tensorflow
memory leak in executeAsync?
"dependencies": { "@tensorflow/tfjs-backend-cpu": "^3.5.0", "@tensorflow/tfjs-backend-webgl": "^3.5.0", "@tensorflow/tfjs-converter": "^3.5.0", "@tensorflow/tfjs-core": "^3.5.0", "fetch-wechat": "^0.0.3", "regenerator-runtime": "^0.14.1" }, "plugins": { "tfjsPlugin": { "version": "0.2.0", "provider": "wx6afed118d9e81df9" } }, WeChat version: 8.0.50 WeChat base API version: 3.5.0 WeChat IDE version: 1.06.2405020 win32-x64
I am trying to use a model with dynamic axis to adapt to different size of input, like [1, 3, 255,255] or [1,3,127,127]. So I have to use executeAsync to get model inference, since there is error using 'predict' method:
nano_tracker.js:36 Preheat failed: Error: This execution contains the node 'StatefulPartitionedCall/assert_equal_5/Assert/AssertGuard/branch_executed/_114', which has the dynamic op 'Merge'. Please use model.executeAsync() instead. Alternatively, to avoid the dynamic ops, specify the inputs [Identity] at e.compile (index.js:17) at e.execute (index.js:17) at e.execute (index.js:17) at e.predict (index.js:17) at ModelBuilder._callee3$ (model_builder.js:54) at s (regeneratorRuntime.js:1) at Generator.<anonymous> (regeneratorRuntime.js:1) at Generator.next (regeneratorRuntime.js:1) at asyncGeneratorStep (asyncToGenerator.js:1) at c (asyncToGenerator.js:1)(env: Windows,mp,1.06.2405020; lib: 3.5.3)
Using executeAsync seems to get correct model inference, while memory increasing with every invoke until it fails.
Is there anyone knows how to fix this? Thank you so much
