note

[For CUDA 16GB] llama.cpp Q4_K batch prefill/decode 5.8x speedup, SGLang NVFP4 MoE & vLLM CUDA Graph optimization — 2026-06-28

This article has been edited and created by AI. llama.cpp Q4_K batch prefill/decode 5.8x speedup, SGLang NVFP4 MoE & vLLM CUDA Graph optimization — 2026-06-28 Today's major news for CUDA 16GB includes ...
note

[For CUDA 16GB] Gemma-4 E4B/E2B GGUF Quantization Released & llama.cpp Qwen3-8B 6x Speedup (CUDA 16GB Support)

LLM trends for CUDA 16GB as of today, June 25th. Most notable are the release of three Gemma-4 series E4B/E2B/GGUF quantized models by Unsloth, and the implementation of dense Q4_K batched ...

Santander publishes AI projects on GitHub

Banco Santander has shared a host of its AI projects under an open source licence in a bid to "ramp up shared innovation".
InfoWorld

Four agentic AI memory systems for smarter LLMs

These third-party projects greatly expand the ways agents and LLMs can draw on facts, documents, and conversations to deliver ...
GitHub

SakamotoDays395/code-graph-rag-cpp

code-graph-rag-cpp/ ├── CMakeLists.txt ← Open this in Visual Studio ├── config/default.yaml ← Application settings ├── examples/hello.py ← Test files for parser ├── src/ │ ├── main.cpp ← Entry point ...
GitHub

llama.cpp-mtp — Fused TBQ4 Flash Attention + MTP + Shared Tensors

Fork of llama.cpp with fused TurboQuant flash attention — the FA kernel reads raw TBQ4_0 K/V blocks directly from global memory and dequants via centroid lookup in the FWHT-rotated domain. No separate ...