Electra-Bert-Base-Discrim-Google: Optimized for Mobile Deployment

Language model for masked language modeling and general-purpose NLP tasks

ELECTRABERT is a lightweight BERT model designed for efficient self-supervised learning of language representations. It can be used for identify unnatural or artificially modified text and as a backbone for various NLP tasks.

This model is an implementation of Electra-Bert-Base-Discrim-Google found here.

This repository provides scripts to run Electra-Bert-Base-Discrim-Google on Qualcomm® devices. More details on model performance across various devices, can be found here.

Model Details

Model Type: Model_use_case.text_generation
Model Stats:
- Model checkpoint: google/electra-base-discriminator
- Input resolution: 1x384
- Number of parameters: 109M
- Model size (float): 417 MB

Model	Precision	Device	Chipset	Target Runtime	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit	Target Model
Electra-Bert-Base-Discrim-Google	float	QCS8275 (Proxy)	Qualcomm® QCS8275 (Proxy)	TFLITE	66.115 ms	0 - 302 MB	NPU	Electra-Bert-Base-Discrim-Google.tflite
Electra-Bert-Base-Discrim-Google	float	QCS8275 (Proxy)	Qualcomm® QCS8275 (Proxy)	QNN_DLC	65.271 ms	0 - 327 MB	NPU	Electra-Bert-Base-Discrim-Google.dlc
Electra-Bert-Base-Discrim-Google	float	QCS8450 (Proxy)	Qualcomm® QCS8450 (Proxy)	TFLITE	23.58 ms	0 - 280 MB	NPU	Electra-Bert-Base-Discrim-Google.tflite
Electra-Bert-Base-Discrim-Google	float	QCS8450 (Proxy)	Qualcomm® QCS8450 (Proxy)	QNN_DLC	24.876 ms	0 - 305 MB	NPU	Electra-Bert-Base-Discrim-Google.dlc
Electra-Bert-Base-Discrim-Google	float	QCS8550 (Proxy)	Qualcomm® QCS8550 (Proxy)	TFLITE	16.824 ms	0 - 29 MB	NPU	Electra-Bert-Base-Discrim-Google.tflite
Electra-Bert-Base-Discrim-Google	float	QCS8550 (Proxy)	Qualcomm® QCS8550 (Proxy)	QNN_DLC	16.974 ms	0 - 41 MB	NPU	Electra-Bert-Base-Discrim-Google.dlc
Electra-Bert-Base-Discrim-Google	float	QCS8550 (Proxy)	Qualcomm® QCS8550 (Proxy)	ONNX	24.091 ms	0 - 256 MB	NPU	Electra-Bert-Base-Discrim-Google.onnx.zip
Electra-Bert-Base-Discrim-Google	float	QCS9075 (Proxy)	Qualcomm® QCS9075 (Proxy)	TFLITE	98.459 ms	0 - 302 MB	NPU	Electra-Bert-Base-Discrim-Google.tflite
Electra-Bert-Base-Discrim-Google	float	QCS9075 (Proxy)	Qualcomm® QCS9075 (Proxy)	QNN_DLC	21.082 ms	0 - 331 MB	NPU	Electra-Bert-Base-Discrim-Google.dlc
Electra-Bert-Base-Discrim-Google	float	SA7255P ADP	Qualcomm® SA7255P	TFLITE	66.115 ms	0 - 302 MB	NPU	Electra-Bert-Base-Discrim-Google.tflite
Electra-Bert-Base-Discrim-Google	float	SA7255P ADP	Qualcomm® SA7255P	QNN_DLC	65.271 ms	0 - 327 MB	NPU	Electra-Bert-Base-Discrim-Google.dlc
Electra-Bert-Base-Discrim-Google	float	SA8255 (Proxy)	Qualcomm® SA8255P (Proxy)	TFLITE	16.128 ms	0 - 15 MB	NPU	Electra-Bert-Base-Discrim-Google.tflite
Electra-Bert-Base-Discrim-Google	float	SA8255 (Proxy)	Qualcomm® SA8255P (Proxy)	QNN_DLC	16.934 ms	0 - 45 MB	NPU	Electra-Bert-Base-Discrim-Google.dlc
Electra-Bert-Base-Discrim-Google	float	SA8295P ADP	Qualcomm® SA8295P	TFLITE	26.152 ms	0 - 273 MB	NPU	Electra-Bert-Base-Discrim-Google.tflite
