Jun 13Vibe with Hermes Agent — Bengaluru · RSVP
ToolsMCPBlogResearchCommunityStar on GitHub
All Tools
T
Dev ToolsFreeOpen Source

TNN

Tencent's cross-platform deep learning inference framework

BSD-3-Clause

ABOUT

Deploying deep learning models across mobile, desktop, and server platforms usually forces a choice between raw speed and portability. TNN, built on the lineage of ncnn and Rapidnet, delivers both: it provides a unified inference runtime across Android, iOS, Windows, Linux, and macOS with architecture-specific optimizations (ARM NEON, x86 AVX, GPU). Model compression and operator pruning reduce the binary size, making TNN suitable for production apps including Tencent's own Mobile QQ and Weishi. Models from PyTorch, ONNX, and TensorFlow can be converted via the TNN converter toolchain.

INSTALL
pip install tnn

INTEGRATION GUIDE

1. Deploy a YOLO object detection model in a mobile app with real-time video processing 2. Run a BERT-based question answering model on-device for privacy-preserving AI 3. Convert and optimize a PyTorch classification model for deployment on Android and iOS 4. Prune and compress a neural network to reduce app binary size while maintaining accuracy 5. Benchmark inference throughput across ARM, x86, and GPU backends from a single model

TAGS

inferencemobiledeep-learningcross-platformoptimizationmodel-compressiontencent