Funding
Funding not disclosed
Founders
Product
Problem
Developing and deploying AI models, especially large language models, requires significant computational resources, leading to high costs and limited accessibility, particularly when using domestically produced Chinese chips. Existing AI software platforms often lack unified support for diverse CPU, GPU, and NPU architectures, creating integration challenges.
Solution
Zhongke Jiahe provides a compilation and optimization framework designed to deliver high-performance computing resources for AI models, with a focus on supporting Chinese-made chips. Their technology offers a generalized and cost-effective solution by vertically integrating model inference architectures across algorithm, system, and chip layers. The framework provides a standardized AI software foundation that enables seamless deployment of models on hardware environments powered by domestic AI chips, supporting both server-side (data center) and edge-side (personal computers, mobile devices) inference.
Target Audience
The primary target audience includes AI developers and enterprises in China who require high-performance, low-cost computing resources for deploying AI models on domestic chips, as well as hardware manufacturers seeking to optimize their chip performance for AI applications.
Features
- Compilation framework supporting multiple backend chips for generalized computing power.
- Model inference optimization tools that enhance performance without requiring user modifications.
- Standardized AI software base compatible across different brands and models of domestic AI chips.
- Operator auto-generation tools that provide CUDA-compatible translation solutions.
- Support for both server-side and edge-side large model inference.
- Tools to reduce AI computing costs.