Open Mynameisyaojiaqi opened 2 years ago
Hi, that message indicates model inference + tensor extraction. On loading any model for the first time, Zetane runs placeholder data through the model in order to extract its runtime properties.
Additionally, the first time any Keras model is loaded, the Tensorflow runtime needs to be initialized which might increase the loading time. On average, the first "Keras: running model" message could take 30 seconds to 2 minutes. Subsequent loads should be much faster.
If the model never loads or takes too long to load for your usecase, we could certainly look into that. What OS are you using, and are you able to share the model (or point me to a similar one)?
You can also join our slack channel (https://zetane.com/join-slack) or email me directly (omar@zetane.com)
In the process of loading the model and testing the data, "keras:running model" is displayed, which takes a long time. Why?