While OP Issue in TFLM

[HemanthSai7]

System information

• OS Platform and Distribution (e.g., Linux Ubuntu 22.04):
• TensorFlow installed from (source or binary): pip
• TensorFlow version (or github SHA if from source): 2.15.0.post1

Background of the issue
I created a conformer transducer model following the Conformer paper. The model consists of encoder, decoder and the joiner network. The Prediction network contains a single LSTM Layer. I converted the model to INT8 using the get_concrete_function().

concrete_func = model.get_concrete_function()
converter = tf.lite.TFLiteConverter.from_concrete_functions([concrete_func])

The model converted successfully and after visualising the model in netron, i verified that none of the ops were missing. I then flashed the model in esp32s3 and kept on getting this error infinitely in while.cc op. I have pasted the screenshot of the same below.

I then created a simple model consisting of only while loop and then flashed it. I faced the same error.
NOTE: AllocateTensors() was passed

Provide the text output from tflite_convert

W tensorflow/compiler/mlir/lite/python/tf_tfl_flatbuffer_helpers.cc:378] Ignored output_format.
W tensorflow/compiler/mlir/lite/python/tf_tfl_flatbuffer_helpers.cc:381] Ignored drop_control_dependency.
I tensorflow/cc/saved_model/reader.cc:83] Reading SavedModel from: /tmp/tmp839tyn2a
I tensorflow/cc/saved_model/reader.cc:51] Reading meta graph with tags { serve }
I tensorflow/cc/saved_model/reader.cc:146] Reading SavedModel debug info (if present) from: /tmp/tmp839tyn2a
I tensorflow/compiler/mlir/mlir_graph_optimization_pass.cc:388] MLIR V1 optimization pass is not enabled
I tensorflow/cc/saved_model/loader.cc:233] Restoring SavedModel bundle.
I tensorflow/cc/saved_model/loader.cc:217] Running initialization op on SavedModel bundle at path: /tmp/tmp839tyn2a
I tensorflow/cc/saved_model/loader.cc:316] SavedModel load for tags { serve }; Status: success: OK. Took 13069 microseconds.
I tensorflow/compiler/mlir/tensorflow/utils/dump_mlir_util.cc:269] disabling MLIR crash reproducer, set env var `MLIR_CRASH_REPRODUCER_DIRECTORY` to enable.
Summary on the non-converted ops:
---------------------------------
 * Accepted dialects: tfl, builtin, func
 * Non-Converted Ops: 2, Total Ops 18, % non-converted = 11.11 %
 * 2 ARITH ops

- arith.constant:    2 occurrences  (i32: 2)

  (i1: 1, i32: 1)


  (i32: 2)
  (i1: 1)
  (i32: 1)
  (i32: 1)

I tensorflow/compiler/mlir/lite/flatbuffer_export.cc:2989] Estimated count of arithmetic ops: 3  ops, equivalently 1  MACs
fully_quantize: 0, inference_type: 6, input_inference_type: FLOAT32, output_inference_type: FLOAT32
I tensorflow/compiler/mlir/lite/flatbuffer_export.cc:2989] Estimated count of arithmetic ops: 3  ops, equivalently 1  MACs

Standalone code to reproduce the issue
This is how I converted the sample model consisting only while loop

import tensorflow as tf

def convert_tflite(model, rep_data):
    converter = tf.lite.TFLiteConverter.from_keras_model(model)
    converter.experimental_new_converter = True
    converter.representative_dataset = rep_data
    converter.optimizations = [tf.lite.Optimize.DEFAULT]
    converter.target_spec.supported_ops = [tf.lite.OpsSet.TFLITE_BUILTINS_INT8, tf.lite.OpsSet.SELECT_TF_OPS]
    converter.allow_custom_ops = True
    converter._experimental_lower_tensor_list_ops = False
    tflite_model = converter.convert()
    with open("model.tflite", "wb") as f:
        f.write(tflite_model)

class SimpleWhileLoopModel(tf.keras.Model):
    def __init__(self):
        super(SimpleWhileLoopModel, self).__init__()

    @tf.function
    def call(self, inputs):
        i, result = tf.constant(0), tf.constant(0)
        while i < inputs:
            result += i * i
            i += 1
        return result

model = SimpleWhileLoopModel()
model.summary()


def representative_dataset_gen():
    for limit in range(1, 11):
        yield [tf.constant(limit, dtype=tf.int32)]

convert_tflite(model, rep_data = representative_dataset_gen)

Also, please include a link to a GraphDef or the model if possible.