Electra-Bert-Base-Discrim-Google	float	SA8295P ADP	Qualcomm® SA8295P	QNN_DLC	26.433 ms	0 - 296 MB	NPU	Electra-Bert-Base-Discrim-Google.dlc
Electra-Bert-Base-Discrim-Google	float	SA8650 (Proxy)	Qualcomm® SA8650P (Proxy)	TFLITE	16.8 ms	0 - 31 MB	NPU	Electra-Bert-Base-Discrim-Google.tflite
Electra-Bert-Base-Discrim-Google	float	SA8650 (Proxy)	Qualcomm® SA8650P (Proxy)	QNN_DLC	16.945 ms	0 - 45 MB	NPU	Electra-Bert-Base-Discrim-Google.dlc
Electra-Bert-Base-Discrim-Google	float	SA8775P ADP	Qualcomm® SA8775P	TFLITE	98.459 ms	0 - 302 MB	NPU	Electra-Bert-Base-Discrim-Google.tflite
Electra-Bert-Base-Discrim-Google	float	SA8775P ADP	Qualcomm® SA8775P	QNN_DLC	21.082 ms	0 - 331 MB	NPU	Electra-Bert-Base-Discrim-Google.dlc
Electra-Bert-Base-Discrim-Google	float	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	TFLITE	11.657 ms	0 - 307 MB	NPU	Electra-Bert-Base-Discrim-Google.tflite
Electra-Bert-Base-Discrim-Google	float	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	QNN_DLC	11.651 ms	0 - 332 MB	NPU	Electra-Bert-Base-Discrim-Google.dlc
Electra-Bert-Base-Discrim-Google	float	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	ONNX	16.753 ms	0 - 359 MB	NPU	Electra-Bert-Base-Discrim-Google.onnx.zip
Electra-Bert-Base-Discrim-Google	float	Samsung Galaxy S25	Snapdragon® 8 Elite For Galaxy Mobile	TFLITE	7.985 ms	0 - 303 MB	NPU	Electra-Bert-Base-Discrim-Google.tflite
Electra-Bert-Base-Discrim-Google	float	Samsung Galaxy S25	Snapdragon® 8 Elite For Galaxy Mobile	QNN_DLC	7.843 ms	0 - 335 MB	NPU	Electra-Bert-Base-Discrim-Google.dlc
Electra-Bert-Base-Discrim-Google	float	Samsung Galaxy S25	Snapdragon® 8 Elite For Galaxy Mobile	ONNX	11.213 ms	0 - 361 MB	NPU	Electra-Bert-Base-Discrim-Google.onnx.zip
Electra-Bert-Base-Discrim-Google	float	Snapdragon 8 Elite Gen 5 QRD	Snapdragon® 8 Elite Gen5 Mobile	TFLITE	6.154 ms	4 - 308 MB	NPU	Electra-Bert-Base-Discrim-Google.tflite
Electra-Bert-Base-Discrim-Google	float	Snapdragon 8 Elite Gen 5 QRD	Snapdragon® 8 Elite Gen5 Mobile	QNN_DLC	5.908 ms	0 - 342 MB	NPU	Electra-Bert-Base-Discrim-Google.dlc
Electra-Bert-Base-Discrim-Google	float	Snapdragon 8 Elite Gen 5 QRD	Snapdragon® 8 Elite Gen5 Mobile	ONNX	8.275 ms	0 - 367 MB	NPU	Electra-Bert-Base-Discrim-Google.onnx.zip
Electra-Bert-Base-Discrim-Google	float	Snapdragon X Elite CRD	Snapdragon® X Elite	QNN_DLC	17.66 ms	1034 - 1034 MB	NPU	Electra-Bert-Base-Discrim-Google.dlc
Electra-Bert-Base-Discrim-Google	float	Snapdragon X Elite CRD	Snapdragon® X Elite	ONNX	24.045 ms	221 - 221 MB	NPU	Electra-Bert-Base-Discrim-Google.onnx.zip

Installation

Install the package via pip:

# NOTE: 3.10 <= PYTHON_VERSION < 3.14 is supported.
pip install "qai-hub-models[electra-bert-base-discrim-google]"

Configure Qualcomm® AI Hub Workbench to run this model on a cloud-hosted device

Sign-in to Qualcomm® AI Hub Workbench with your Qualcomm® ID. Once signed in navigate to Account -> Settings -> API Token.

