A deterministic PyTorch autograd verification trap for catching silent KV-cache routing and block-alignment failures in vLLM and SGLang serving infrastructure.
cli inference pytorch autograd multi-agent fuzzing sympy formal-verification mlops kv-cache llm-serving vllm pagedattention sglang agentic-workflow ml-infra radixattention
-
Updated
Jun 7, 2026 - Python