ggml-org/llama.cpp

ggml-org/llama.cpp shipped b8868-b8870. Added Vulkan F16 support, fixed multi-token mask decoding. Expands efficient inference to non-NVIDIA GPUs. qdrant/qdrant added TQ-4-bit-SIMD for hardware-accelerated vector search.

Contains the exact GitHub repo slug.

ggml-org/llama.cpp released b8857 with major WebGPU matrix-vector multiplication improvements. These specific updates accelerate browser-based LLM inference, reducing reliance on local hardware.

Contains the exact GitHub repo slug.

ggml-org/llama.cpp released b8851 with cpp-httplib v0.42.0. New branch `gg/common-log-to-base` suggests deep math optimizations for quantization efficiency, keeping it the default for on-device LLMs.

Contains the exact GitHub repo slug.

ggml-org/llama.cpp pushed versions b8841-b8850. Highlights include RPC transport refactoring, CPU subgraph caching, and crucial AMD GPU fixes (CDNA/RDNA). Also improved tool parsing compatibility. A relentless push for robust local inference speeds.

Contains the exact GitHub repo slug.