Skip to content

Spacy Inference Pipeline

This example showcases how to run multi-step ML pipeline to prepare data and run spacy Named Entity Recognition (NER) model inference within Hera / Argo Workflows! Step 1: Prepares dataset using Spacy example sentences library and saves dataset the volume path /mnt/data Step 2: Performs inference on the dataset in the volume path /mnt/data using Spacy Named Entity Recognition (NER) LLM model

  1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
from hera.workflows import (
    Resources,
    Steps,
    Volume,
    Workflow,
    models as m,
    script,
)


@script(
    image="jupyter/datascience-notebook:latest",
    resources=Resources(cpu_request=0.5, memory_request="1Gi"),
    volume_mounts=[m.VolumeMount(name="data-dir", mount_path="/mnt/data")],
)
def data_prep() -> None:
    import json
    import subprocess

    from spacy.lang.en.examples import sentences

    print(subprocess.run("cd /mnt/data && ls -l", shell=True, capture_output=True).stdout.decode())
    """ the used image does not have `spacy` installed, so we need to install it first! """
    subprocess.run(
        ["pip", "install", "spacy"],
        stdout=subprocess.PIPE,
        universal_newlines=True,
    )

    """ dumping spacy example sentences data into a file
        replace this with real dataset """
    with open("/mnt/data/input_data.json", "w") as json_file:
        json.dump(sentences, json_file)
    print("Data preparation completed")
    print(subprocess.run("cd /mnt/data && ls -l", shell=True, capture_output=True).stdout.decode())


@script(
    image="jupyter/datascience-notebook:latest",
    resources=Resources(cpu_request=0.5, memory_request="1Gi"),
    volume_mounts=[m.VolumeMount(name="data-dir", mount_path="/mnt/data")],
)
def inference_spacy() -> None:
    import subprocess

    """ the used image does not have `spacy` installed, so we need to install it first! """
    subprocess.run(
        ["pip", "install", "spacy"],
        stdout=subprocess.PIPE,
        universal_newlines=True,
    )
    print(subprocess.run("cd /mnt/data && ls -l ", shell=True, capture_output=True).stdout.decode())
    import json
    from typing import List

    import pydantic
    import spacy
    from pydantic import BaseModel
    from spacy.cli import download

    """ download and load spacy model https://spacy.io/models/en#en_core_web_lg """
    spacy_model_name = "en_core_web_lg"
    download(spacy_model_name)
    nlp = spacy.load(spacy_model_name)

    """ build pydantic model """
    print(pydantic.version.version_info())

    class NEROutput(BaseModel):
        input_text: str
        ner_entities: List[str] = []

    ner_output_list: List[NEROutput] = []

    """ read data prepared from previous step data_prep """
    with open("/mnt/data/input_data.json", "r") as json_file:
        input_data = json.load(json_file)
        print(input_data)
        """ iterate each sentence in the data and perform NER """
        for sentence in input_data:
            print("input text: " + sentence)
            doc = nlp(sentence)
            print("output NER:")
            ner_entities: List[str] = []
            for entity in doc.ents:
                """ Print the entity text and its NER label """
                ner_entity = entity.text + " is " + entity.label_
                print(ner_entity)
                ner_entities.append(ner_entity)
            print("ner_entities = + " + ner_entities)
            ner_output = NEROutput(input_text=sentence, ner_entities=ner_entities)
            ner_output_list.append(dict(ner_output))
        print("ner_output_list = " + ner_output_list)
    print("Inference completed")
    """ save output in a file """
    with open("/mnt/data/output_data.json", "w") as json_file:
        json.dump(ner_output_list, json_file)
    print(subprocess.run("cd /mnt/data && ls -l ", shell=True, capture_output=True).stdout.decode())


with Workflow(
    generate_name="spacy-inference-pipeline-",
    entrypoint="spacy-inference-pipeline",
    volumes=[Volume(name="data-dir", size="1Gi", mount_path="/mnt/data")],
    service_account_name="hera",
    namespace="argo",
) as w:
    with Steps(name="spacy-inference-pipeline") as steps:
        data_prep(name="data-prep")
        inference_spacy(name="inference-spacy")

