nota-netspresso / pynetspresso Goto Github PK

Description

• Add convert (TFLite int8) and benchmark for Alif Ensemble DevKit Gen2 and Renesas RA8D1

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.

Todo

• Add int8 data type for TFLite converting
• Add Alif Ensemble DevKit Gen2 adn Renesas RA8D1 on target devices

Description

• Upload the docker image to the hub so that users can use it directly as a docker image instead of a pip install.

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.

Description

• There is a bug in the process of assigning the recommendation result to the available layer.

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.

Describe the bug

The recent libraries with FastAPI require an up-to-date typing_extensions (e.g. 4.8.0)

The version requirements are too specific, so it is hard to install with other libraries...

Plus, both pydantic==1.10.4 and fastapi, other dependencies on the library, are satisfied with typing-extensions==4.8.0.
However, the current dependency typing_extensions==4.5.0 shows an error.

This would be solved if the requirements.txt should be fixed to:

pydantic==1.10.8
typing-extensions>=4.8.0

Have you searched existing issues? 🔎

• I have searched and found no existing issues

Reproduction

Just

pip install netspresso

(Use with library)

throws an error.

Screenshot

No response

Logs

ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
fastapi 0.104.1 requires typing-extensions>=4.8.0, but you have typing-extensions 4.5.0 which is incompatible.
pydantic-core 2.14.3 requires typing-extensions!=4.7.0,>=4.6.0, but you have typing-extensions 4.5.0 which is incompatible.

System Info

Python 3.8
General with all OS systems

• I have searched to see if a similar issue already exists.

Is there any feature that you would like to add? (추가했으면 하는 feature가 어떤 것일까요?)
Community Standards을 위한 파일 추가

• Github Issue & PR Template
• CODE_OF_CONDUCT
• CONTRIBUTING
• SECURITY

Is there any suggestion (or solution) to solve this issue? (이를 해결하기 위한 아이디어가 있으신가요?)
Please write a clear and concise description (간결하고 명확하게 작성해주세요)

Additional context (전후 맥락을 돕기 위한 내용들)
All other contexts or screenshots are welcome. (노트, 토의 내용, 스크린샷 등 모든 추가 정보를 넣어주세요)

Description

• Modify to build trainer instance with original trainer config for retraining.

ex)

# For Training
trainer = Trainer(task=Task.OBJECT_DETECTION)

# For Retraining
retrainer = Trainer(task=Task.OBJECT_DETECTION, from_config="config_path")

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.

Description

• Currently, we are receiving output_path as file path, but we will modify it to receive it as folder path.

As-is

compressed_model = compressor.automatic_compression(
    ...
    output_path="./outputs/compressed/compressed_model.h5"
)

To-be

compressed_model = compressor.automatic_compression(
    ...
    output_path="./outputs/compressed/automatic_05"
)

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.

Description

• Add logic that automatically creates folders if there are no folders in the path you entered when you download the model

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.

Description

• Add exporting to ONNX for immediate use with the converter and benchmark after compression

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.

Description

• Change task & framework parameters to optional in compressor
  • The default for the task is other.
  • The default for the framework is pytorch.

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.

Description

• Remove the model_name parameter to compression function.

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.

Description

• Remove validate_token decorator.
  • It is inconvenient to check the parameters and return values of the function because of the decoder

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.

Description

• Change img_size from set_model_config to set_model_config parameter.

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.

Description

• Add pypi packaging with github workflow

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.

Description

• Change the default value of number_of_layers from 0 to None.

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.

Description

• Add input_model_framework parameter for convert & benchmark.

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.

Description

• Implementation of functions that manage the state of train, compress, convert, benchmark

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.

Description

• Implement a function that manages json data to be used by GUI.

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.

Description

• Add assertion for file_path or dir_path.

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.

Description

• Enable training with netspresso-trainer on pynp.

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.

Description

• Improve the user's guide of PyNetsPresso.

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.

Description

• Show the remaining credits of the user after using the module.

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.

Description

• The input is received in the file_path, and the output is received in the dir_path.

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.

Description

• Add visualization of result on Jupyter notebook.(ex. latency, metrics, flops, etc)

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.

Description

• Move the enum file in the trainer to a common enum.

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.

Description

• Currently, PyTorch does not support Nuclear Norm(PR_NN).

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.

Description

• Add npy file transfer for int8 quantization.

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.

Description

• Change it to Pathlib because we are using both os.path and Pathlib.

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.

• I have searched to see if a similar issue already exists.

Is there any feature that you would like to add?
PyNetspresso does not have a release note. It should add a release note containing a previous version.

