- [Perf][DSV4.1] Optimize MegaMoE staging and NVFP4 cache gathers9/10vllm-project/vllm·performance·+130-62GPU kernel optimizationBatch processingDynamic resource allocationPerformance benchmarking
- metal : add MoE and SSM_CONV fusion optimizations9/10ggml-org/llama.cpp·performance·+1418-174Function constants in Metal shading languageFusion optimization patternsVector-based operation sequencesMemory aliasing guards
- today's pick · read the full story[FLINK-40472][python] Support Arrow vectorized UDFs in DataFrame API9/10apache/flink·feature·+1432-246Apache Arrow integrationVectorized user-defined functions (UDFs)Python decorators and annotationsData serialization and schema validation
- [Feat][Model] Support encoder-side SWA-bounded replay for DeepSeek-V4.1-Flash8/10vllm-project/vllm·feature·+1067-38Sliding window cachingPrefix caching and replayAsynchronous schedulingModel-specific attention metadata
- opencl: add support for bin kernel `flash_attn_f32_f16_bin`8/10ggml-org/llama.cpp·feature·+576-0OpenCL kernel integrationBinary kernel usageFlash attention mechanismC++ interoperability with GPU computing
- chat: add opt-in persistent working progress8/10microsoft/vscode·feature·+7826-536TypeScript enumsConfiguration settings managementReact component state and renderingAnimation handling in UI
- Stamp Airflow run identity onto agent runs and traces8/10apache/airflow·feature·+401-50Pydantic model parameterizationDistributed tracing with span metadataTask instance identification and retry handlingInter-task communication via XCom in Airflow
- [wasm][coreclr] ReadyToRun for CoreCLR browser-wasm8/10dotnet/runtime·feature·+707-64Ahead-of-time (AOT) compilationConditional compilation and feature gatingBuild and publish pipeline integrationCross-module optimization and validation
- Populate runtime-async DiagnosticIP for Wasm R2R8/10dotnet/runtime·feature·+775-56ReadyToRun (R2R) format enhancementsULEB128 encoding for compact data representationVirtual instruction pointer (IP) managementVersioning and backward compatibility in runtime formats
- Support the euclidian operator `<->`8/10duckdb/duckdb·feature·+1195-1153Operator overloadingUser-defined SQL functionsVector arithmeticEuclidean distance calculation