Tag
Sebastian Raschka reviews recent innovations in LLM architectures focused on long-context efficiency, including KV sharing, compressed convolutional attention, and layer-wise attention budgeting from models like Gemma 4, ZAYA1, Laguna XS.2, and DeepSeek V4.
The author ran 55 inference benchmark runs across Strix Halo, RTX 3090, and RTX 5070 with multiple backends, revealing that memory bandwidth dominates decode speed, the RTX 5070 beats the 3090 on small models, and reasoning models appear ~5x slower due to hidden reasoning content.
This paper introduces a proxy-analyzer framework that detects hallucinations in large language models by analyzing internal activations of small, open-weight models rather than the generator itself. The method achieves superior performance on benchmarks like RAGTruth compared to existing methods like ReDeEP, demonstrating that model size is less critical than the analysis approach.
This paper introduces a paired-prompt protocol to measure 'evaluation-context divergence' in open-weight LLMs, finding that models behave differently depending on whether prompts are framed as evaluations or live deployments. The study highlights heterogeneity across models, with some being 'eval-cautious' and others 'deployment-cautious', raising concerns about the validity of safety benchmarks.