Any other info / logs

This is what I meant by the loop getting stuck

E (331202) task_wdt: Task watchdog got triggered. The following tasks did not reset the watchdog in time:
E (331202) task_wdt:  - IDLE (CPU 0)
E (331202) task_wdt: Tasks currently running:
E (331202) task_wdt: CPU 0: main
E (331202) task_wdt: CPU 1: IDLE
E (331202) task_wdt: Print CPU 0 (current core) backtrace


Backtrace: 0x4204E152:0x3FC92460 0x403770C9:0x3FC92480 0x42041845:0x3FCF39E0 0x4201AA06:0x3FCF3A00 0x42030772:0x3FCF3A40 0x4200927E:0x3FCF3A70 0x42008F5E:0x3FCF3AA0 0x42006CA3:0x3FCF3AC0 0x42006B36:0x3FCF3AE0 0x42006B2B:0x3FCF3B00 0x4206162C:0x3FCF3B20 0x4037C795:0x3FCF3B40
0x4204e152: task_wdt_isr at esp-adf/esp-idf/components/esp_system/task_wdt.c:183 (discriminator 3)

0x403770c9: _xt_lowint1 at esp-adf/esp-idf/components/freertos/port/xtensa/xtensa_vectors.S:1114

0x42041845: tflite::TfLiteTypeSizeOf(TfLiteType, unsigned int*) at dependency/esp-tflite-micro/tensorflow/lite/micro/memory_helpers.cc:87
 (inlined by) tflite::TfLiteEvalTensorByteLength(TfLiteEvalTensor const*, unsigned int*) at dependency/esp-tflite-micro/tensorflow/lite/micro/memory_helpers.cc:135

0x4201aa06: tflite::micro::ValidateAndGetTensorSizes(TfLiteEvalTensor const*, TfLiteEvalTensor const*) at dependency/esp-tflite-micro/tensorflow/lite/micro/kernels/kernel_util.cc:156
 (inlined by) tflite::micro::CopySubgraphOutputsToOpOutputs(TfLiteContext*, TfLiteNode*, tflite::MicroGraph*, int) at dependency/esp-tflite-micro/tensorflow/lite/micro/kernels/kernel_util.cc:262

0x42030772: tflite::(anonymous namespace)::WhileEval(TfLiteContext*, TfLiteNode*) at dependency/esp-tflite-micro/tensorflow/lite/micro/kernels/while.cc:110

0x4200927e: tflite::MicroInterpreterGraph::InvokeSubgraph(int) at dependency/esp-tflite-micro/tensorflow/lite/micro/micro_interpreter_graph.cc:194

0x42008f5e: tflite::MicroInterpreter::Invoke() at dependency/esp-tflite-micro/tensorflow/lite/micro/micro_interpreter.cc:294

0x42006ca3: loop at application/ml_testing/src/main_functions.cc:259

0x42006b36: app_init at application/ml_testing/src/ml_testing.cpp:11 (discriminator 1)

0x42006b2b: app_main at main/src/audio_frame_work.cpp:10

0x4206162c: main_task at esp-adf/esp-idf/components/freertos/port/port_common.c:141 (discriminator 2)

0x4037c795: vPortTaskWrapper at esp-adf/esp-idf/components/freertos/port/xtensa/port.c:142


E (331202) task_wdt: Print CPU 1 backtrace


Backtrace: 0x40378659:0x3FC92A60 0x403770C9:0x3FC92A80 0x400559DD:0x3FCF4940 |<-CORRUPTED
0x40378659: esp_crosscore_isr at esp-adf/esp-idf/components/esp_system/crosscore_int.c:92

0x403770c9: _xt_lowint1 at esp-adf/esp-idf/components/freertos/port/xtensa/xtensa_vectors.S:1114


E (336202) task_wdt: Task watchdog got triggered. The following tasks did not reset the watchdog in time:
E (336202) task_wdt:  - IDLE (CPU 0)
E (336202) task_wdt: Tasks currently running:
E (336202) task_wdt: CPU 0: main
E (336202) task_wdt: CPU 1: IDLE
E (336202) task_wdt: Print CPU 0 (current core) backtrace


Backtrace: 0x4204E152:0x3FC92460 0x403770C9:0x3FC92480 0x420326B6:0x3FCF39F0 0x4200927E:0x3FCF3A10 0x42030762:0x3FCF3A40 0x4200927E:0x3FCF3A70 0x42008F5E:0x3FCF3AA0 0x42006CA3:0x3FCF3AC0 0x42006B36:0x3FCF3AE0 0x42006B2B:0x3FCF3B00 0x4206162C:0x3FCF3B20 0x4037C795:0x3FCF3B40
0x4204e152: task_wdt_isr at esp-adf/esp-idf/components/esp_system/task_wdt.c:183 (discriminator 3)

0x403770c9: _xt_lowint1 at esp-adf/esp-idf/components/freertos/port/xtensa/xtensa_vectors.S:1114

0x420326b6: tflite::AddEval(TfLiteContext*, TfLiteNode*) at dependency/esp-tflite-micro/tensorflow/lite/micro/kernels/esp_nn/add.cc:213

0x4200927e: tflite::MicroInterpreterGraph::InvokeSubgraph(int) at dependency/esp-tflite-micro/tensorflow/lite/micro/micro_interpreter_graph.cc:194