  1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
apiVersion: argoproj.io/v1alpha1
kind: Workflow
metadata:
  generateName: spacy-inference-pipeline-
  namespace: argo
spec:
  entrypoint: spacy-inference-pipeline
  serviceAccountName: hera
  templates:
  - name: spacy-inference-pipeline
    steps:
    - - name: data-prep
        template: data-prep
    - - name: inference-spacy
        template: inference-spacy
  - name: data-prep
    script:
      image: jupyter/datascience-notebook:latest
      source: |-
        import os
        import sys
        sys.path.append(os.getcwd())
        import json
        import subprocess
        from spacy.lang.en.examples import sentences
        print(subprocess.run('cd /mnt/data && ls -l', shell=True, capture_output=True).stdout.decode())
        ' the used image does not have `spacy` installed, so we need to install it first! '
        subprocess.run(['pip', 'install', 'spacy'], stdout=subprocess.PIPE, universal_newlines=True)
        ' dumping spacy example sentences data into a file\n        replace this with real dataset '
        with open('/mnt/data/input_data.json', 'w') as json_file:
            json.dump(sentences, json_file)
        print('Data preparation completed')
        print(subprocess.run('cd /mnt/data && ls -l', shell=True, capture_output=True).stdout.decode())
      command:
      - python
      volumeMounts:
      - name: data-dir
        mountPath: /mnt/data
      resources:
        requests:
          cpu: '0.5'
          memory: 1Gi
  - name: inference-spacy
    script:
      image: jupyter/datascience-notebook:latest
      source: |-
        import os
        import sys
        sys.path.append(os.getcwd())
        import subprocess
        ' the used image does not have `spacy` installed, so we need to install it first! '
        subprocess.run(['pip', 'install', 'spacy'], stdout=subprocess.PIPE, universal_newlines=True)
        print(subprocess.run('cd /mnt/data && ls -l ', shell=True, capture_output=True).stdout.decode())
        import json
        from typing import List
        import pydantic
        import spacy
        from pydantic import BaseModel
        from spacy.cli import download
        ' download and load spacy model https://spacy.io/models/en#en_core_web_lg '
        spacy_model_name = 'en_core_web_lg'
        download(spacy_model_name)
        nlp = spacy.load(spacy_model_name)
        ' build pydantic model '
        print(pydantic.version.version_info())

        class NEROutput(BaseModel):
            input_text: str
            ner_entities: List[str] = []
        ner_output_list: List[NEROutput] = []
        ' read data prepared from previous step data_prep '
        with open('/mnt/data/input_data.json', 'r') as json_file:
            input_data = json.load(json_file)
            print(input_data)
            ' iterate each sentence in the data and perform NER '
            for sentence in input_data:
                print('input text: ' + sentence)
                doc = nlp(sentence)
                print('output NER:')
                ner_entities: List[str] = []
                for entity in doc.ents:
                    ' Print the entity text and its NER label '
                    ner_entity = entity.text + ' is ' + entity.label_
                    print(ner_entity)
                    ner_entities.append(ner_entity)
                print('ner_entities = + ' + ner_entities)
                ner_output = NEROutput(input_text=sentence, ner_entities=ner_entities)
                ner_output_list.append(dict(ner_output))
            print('ner_output_list = ' + ner_output_list)
        print('Inference completed')
        ' save output in a file '
        with open('/mnt/data/output_data.json', 'w') as json_file:
            json.dump(ner_output_list, json_file)
        print(subprocess.run('cd /mnt/data && ls -l ', shell=True, capture_output=True).stdout.decode())
      command:
      - python
      volumeMounts:
      - name: data-dir
        mountPath: /mnt/data
      resources:
        requests:
          cpu: '0.5'
          memory: 1Gi
  volumeClaimTemplates:
  - metadata:
      name: data-dir
    spec:
      accessModes:
      - ReadWriteOnce
      resources:
        requests:
          storage: 1Gi

Comments