[ 仅 API 调用方式不一致 ]torch.cuda.StreamContext

torch.cuda.StreamContext

torch.cuda.StreamContext(stream)

paddle.device.stream_guard

paddle.device.stream_guard(stream)

两者功能一致,但调用方式不一致,具体如下:

转写示例

# PyTorch 写法
with torch.cuda.StreamContext(stream=s1):
    result = a + b

# Paddle 写法
with paddle.device.stream_guard(stream=s1):
    result = a + b