Is there any suggestion (or solution) to solve this issue?
Please write a clear and concise description.

Additional context
All other contexts or screenshots are welcome.

Problem

From @illian01

Compressor

• Current API gets input tensor as input_shape
  • Corresponding line
  • Generate input tensor with float32 with the given size
• Some models can be saved with half() (i.e. fp16), which cannot infer with float32 input tensor .
  • This model should give the example input tensor whose dtype is same with the model (float16)
• Need additional feature for users to select the data type of example input tensor
• Our back-end has already supported different input types of tensors, and it doesn't affect to compress the model

Related Link

Slack: https://nota-workspace.slack.com/archives/C040F65LSAJ/p1693546735327899

Description

• Improve netspresso interface

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.

Description

• The module's artifacts are stored locally and the lookup function is no longer required.
  • ex) get_models, get_compression .. etc

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.

Description

• Add github_download to download netspresso-trainer config folder

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.

Description

• Improve compression, convert, and benchmark usability.

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.

Description

• The compression option value for the example code must be set to Default.

As-is

OPTIONS = Options(
    policy=Policy.AVERAGE, layer_norm=LayerNorm.TSS_NORM, group_policy=GroupPolicy.COUNT, reshape_channel_axis=-1
)

To-be

Options(
    policy=Policy.AVERAGE, layer_norm=LayerNorm.STANDARD_SCORE, group_policy=GroupPolicy.AVERAGE, reshape_channel_axis=-1
)

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.

Description

• Apply the v0.1.2 version of the trainer.

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.

Description

• Rename model_trainer.py to trainer.py.

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.

Compressor API requests the specific type of pytorch checkpoint, fx graphmodule, but there's no guide in README

According to docs, the model can be converted with the following commands:

Compressor

• pytorch checkpoint (state_dict) -> fx graphmodule

import torch.fx
from torchvision.models import resnet18, ResNet18_Weights

model = resnet18(weights=ResNet18_Weights)
graph = torch.fx.Tracer().trace(model)
traced_model = torch.fx.GraphModule(model, graph)
torch.save(traced_model, "resnet18.pt")

Launcher

• pytorch checkpoint (state_dict) -> onnx

from torchvision.models import resnet18
import torch
from torch.onnx import TrainingMode

input_tensor = torch.rand(torch.Size([1, 3, 224, 224]))
model = resnet18(pretrained=True)
dummy_output = model(input_tensor)
torch.onnx.export(model, input_tensor, "resnet18.onnx", verbose=True, training=TrainingMode.TRAINING)

I'll add some descriptions about torch checkpoint and checkpoint converting features if needed.

Description

• Add typing to a function or parameter.

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.

Description

• Divide Launcher into Converter and Benchmarker.
  • Launcher -> Converter & Benchmarker

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.

Description

• Change YOLOX to YOLOX-S.
  • Currently, trainer only provides YOLOX-S pre-trained weight.

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.

Description

• Fix reissue_token endpoint
  • token -> auth/token
• Remove instance variable(email & password)

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.

Description

• Add the paths of each module's artifacts to metadata.
  • ex) best_pt_path, best_onnx_path, compressed_model_path, converted_model_path .. etc

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.

I'm trying to use the Netspresso Python client to log in to my account using the SessionClient class, but I'm facing an SSL certificate verification error. Here are the details:

I'm using the latest version of the Netspresso Python client.
I'm using my Netspresso account email and password to log in.
I'm getting the following error message:

Login failed. Error: HTTPSConnectionPool(host='searcher.netspresso.ai', port=443): Max retries exceeded with url: /api/v1/auth/local/login (Caused by SSLError(SSLCertVerificationError(1, '[SSL: CERTIFICATE_VERIFY_FAILED] certificate verify failed: unable to get local issuer certificate (_ssl.c:1129)')))

I believe this error is related to SSL certificate verification, but I'm not sure how to fix it. Can someone from the Netspresso team help me troubleshoot this issue?

Thanks in advance for your help!

Description

• Refactoring API Clients(Compressor, Launcher)

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.

Description

• Provides readable messages and improve tone.
• Provide solutions and examples.

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.

Description

• Update example code.

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.

Description

• Remove the "Model" prefix from the name of each module.
• ex) ModelTrainer -> Trainer

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.

Description

• Remove enum value from return of module.
  • Instead of enum value, it would be nice to make it look like a string value.

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.

Description

• Add jupyter notebook examples

Checklists

• I would create a corresponding branch for this issue from the designated (mostly develop) branch.
• This issue only contains a preceding agreement between project owners.