The primary signal is desiredSize on the controller. It can be positive (wants data), zero (at capacity), negative (over capacity), or null (closed). Producers are supposed to check this value and stop enqueueing when it's not positive. But there's nothing enforcing this: controller.enqueue() always succeeds, even when desiredSize is deeply negative.
В России допустили «второй Чернобыль» в Иране22:31
，这一点在爱思助手下载最新版本中也有详细论述
На МКАД загорелись две машины14:46
ONNX 导出失败的根因是图里某处会把一个标量常量以没有 dtype（即 None） 的形式传给了 ONNX 导出器，导致 torch.onnx。_type_utils.JitScalarType.from_name 收到 None 并抛出 ValueError: Scalar type name cannot be None。这类情况常在用高级索引/原地赋值（tensor[index] = other、index_put、masked_scatter 等）时出现，导出器有时会把标量常量漏掉 dtype。