25.1.RC1

**Table 1** Features and documents related to 25.1.RC1
Feature	Document	Description
Kunpeng Recall Algorithm Library	Kunpeng Recall Algorithm Library Feature Documentation	Optimized the KBest algorithm performance. Added the SetEarlyStoppingParams API and modified the Add API and KBest constructors. Optimized the KScaNN algorithm performance and added the implementation for connecting KScaNN to Milvus.
Kunpeng Inference Acceleration Kit	Kunpeng Inference Acceleration Kit Feature Documentation	Added the KONNX sub-library.
Kunpeng Artificial Intelligence Library	Kunpeng Artificial Intelligence Library Feature Documentation	Added support for the Pool, Batch Normalization (bnormal), Local Response Normalization (lrn), Reduction, PReLU, Binary, and RNN deep neural network operators on the Kunpeng platform. Added support for the new Kunpeng 920 processor model.
TensorFlow Serving Thread Scheduling Optimization	TensorFlow Serving Thread Scheduling Optimization Feature Documentation	When multiple TensorFlow sessions run concurrently and share operators, they rely on a common thread pool for execution. This can lead to resource contention and significantly reduce the performance. The TensorFlow Serving thread scheduling optimization feature is provided by Kunpeng to improve graph execution efficiency under the high-concurrency scenarios. The operator scheduling algorithm is improved and thread management is optimized, which greatly improves the model inference throughput in the high-concurrency scenarios.