Controller Mapping

---

🛠️ For Developers & Tinkers

If you want to dive deeper into the code, explore the following:

• Building: See the Building guide.
• Porting: Check Porting and Research docs.
• CLI: Integration with opencode, Codex, or Claude Code is detailed in docs/CLI.md.

Special Thanks:

• abhi245y: Created the original native GPU app (PS5LM Plus).
• Mihawk-99: For PS5_Vulkan and RADV.
• PS5CEMU-HAR: For the Mesa driver patches.
• BlackBearReloaded: For the ps5-homebrew-ui.