With this API token, you can configure your client to run models on the cloud hosted devices.

qai-hub configure --api_token API_TOKEN

Navigate to docs for more information.

Demo off target

The package contains a simple end-to-end demo that downloads pre-trained weights and runs this model on a sample input.

python -m qai_hub_models.models.electra_bert_base_discrim_google.demo

The above demo runs a reference implementation of pre-processing, model inference, and post processing.

NOTE: If you want running in a Jupyter Notebook or Google Colab like environment, please add the following to your cell (instead of the above).

%run -m qai_hub_models.models.electra_bert_base_discrim_google.demo

Run model on a cloud-hosted device

In addition to the demo, you can also run the model on a cloud-hosted Qualcomm® device. This script does the following:

Performance check on-device on a cloud-hosted device
Downloads compiled assets that can be deployed on-device for Android.
Accuracy check between PyTorch and on-device outputs.

python -m qai_hub_models.models.electra_bert_base_discrim_google.export

How does this work?

This export script leverages Qualcomm® AI Hub to optimize, validate, and deploy this model on-device. Lets go through each step below in detail:

Step 1: Compile model for on-device deployment

To compile a PyTorch model for on-device deployment, we first trace the model in memory using the jit.trace and then call the submit_compile_job API.

import torch

import qai_hub as hub
from qai_hub_models.models.electra_bert_base_discrim_google import Model

# Load the model
torch_model = Model.from_pretrained()

# Device
device = hub.Device("Samsung Galaxy S25")

# Trace model
input_shape = torch_model.get_input_spec()
sample_inputs = torch_model.sample_inputs()

pt_model = torch.jit.trace(torch_model, [torch.tensor(data[0]) for _, data in sample_inputs.items()])

# Compile model on a specific device
compile_job = hub.submit_compile_job(
    model=pt_model,
    device=device,
    input_specs=torch_model.get_input_spec(),
)

# Get target model to run on-device
target_model = compile_job.get_target_model()

Step 2: Performance profiling on cloud-hosted device

After compiling models from step 1. Models can be profiled model on-device using the target_model. Note that this scripts runs the model on a device automatically provisioned in the cloud. Once the job is submitted, you can navigate to a provided job URL to view a variety of on-device performance metrics.

profile_job = hub.submit_profile_job(
    model=target_model,
    device=device,
)

Step 3: Verify on-device accuracy

To verify the accuracy of the model on-device, you can run on-device inference on sample input data on the same cloud hosted device.

input_data = torch_model.sample_inputs()
inference_job = hub.submit_inference_job(
    model=target_model,
    device=device,
    inputs=input_data,
)
    on_device_output = inference_job.download_output_data()

With the output of the model, you can compute like PSNR, relative errors or spot check the output with expected output.

Note: This on-device profiling and inference requires access to Qualcomm® AI Hub Workbench. Sign up for access.

Run demo on a cloud-hosted device

You can also run the demo on-device.

python -m qai_hub_models.models.electra_bert_base_discrim_google.demo --eval-mode on-device

NOTE: If you want running in a Jupyter Notebook or Google Colab like environment, please add the following to your cell (instead of the above).

%run -m qai_hub_models.models.electra_bert_base_discrim_google.demo -- --eval-mode on-device

Deploying compiled model to Android

The models can be deployed using multiple runtimes:

TensorFlow Lite (.tflite export): This tutorial provides a guide to deploy the .tflite model in an Android application.
QNN (.so export ): This sample app provides instructions on how to use the .so shared library in an Android application.

View on Qualcomm® AI Hub

Get more details on Electra-Bert-Base-Discrim-Google's performance across various devices here. Explore all available models on Qualcomm® AI Hub

License

The license for the original implementation of Electra-Bert-Base-Discrim-Google can be found here.
The license for the compiled assets for on-device deployment can be found here

References

Community

Join our AI Hub Slack community to collaborate, post questions and learn more about on-device AI.
For questions or feedback please reach out to us.

Downloads last month: 7

Paper for qualcomm/Electra-Bert-Base-Discrim-Google

ELECTRA: Pre-training Text Encoders as Discriminators Rather Than Generators

Paper • 2003.10555 • Published Mar 23, 2020