Opportunistic Expert Activation: Batch-Aware Expert Routing for Faster Decode Without Retraining

IBM Released new Granite 4.0 Models with a Novel Hybrid Mamba-2/Transformer Architecture: Drastically Reducing Memory Use without Sacrificing Performance

Qwen3 von Grund auf: Ein Leitfaden zum führenden Open‑Source‑LLM

Learning to Shard: RL for Co-optimizing the Parallelism Degrees and Per-operator Sharding Dimensions in Distributed LLM Inference

FLUX: Federiertes Feintuning sparsamer LLMs auf ressourcenbeschränkten Geräten

Huawei präsentiert CloudMatrix: Peer-to-Peer-Architektur skalierbare LLM-Server

Gemini API File Search: The Easy Way to Build RAG