kserve/docs/samples/v1beta1/torchserve/v2/bert/README.md

TorchServe example with Huggingface bert model

In this example we will show how to serve Huggingface Transformers with TorchServe on KServe.

Model archive file creation

Clone pytorch/serve repository, navigate to examples/Huggingface_Transformers and follow the steps for creating the MAR file including serialized model and other dependent files. TorchServe supports both eager model and torchscript and here we save as the pretrained model. Download the preprocess script from here

torch-model-archiver --model-name BERTSeqClassification --version 1.0 \
--serialized-file Transformer_model/pytorch_model.bin \
--handler ./Transformer_kserve_handler.py \
--extra-files "Transformer_model/config.json,./setup_config.json,./Seq_classification_artifacts/index_to_name.json,./Transformer_handler_generalized.py"

Create the InferenceService

Apply the CRD

kubectl apply -f bert.yaml

Expected Output

$inferenceservice.serving.kserve.io/torchserve-bert-v2 created

For deploying it in gpu use

kubectl apply -f bert_gpu.yaml

Run a prediction

The first step is to determine the ingress IP and ports and set INGRESS_HOST and INGRESS_PORT

Setting variables

MODEL_NAME=BERTSeqClassification
SERVICE_HOSTNAME=$(kubectl get inferenceservice torchserve-bert-v2 -o jsonpath='{.status.url}' | cut -d "/" -f 3)

curl -v -H "Host: ${SERVICE_HOSTNAME}" http://${INGRESS_HOST}:${INGRESS_PORT}/v2/models/BERTSeqClassification/infer -d @./sequence_classification/bytes/bert_v2.json

Expected Output

{"id": "d3b15cad-50a2-4eaf-80ce-8b0a428bd298", "model_name": "BERTSeqClassification", "model_version": "1.0", "outputs": [{"name": "predict", "shape": [], "datatype": "BYTES", "data": ["Not Accepted"]}]}

For tensor input

curl -v -H "Host: ${SERVICE_HOSTNAME}" http://${INGRESS_HOST}:${INGRESS_PORT}/v2/models/BERTSeqClassification/infer -d @./sequence_classification/tensor/bert_v2.json

Expected output

{"id": "33abc661-7265-42fc-b7d9-44e5f79a7a67", "model_name": "BERTSeqClassification", "model_version": "1.0", "outputs": [{"name": "predict", "shape": [], "datatype": "BYTES", "data": ["Not Accepted"]}]}

Captum Explanations

In order to understand the word importances and attributions when we make an explanation Request, we use Captum Insights for the HuggingFace Transformers pre-trained model.

curl -v -H "Host: ${SERVICE_HOSTNAME}" http://${INGRESS_HOST}:${INGRESS_PORT}/v2/models/BERTSeqClassification/explain -d ./sequence_classification/bytes/bert_v2.json

Expected output

{"id": "d3b15cad-50a2-4eaf-80ce-8b0a428bd298", "model_name": "BERTSeqClassification", "model_version": "1.0", "outputs": [{"name": "explain", "shape": [], "datatype": "BYTES", "data": [{"words": ["[CLS]", "bloomberg", "has", "decided", "to", "publish", "a", "new", "report", "on", "the", "global", "economy", ".", "[SEP]"], "importances": [0.0, -0.43571255624310423, -0.11062097534384648, 0.11323803203829622, 0.05438679692935377, -0.11364841625009202, 0.15214504085858935, -0.0013061684457894148, 0.05712844103997178, -0.02296408323390218, 0.1937543236757826, -0.12138265438655091, 0.20713335609474381, -0.8044260616647264, 0.0], "delta": -0.019047775223331675}]}]}