Predict is aborted with a .tflite model

[lhofinger]

Hello,

System information

• I have written custom code
• OS: Windows 11 23H2
• TensorFlow.js installed from script link https://cdn.jsdelivr.net/npm/@tensorflow/tfjs@latest/dist/tf.js
• TensorFlow.js version: latest
• Browser version: Chrome 130.0
• Tensorflow.js Converter Version: /

Describe the current behavior
Trying to use BirdNET model https://github.com/rdz-oss/BattyBirdNET-Analyzer/raw/refs/heads/main/checkpoints/V2.4/BirdNET_GLOBAL_6K_V2.4_Model_FP32.tflite

But I get this error:
classif.js:22 Error: RuntimeError: Aborted(). Build with -sASSERTIONS for more info.
at abort (tflite_web_api_cc_simd.js:9:6515)
at abort (tflite_web_api_cc_simd.js:9:60528)
at tflite_web_api_cc_simd.wasm:0x1baad
at tflite_web_api_cc_simd.wasm:0x2e1f87
at tflite_web_api_cc_simd.wasm:0x2f907
at tflite_web_api_cc_simd.wasm:0x522f0
at tflite_web_api_cc_simd.wasm:0x32b150
at tflite_web_api_cc_simd.wasm:0x17e14
at TFLiteWebModelRunner.TFLiteWebModelRunner$Infer [as Infer] (eval at new (tflite_web_api_cc_simd.js:9:33059), :8:10)
at module$exports$google3$third_party$tensorflow_lite_support$web$tflite_web_api_client.TFLiteWebModelRunner.infer (tflite_web_api_client.js:2714:188)

The same code, with a simple model is not aborting.

Standalone code to reproduce the issue
Simple code with issue: https://www.paludour.net/bats/classification.html

[shmishra99]

Hi @lhofinger ,

I am able to reproduce the issue you're facing. I created a dummy dataset of 144,000 points and encountered the same error.

I've noticed that the model sometimes halts for a while before throwing the error. Additionally, models with fewer tensors seem to work without problems.
While the model is running, I've observed that browser memory consumption becomes very high. I suspect this could be an 'out-of-memory' issue due to the execution of large tensors.

Thank You!!

[lhofinger]

Hello and thank you. Great that you have reproduced the problem. I'm not sure it is a memory problem. This model works great in Python with not too much memory. Can anyone look deeper?