Telemetry-based taxonomy of how LLMs strain, drift, and hallucinate — measured from layer activations, attention, KV cache, and MoE routing.
telemetry pytorch artificial-intelligence llama observability interpretability mixture-of-experts mechanistic-interpretability regime-classification qwen hallucination-detection llm-observability regime-detection ai-telemetry forensic-telemetry model-telemetry
-
Updated
May 21, 2026 - Python