SYCL Llama Manager

PySide6ベースのGUIから llama.cpp (SYCL) を管理し、 OpenAI互換API経由で Open WebUI や各種OpenAIクライアントから利用できるデスクトップアプリケーションです。

Intel Arc GPU + oneAPI を前提に設計されています。

A PySide6 desktop application for managing llama.cpp (SYCL) with an OpenAI-compatible API.

---

Features

• 🖥️ PySide6 GUI
• 🤖 OpenAI Compatible API
• 🔄 Automatic model switching
• 📦 GGUF model management
• ⚡ Real-time Prompt / Decode speed monitor
• 💾 Real-time VRAM monitor
• 🚀 Startup Model
• 🔍 SYCL Diagnostics
• 🌐 Open WebUI compatible
• ⚙️ Setup Wizard

---

Architecture

Open WebUI
      │
      ▼
SYCL Llama Manager
      │
      ▼
llama-server (SYCL)
      │
      ▼
Intel Arc GPU

The manager launches and controls llama-server, exposes an OpenAI-compatible API, and automatically switches models when requested by clients.

---

Screenshots

Main Window

Open WebUI

Settings

---

Requirements

• Linux Desktop
• Python 3.12+
• uv
• Intel oneAPI
• llama.cpp (SYCL build)
• GGUF models

SYCL Llama Manager does not install oneAPI, GPU drivers or llama.cpp.

It only manages an existing environment.

---

Installation

git clone https://github.com/nuko6925/SYCL-Llama-Manager-GUI
cd SYCL-Llama-Manager-GUI
./install.sh

---

Quick Start

1. Run Setup Wizard
2. Configure:
   • llama-server
   • llama-cli
   • oneAPI
   • sycl-ls
   • GGUF root
3. Register a model
4. Press Start
5. Connect Open WebUI

---

OpenAI API

Default endpoint

http://localhost:10000/v1

This endpoint is compatible with the OpenAI API and can be used by Open WebUI and other compatible clients.

Health check

curl http://localhost:10000/health

Models

curl http://localhost:10000/v1/models

---

Main Features

Model Management

• GGUF registration
• Recursive scan
• Custom chat templates
• Startup model
• Automatic switching

Monitoring

• VRAM usage
• Prompt speed
• Decode speed
• Prompt tokens
• Generated tokens
• Runtime logs

API

• OpenAI compatible
• Chat Completions
• Completions
• Embeddings
• Streaming (SSE)

Startup

• Start with OS
• System Tray
• Retry
• Startup Model

---

Documentation

More detailed documentation can be found in:

• docs/ACCEPTANCE_TESTS.md
• docs/TROUBLESHOOTING.md
• sycl_llama_manager_spec_v1.md

---

License

MIT License