Access model before conversion to .tlt OR decode .tlt to .hdf5/.pb

amogh.dabholkar · August 2, 2023, 5:21pm

Please provide the following information when requesting support.

• Hardware (T4/V100/Xavier/Nano/etc)
• Network Type (Detectnet_v2/Faster_rcnn/Yolo_v4/LPRnet/Mask_rcnn/Classification/etc) Classification_tf2
• TLT Version (Please run “tlt info --verbose” and share “docker_tag” here) v5.0.0
• Training spec file(If have, please share here)
• How to reproduce the issue ? (This is for errors. Please share the command line and the detailed log here.)

I would like to access the model file before it is internally converted to .tlt file, or try to decode the .tlt file to .hdf5 or .pb. Can you tell me how to do that? Exporting the .tlt model to .etlt and then to ONNX only to convert it back to tflite for deployment is causing a lot of non-native ops to be introduced and sometimes even failure in conversion. I see that the code for this exists, but I am not sure how to use it.

github.com

NVIDIA/tao_tensorflow2_backend/blob/main/internal/decode_eff.py

import argparse
import os
import zipfile

from eff.core import Archive, File
from eff.callbacks import BinaryContentCallback


def decode_eff(eff_model_path, output_path=None, passphrase=None):
    """Decode EFF to saved_model directory.

    Args:
        eff_model_path (str): Path to eff model
        passphrase (str, optional): Encryption key. Defaults to None.

    Returns:
        str: Path to the saved_model
    """
    # Decrypt EFF
    eff_filename = os.path.basename(eff_model_path)

This file has been truncated. show original

Morganh · August 3, 2023, 3:29am

You can try to run it by setting -m your.tlt , -o out.hdf5 and -k yourkey.

amogh.dabholkar · August 3, 2023, 1:31pm

Which mode do I use for this?
!Tao deploy -m mytlt -o out.hdf5 -k mykey?

Morganh · August 3, 2023, 2:19pm

Please run the python file directly inside the docker nvcr.io/nvidia/tao/tao-toolkit:5.0.0-tf2.11.0

system · August 17, 2023, 2:20pm

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.