TNN

Overview

TNN is a cross-platform deep learning inference framework developed by Tencent Youtu Lab and Guangying Lab. It optimizes model performance for mobile, desktop, and server environments through features like model compression and code pruning, building on ncnn and Rapidnet. The framework is deployed across multiple Tencent applications, including Mobile QQ and Weishi.

Best for

Best for
Developers needing optimized deep learning inference specifically on mobile devices, especially within Tencent ecosystem integrations.

Use cases

• Deploying low-latency deep learning inference on mobile devices
• Compressing and pruning models for efficient on-device execution
• Running inference across iOS, Android, and desktop platforms with a unified engine

Notes

TNN is a cross-platform deep learning inference framework developed by Tencent Youtu Lab and Guangying Lab. It optimizes model performance for mobile, desktop, and server environments through features like model compression and code pruning, building on ncnn and Rapidnet. The framework is deployed across multiple Tencent applications, including Mobile QQ and Weishi.

4,634 stars on GitHub. Last updated 2025-05-09.

Use cases

• Deploying low-latency deep learning inference on mobile devices
• Compressing and pruning models for efficient on-device execution
• Running inference across iOS, Android, and desktop platforms with a unified engine

Pros

• High inference performance on mobile hardware through targeted optimizations
• Built-in model compression and code pruning reduce memory footprint
• Cross-platform support covers mobile, desktop, and server targets

Cons

• Smaller community and fewer third-party resources compared to TensorFlow Lite or ONNX Runtime
• Integration may require custom effort for non-Tencent use cases
• Limited to inference; does not include training capabilities

Indexed from awesome-llmops and enriched against its public facts.