- ggml-zdnn: introduce Q8_0 mat mul9/10ggml-org/llama.cpp·performance·+701-17Quantized data typesMatrix multiplication optimizationEnvironment variables for configurationPerformance benchmarking
- today's pick · read the full storyfeat: accelerate regex filtering with FMIndex candidates9/10milvus-io/milvus·feature·+2826-138Regular expressionsIndex-based filteringBitmaps for candidate selectionAdapter pattern
- Add a tap middleware9/10traefik/traefik·feature·+5234-16Middleware patternHTTP request and response handlingJSON serializationConfigurable service routing
- [Feature][Pooling] Add offline fixed-length late chunking8/10vllm-project/vllm·feature·+1147-7Mean pooling of token embeddingsFunction parameterization with dataclassesOffline batch processingTokenization and offset mapping
- feat: support analyzer char filter8/10milvus-io/milvus·feature·+2223-2Text tokenization pipelineCharacter filtering in text processingOffset correction in token streamsCross-language testing (Rust and Go)