Model Training and Management: Documentation

[KDwevedi]

Description

This ticket is to be used to collate documentation being generated for #62

Model Specification Sheet

Documented Entities:

• Tasks
• Models
• Datasets
• Label Studio UI Elements
• Internal Autotune JSON schemas being used to store Tasks, Models
• API interfaces
• JSON Schemas

[KDwevedi]

Autotune Docs

Train

"model"

Workflow

Models

Dataset

LStudio Components | Config

NER

{
  "model": "",
  "task": "NER",
  "labelStudioElement": {
    "name": "Named Entity Recognition",
    "config": {
      "labels":[
        {
          "name": "",
          "value": ""
        }
        ]
    
    }
  },
  "telemetryDataField": {
    "input": "",
    "output": ""
  }
}

Translation

{
  "model": "",
  "task": "Translation",
  "labelStudioElement": {
    "name": "Translation",
    "config": {
      "from": "english",
      "to": "odia"
    }
  },
  "telemetryDataField": {
    "input": "inputFieldName",
    "output": "outputFieldName"
  }
}

Document Retrieval

{
  "model": "",
  "task": "",
  "labelStudioElement": {
    "name": "Document Retrieval",
    "config": {  
    }
  },
  "telemetryDataField": {
    "input": "inputFieldName",
    "output": "outputFieldName"
}

Text Classification

{
  "model": "",
  "task": "",
  "labelStudioElement": {
    "name": "Text Classification",
    "config": {
      "choices": ["Choice 1", "Choice 2"]
    }
  },
  "telemetryDataField": {
    "input": "inputFieldName",
    "output": "outputFieldName"
}

[KDwevedi]

Model	Task	Input	Input Type	Input Example	Output	Output Type	Output Example	Is Being Recorded	Telemetry Event ID	Telemetry Event Field
symspell	Spell check (agri-Odia,English)		text		text		No

[KDwevedi]

Fields In Admin Panel Valid for feedback:

• translatedQuery
• spellCorrectedText
• translatedResponse
• coreferencedText
• queryClass
• NER
• response
• detectedLanguage
• similarChunks

C.C. @prtkjakhar
@Shryansh107 @Karanpal97
Can we start adding Label Studio elements and their configs here like so:

Label Studio Elements

Translation Component: Machine Translation

Labeling Configuration | Playground

Frontend Return Type:

{ 


  "Record ID":"",
  "Query":"Hello"
}

Example JSON Response

1. Input Preview

{
    "english": "To have faith is to trust yourself to the water"
}

2. Output Preview

[
    {
        "value": {
            "text": [
                "ବିଶ୍ୱାସ ରହିବା ହେଉଛି ନିଜକୁ ପାଣି ଉପରେ ବିଶ୍ୱାସ କରିବା |"
            ]
        },
        "meta": {
            "lead_time": 1.408
        },
        "id": "O91KmzN127",
        "from_name": "odia",
        "to_name": "english",
        "type": "textarea"
    }
]

[KDwevedi]

closing this as outdated