pavanimajety

Debug inference model accuracy and numerical correctness in vLLM/pm-vllm. Use when investigating wrong outputs, numerical mismatches vs baseline, kernel/layer accuracy, quantization or checkpoint issues, or when the user mentions accuracy debugging, minimum reproducer, or inference correctness.