Open
Description
I'm encountering an undefined variable issue while running inference with the GraphModel in TensorFlow.js. It's occurring in tfjs-converter/src/operations/executors/utils.ts#L138.
System Information:
- OS: Ubuntu 24
- TensorFlow.js Version: v4.22.0 / v4.5.0 (installed via npm)
- Browser: Firefox 134.0.2
- TensorFlow.js Converter Version: v4.5.0
- TensorFlow Version: v2.1.0
Current Behavior:
Inference fails due to an undefined variable, leading to a TypeError.
Expected Behavior:
The GraphModel should execute inference without errors.
Minimal Reproducible Code (in HTML):
<!DOCTYPE html>
<html lang="en">
<head>
<script src="https://cdn.jsdelivr.net/npm/@tensorflow/tfjs/dist/tf.min.js"></script>
</head>
<body>
<script>
async function run() {
const model = await tf.loadGraphModel('./model.json');
const tensor = tf.zeros([1, 320, 320, 3]);
const result = model.predict({"input": tensor});
}
run();
</script>
</body>
</html>
I have attached the model to reproduce the issue
Error Log (from Node):
TypeError: Cannot read properties of undefined (reading 'split')
at parseNodeName (tfjs-converter/src/operations/executors/utils.ts:138:22)
at getTensor (tfjs-converter/src/operations/executors/utils.ts:74:29)
at getParamValue (tfjs-converter/src/operations/executors/utils.ts:53:20)
...
Any insights on resolving this issue?