What's New

1

Added the Kunpeng DiskANN documents.

• Based on the open-source DiskANN algorithm library, the disk retrieval performance is improved through NEON SIMD vectorization and data layout optimization.
• Provides optimization patch introduction, API reference, and patch enabling documents.

BoostKit BoostSRA Kunpeng DiskANN documents

2

Added the Kunpeng Embedding Lookup documents.

• Deeply adapted and optimized the core Embedding Lookup module of the open-source Monolith large-scale real-time recommendation system.
• Provides optimization patch introduction, API reference, and patch enabling documents.

BoostKit BoostSRA Embedding Lookup documents

3

Updated the KRL documents.

Added SVE2 instruction implementation for the 4-bit table lookup-based accumulation operators and some FP16 distance computation operators.

BoostKit BoostSRA KRL documents

4

Updated the Kunpeng Inference Acceleration Kit documents.

Modified the matrix partitioning method for matrix computation in ONNX Runtime.

BoostKit BoostSRA Kunpeng Inference Acceleration Kit Documentation

5

Updated the Kunpeng TensorFlow inference optimization documents.

• Added the TensorFlow ANNC static graph fusion feature, adapted to the Kunpeng 950 processor, and supporting operators such as KPFusedGather and KPFusedSparseReshape.
• Added constant folding optimization to the TensorFlow ANNC graph compilation optimization feature, adapted to the Kunpeng 950 processor.

BoostKit BoostSRA TensorFlow inference optimization documents

6

Added the Kunpeng TensorRT-LLM documents for TensorFlow inference optimizations.

• Based on open-source TensorRT-LLM, it focuses on efficient execution in Large Language Model (LLM) inference scenarios. Through deep performance enhancements including kernel optimization, memory access optimization, and parameter tuning, it significantly increases inference throughput and reduces latency.
• Provides optimization patch introduction, API reference, and patch enabling documents.

BoostKit BoostSRA Kunpeng TensorRT-LLM documents

1

Migrated the KBest documents to the GitCode open-source platform.

Binary software packages and documents are provided on GitCode.

BoostKit BoostSRA KBest documents

2

Migrated the KScaNN documents to the GitCode open-source platform.

Binary software packages and documents are provided on GitCode.

BoostKit BoostSRA KScaNN documents

3

Migrated the KVecTurbo documents to the GitCode open-source platform.

Open-source software packages and documents are provided on GitCode.

BoostKit BoostSRA KVecTurbo documents

4

Migrated the KNewPfordelta documents to the GitCode open-source platform.

Patch packages and documents for the open-source PForDelta algorithm are provided on GitCode.

BoostKit BoostSRA KNewPfordelta documents

5

Migrated the Kunpeng hnswlib documents to the GitCode open-source platform.

• Patch packages and documents for the open-source hnswlib algorithm are provided on GitCode.
• Supports the FP 16 data type, NEON vectorization, prefetching, and database ID renumbering optimization. The reconstructed hnswlib simplifies the code structure and improves usability.

BoostKit BoostSRA Kunpeng hnswlib documents

6

Migrated the Kunpeng Faiss documents to the GitCode open-source platform.

• Patch packages and documents for the open-source Faiss algorithm are provided on GitCode.
• Provides non-equivalence optimization patch and equivalence optimization patch. The non-equivalence optimization patch further optimizes the IVFPQ algorithm and supports the HNSW FP16 interface.

BoostKit BoostSRA Faiss documents

7

Added Kunpeng RaBitQ.

• Intrusive modifications are made to the open-source RaBitQ codebase. This extends its support to the AArch64 architecture, introducing performance optimizations and functional enhancements. The optimizations include FP16 precision optimization, NEON SIMD vectorization, assembly-level Lookup Table (LUT) acceleration, Spilling with Orthogonality-Amplified Residuals (SOAR) spilled vector assignment, and ML-based adaptive nprobe.
• Patch packages and documents for the open-source RaBitQ are provided on GitCode.

BoostKit BoostSRA Kunpeng RaBitQ documents

8

Migrated the KRL documents to the GitCode open-source platform.

• Binary software packages and documents are provided on GitCode.

BoostKit BoostSRA KRL documents

9

Migrated the Kunpeng TensorFlow inference optimization documents to the GitCode open-source platform.

• The TensorFlow Serving thread scheduling optimization and the TensorFlow Serving ANNC graph compilation optimization are collectively referred to as "Kunpeng TensorFlow inference optimization."
• Patch packages and documents for the open-source TensorFlow are provided on GitCode.

BoostKit BoostSRA TensorFlow inference optimization documents