0x42030762: tflite::(anonymous namespace)::WhileEval(TfLiteContext*, TfLiteNode*) at dependency/esp-tflite-micro/tensorflow/lite/micro/kernels/while.cc:107

0x4200927e: tflite::MicroInterpreterGraph::InvokeSubgraph(int) at dependency/esp-tflite-micro/tensorflow/lite/micro/micro_interpreter_graph.cc:194

0x42008f5e: tflite::MicroInterpreter::Invoke() at dependency/esp-tflite-micro/tensorflow/lite/micro/micro_interpreter.cc:294

0x42006ca3: loop at application/ml_testing/src/main_functions.cc:259

0x42006b36: app_init at application/ml_testing/src/ml_testing.cpp:11 (discriminator 1)

0x42006b2b: app_main at main/src/audio_frame_work.cpp:10

0x4206162c: main_task at esp-adf/esp-idf/components/freertos/port/port_common.c:141 (discriminator 2)

0x4037c795: vPortTaskWrapper at esp-adf/esp-idf/components/freertos/port/xtensa/port.c:142


E (336202) task_wdt: Print CPU 1 backtrace


Backtrace: 0x40378659:0x3FC92A60 0x403770C9:0x3FC92A80 0x400559DD:0x3FCF4940 |<-CORRUPTED
0x40378659: esp_crosscore_isr at esp-adf/esp-idf/components/esp_system/crosscore_int.c:92

0x403770c9: _xt_lowint1 at esp-adf/esp-idf/components/freertos/port/xtensa/xtensa_vectors.S:1114


E (341202) task_wdt: Task watchdog got triggered. The following tasks did not reset the watchdog in time:
E (341202) task_wdt:  - IDLE (CPU 0)
E (341202) task_wdt: Tasks currently running:
E (341202) task_wdt: CPU 0: main
E (341202) task_wdt: CPU 1: IDLE
E (341202) task_wdt: Print CPU 0 (current core) backtrace


Backtrace: 0x4204E152:0x3FC92460 0x403770C9:0x3FC92480 0x4200924C:0x3FCF3A10 0x42030762:0x3FCF3A40 0x4200927E:0x3FCF3A70 0x42008F5E:0x3FCF3AA0 0x42006CA3:0x3FCF3AC0 0x42006B36:0x3FCF3AE0 0x42006B2B:0x3FCF3B00 0x4206162C:0x3FCF3B20 0x4037C795:0x3FCF3B40
0x4204e152: task_wdt_isr at esp-adf/esp-idf/components/esp_system/task_wdt.c:183 (discriminator 3)

0x403770c9: _xt_lowint1 at esp-adf/esp-idf/components/freertos/port/xtensa/xtensa_vectors.S:1114

0x4200924c: tflite::EnumNameBuiltinOperator(tflite::BuiltinOperator) at dependency/esp-tflite-micro/tensorflow/lite/schema/schema_generated.h:1641
 (inlined by) OpNameFromRegistration at dependency/esp-tflite-micro/tensorflow/lite/micro/micro_interpreter_graph.cc:34
 (inlined by) tflite::MicroInterpreterGraph::InvokeSubgraph(int) at dependency/esp-tflite-micro/tensorflow/lite/micro/micro_interpreter_graph.cc:190

0x42030762: tflite::(anonymous namespace)::WhileEval(TfLiteContext*, TfLiteNode*) at dependency/esp-tflite-micro/tensorflow/lite/micro/kernels/while.cc:107

0x4200927e: tflite::MicroInterpreterGraph::InvokeSubgraph(int) at dependency/esp-tflite-micro/tensorflow/lite/micro/micro_interpreter_graph.cc:194

0x42008f5e: tflite::MicroInterpreter::Invoke() at dependency/esp-tflite-micro/tensorflow/lite/micro/micro_interpreter.cc:294

0x42006ca3: loop at application/ml_testing/src/main_functions.cc:259

0x42006b36: app_init at application/ml_testing/src/ml_testing.cpp:11 (discriminator 1)

0x42006b2b: app_main at main/src/audio_frame_work.cpp:10

0x4206162c: main_task at esp-adf/esp-idf/components/freertos/port/port_common.c:141 (discriminator 2)

0x4037c795: vPortTaskWrapper at esp-adf/esp-idf/components/freertos/port/xtensa/port.c:142


E (341202) task_wdt: Print CPU 1 backtrace


Backtrace: 0x40378659:0x3FC92A60 0x403770C9:0x3FC92A80 0x400559DD:0x3FCF4940 |<-CORRUPTED
0x40378659: esp_crosscore_isr at esp-adf/esp-idf/components/esp_system/crosscore_int.c:92

0x403770c9: _xt_lowint1 at esp-adf/esp-idf/components/freertos/port/xtensa/xtensa_vectors.S:1114

[github-actions]

"This issue is being marked as stale due to inactivity. Remove label or comment to prevent closure in 5 days."

[github-actions]

"This issue is being closed because it has been marked as
stale for 5 days with no further activity."