- [Perf][Kernel] Add a CuteDSL BF16x3 router GEMM for fp32 router weights on Hopper9/10vllm-project/vllm·performance·+921-36CUDA kernel programmingMixed precision computation (BF16 and FP32)Just-In-Time (JIT) compilationPerformance optimization for GPU workloads
- ggml-cpu : rewrite concat as one row kernel for all types9/10ggml-org/llama.cpp·performance·+60-176Template metaprogramming for type-generic codeMultithreading with work distribution across rowsMemory optimization using tiled copying and memcpyPerformance benchmarking and testing with varied data layouts
- today's pick · read the full storyfeat: extract QueryView coordinator components9/10milvus-io/milvus·feature·+24531-2789Immutable data structuresSnapshot isolationBackground batch processingConfiguration-driven resource management
- Many-core/NUMA improvements9/10duckdb/duckdb·performance·+1146-191NUMA-aware memory allocationPrefetching for cache optimizationWork stealing for load balancingMemory reuse and allocation caching
- server: add router device memory margin parameter for dynamic unloading8/10ggml-org/llama.cpp·feature·+1322-356Command-line argument parsingDynamic memory managementResource tracking and evictionServer-side model loading