SparseML Question Answering Pipelines integrate with Hugging Face’s Transformers library to enable the sparsification of a large set of transformers models. Sparsification is a powerful technique that results in faster, smaller, and cheaper deployable models. A sparse model can be deployed with DeepSparse for GPU-class performance directly on your CPU.

This integration enables you to create a sparse model in two ways. Each option is useful in different situations:

Sparsification of Popular Transformer Models—Sparsify any popular Hugging Face Transformer model from scratch. This enables you to create a sparse version of any model (even those not in the SparseZoo), but requires hand-tuning the hyperparameters of the sparsification algorithm.
Sparse Transfer Learning—Fine-tune a sparse model (or use one of our sparse pre-trained models) on your own private dataset. This is the easiest path to creating a sparse model trained on your data. Simply pull a pre-sparsified model and transfer learning recipe from the SparseZoo and fine-tune on your data with a single command.

Installation Requirements

This use case requires installation of:

It is recommended to run Python 3.8 as some of the scripts within the Transformers repository require it.

Transformers will not immediately install with this command. Instead, a sparsification-compatible version of Transformers will install on the first invocation of the Transformers code in SparseML.

Tutorials

Here are additional tutorials for this functionality.

Getting Started

Sparsifying Popular Transformer Models

In the example below, a dense BERT model is sparsified and fine-tuned on the SQuAD dataset.

1sparseml.transformers.question_answering \
2  --model_name_or_path bert-base-uncased \
3  --dataset_name squad \
4  --do_train \
5  --do_eval \
6  --output_dir './output' \
7  --cache_dir cache \
8  --distill_teacher disable \
9  --recipe zoo:nlp/question_answering/bert-base/pytorch/huggingface/squad/pruned-aggressive_98

The SparseML train script is a wrapper around a Hugging Face script, and usage for most arguments follows the Hugging Face. The most important arguments for SparseML are:

--model_name_or_path indicates the model from which to start the pruning process. It can be a SparseZoo stub, HF model identifier, or a path to a local model.
--recipe points to a recipe file containing the sparsification hyperparameters. It can be a SparseZoo stub or a local file. See Creating Sparsification Recipes for more information.
--dataset_name indicates that we should fine-tune on the SQuAD dataset.

To utilize a custom dataset, use the --train_file and --validation_file arguments. To use a dataset from the Hugging Face hub, use --dataset_name. See the Hugging Face documentation for more details.

Run the following to see the full list of options:

$ sparseml.transformers.question_answering -h

Sparse Transfer Learning

SparseML also enables you to fine-tune a pre-sparsified model onto your own dataset. While you are free to use your backbone, we encourage you to leverage one of our sparse pre-trained models to boost your productivity!

In the example below, we fetch a pruned, quantized BERT model, pre-trained on Wikipedia and Bookcorpus datasets. We then fine-tune the model to the SQuAD dataset.

1sparseml.transformers.question_answering \
2    --model_name_or_path zoo:nlp/masked_language_modeling/bert-base/pytorch/huggingface/wikipedia_bookcorpus/12layer_pruned80_quant-none-vnni \
3    --dataset_name squad \
4    --do_train \
5    --do_eval \
6    --output_dir './output' \
7    --distill_teacher disable \
8    --recipe zoo:nlp/masked_language_modeling/bert-base/pytorch/huggingface/wikipedia_bookcorpus/12layer_pruned80_quant-none-vnni?recipe_type=transfer-question_answering

The usage of the script is the same as for Sparsifying Popular Transformer Models, above. However, in this example, the starting model is a pruned-quantized version of BERT from the SparseZoo (rather than a dense BERT model) and the recipe is a transfer learning recipe, which instructs Transformers to maintain sparsity as it fine-tunes (rather than a recipe that sparsifies a model from scratch).

Knowledge Distillation

By modifying the distill_teacher argument, you can enable Knowledge Distillation (KD) functionality. KD provides additional support to the sparsification or transfer learning process, enabling higher accuracy at higher levels of sparsity.

For example, the --distill_teacher argument can be set to pull a dense SQuAD model from the SparseZoo to support the training process:

--distill_teacher zoo:nlp/question_answering/bert-base/pytorch/huggingface/squad/base-none

Alternatively, SparseML enables you to use your a custom dense teacher model. The following command uses the dense BERT base model from the SparseZoo and fine-tunes it on the SQuAD dataset for use as a dense teacher.

1sparseml.transformers.question_answering \
2    --model_name_or_path zoo:nlp/masked_language_modeling/bert-base/pytorch/huggingface/wikipedia_bookcorpus/base-none \
3    --dataset_name squad \
4    --do_train \
5    --do_eval \
6    --output_dir models/teacher \
7    --recipe zoo:nlp/masked_language_modeling/bert-base/pytorch/huggingface/wikipedia_bookcorpus/base-none?recipe_type=transfer-question_answering

Once the dense teacher is trained, you may reuse it for KD in sparsification or sparse transfer learning. Simply pass the path to the directory with the teacher model to the --distill_teacher argument. For example:

--distill_teacher models/teacher

SparseML CLI

The SparseML installation provides a CLI for sparsifying your models for a specific task. Appending the --help argument displays a full list of options for training in SparseML:

sparseml.transformers.question_answering --help

The output is:

1  --model_name_or_path MODEL_NAME_OR_PATH
2                        Path to pre-trained model or model identifier from huggingface.co/models
3  --distill_teacher DISTILL_TEACHER
4                        Teacher model which needs to be a trained QA model
5  --cache_dir CACHE_DIR
6                        Directory path to store the pre-trained models downloaded from huggingface.co
7  --recipe RECIPE
8                        Path to a SparseML sparsification recipe, see https://github.com/neuralmagic/sparseml for more information
9  --dataset_name DATASET_NAME
10                        The name of the dataset to use (via the datasets library).
11  ...

To learn about the Hugging Face Transformers run-scripts in more detail, refer to Hugging Face Transformers documentation.

Deploying with DeepSparse

The artifacts of the training process are saved to the directory --output_dir. Once the script terminates, the directory will have everything required to deploy or further modify the model such as:

The recipe (with the full description of the sparsification attributes)
Checkpoint files (saved in the appropriate framework format)
Additional configuration files (e.g., tokenizer, dataset info)

Exporting the Sparse Model to ONNX

DeepSparse uses the ONNX format to load neural networks and then deliver breakthrough performance for CPUs by leveraging the sparsity and quantization within a network.

The SparseML installation provides a sparseml.transformers.export_onnx command that you can use to load the training model folder and create a new model.onnx file within. Be sure the --model_path argument points to your trained model.

1sparseml.transformers.export_onnx \
2    --model_path './output' \
3    --task 'question-answering'

DeepSparse Deployment

Once the model is exported in the ONNX format, it is ready for deployment with DeepSparse.

The deployment is intuitive due to the DeepSparse Python API.

1from deepsparse import Pipeline
2
3qa_pipeline = Pipeline.create(
4  task="question-answering",
5  model_path='./output'
6)
7
8inference = qa_pipeline(question="What's my name?", context="My name is Snorlax")
>> {'score': 0.9947717785835266, 'start': 11, 'end': 18, 'answer': 'Snorlax'}

To learn more, refer to the appropriate documentation in the DeepSparse repository.

Support

For Neural Magic Support, sign up or log into our Neural Magic Community Slack. Bugs, feature requests, or additional questions can also be posted to our GitHub Issue Queue.

Deploy an Object Detection Model

NLP Text Classification