Hi, 目前有一个MNIST Inference的Demo[1]是使用GPU的，但是没有用到TensorFlow. 在flink-ai-extended项目中有个TensorFlow训练MNIST的例子[2]，但不确定能否直接用GPU版TF执行。我帮你involve一下Becket和Wei来确认下。
[1] https://github.com/KarmaGYZ/flink-mnist [2] https://github.com/alibaba/flink-ai-extended/tree/master/deep-learning-on-flink/flink-ml-examples/src/main/java/com/alibaba/flink/ml/examples/tensorflow/mnist Best, Yangze Guo On Mon, Apr 12, 2021 at 2:35 PM 张颖 <[email protected]> wrote: > > HI,I am running a tf inference task on my cluster,but I flind it took so long > a time to get response, becase it is a bert model and I run it on cpu > machine.My componey has gpu k8s cluster,and I read the document > https://ci.apache.org/projects/flink/flink-docs-master/docs/deployment/advanced/external_resources/ > > > > count you give me a demo?Including tf inference on gpu and train on gpu? > > I use alink in some of my task, is there a demo for alink on gpu? > > > this is part of the answer： > https://issues.apache.org/jira/browse/FLINK-22205
