qualcomm / GenieX

Unified on-device LLM inference runtime with deep hardware optimization for Qualcomm's NPU, GPU, and CPU across multiple programming interfaces

活跃维护 BSD-3 Rust Tracked
8.3k 1.0k 1 天前
CI
llm on-device-ai sdk vlm go gpt-oss llama3 qwen3 granite4 qwen3vl gemma4 hexagon local-ai qualcomm snapdragon

星标趋势

数据积累中,暂无足够数据生成趋势图

AI 分析

项目摘要

GenieX is Qualcomm's open-source on-device GenAI inference runtime that enables running frontier LLMs and VLMs locally on Snapdragon devices across Hexagon NPU, Adreno GPU, and CPU. It provides a unified C SDK with multiple interfaces including CLI, Python, Kotlin/Java, and an OpenAI-compatible server, supporting GGUF models from Hugging Face or pre-compiled bundles from Qualcomm AI Hub.

为什么值得关注

It brings optimized, hardware-accelerated LLM inference to Qualcomm's ecosystem with a developer-friendly multi-language approach, filling a critical gap for edge AI on Snapdragon platforms where cloud connectivity is limited or undesirable.

优势

  • Hardware-specific optimization for Qualcomm's Hexagon NPU, Adreno GPU, and CPU with unified runtime
  • Multi-language support (Python, Kotlin/Java, C, CLI) and OpenAI-compatible server interface
  • Active development with 12 releases in 6 months, strong CI/CD, and comprehensive documentation

局限性

  • Restricted exclusively to Qualcomm Snapdragon hardware, limiting broader applicability
  • Developer Preview status indicates it's not yet production-ready despite active development

使用场景

  • Running local LLMs on Snapdragon-powered Windows ARM64 laptops for offline AI assistance
  • Deploying on-device AI in Android mobile applications using Qualcomm's latest mobile chipsets
  • Edge AI inference on IoT devices powered by Snapdragon Dragonwing platforms
目标用户: Edge AI developers, mobile developers, and IoT engineers building on Qualcomm Snapdragon platforms who need local LLM/VLM inference without cloud dependency
学习曲线:
分析模型:LongCat-2.0 | 分析时间:29 天前