Skip to content

api-reference.md

Latest commit

 

History

History
257 lines (202 loc) · 4.1 KB

api-reference.md

File metadata and controls

257 lines (202 loc) · 4.1 KB
Raw

API Reference

Overview

BERT Studio provides a RESTful API built with FastAPI for all backend operations. The API handles model management, task execution, custom code processing, and MongoDB operations.

Base URL

  • Development: http://localhost:8000
  • Production: https://your-domain.com/api

Authentication

Most endpoints require API key authentication via session management.

Login

POST /auth/login
Content-Type: application/json

{
  "api_key": "your-api-key"
}

Logout

POST /auth/logout

Model Management

List Available Models

GET /models

Download Model

POST /models/download
Content-Type: application/json

{
  "model_name": "distilbert-base-uncased",
  "task_type": "classification"
}

Get Model Info

GET /models/{model_name}/info

Task Execution

Text Classification

POST /tasks/classification
Content-Type: application/json

{
  "text": "This is a great product!",
  "model_name": "distilbert-base-uncased"
}

Question Answering

POST /tasks/qa
Content-Type: application/json

{
  "question": "What is BERT?",
  "context": "BERT is a transformer model...",
  "model_name": "bert-large-uncased-whole-word-masking-finetuned-squad"
}

Named Entity Recognition

POST /tasks/ner
Content-Type: application/json

{
  "text": "John works at Google in California",
  "model_name": "dbmdz/bert-large-cased-finetuned-conll03-english"
}

Text Summarization

POST /tasks/summarization
Content-Type: application/json

{
  "text": "Long text to summarize...",
  "model_name": "facebook/bart-large-cnn",
  "max_length": 150,
  "min_length": 50
}

Generate Embeddings

POST /tasks/embeddings
Content-Type: application/json

{
  "text": "Text to embed",
  "model_name": "sentence-transformers/all-MiniLM-L6-v2"
}

Custom Tasks

Execute Custom Code

POST /tasks/custom
Content-Type: application/json

{
  "tokenizer_code": "tokenizer = AutoTokenizer.from_pretrained('model-name')",
  "model_code": "model = AutoModel.from_pretrained('model-name')",
  "function_code": "def custom_function(text): return {'result': 'processed'}",
  "input_text": "Input text here"
}

Save Custom Task

POST /custom-tasks
Content-Type: application/json

{
  "name": "My Custom Task",
  "description": "Description of the task",
  "tags": ["nlp", "classification"],
  "tokenizer_code": "...",
  "model_code": "...",
  "function_code": "...",
  "model_name": "distilbert-base-uncased"
}

List Custom Tasks

GET /custom-tasks?search=query&tags=nlp,classification&model=distilbert-base-uncased

Get Custom Task

GET /custom-tasks/{task_id}

Update Custom Task

PUT /custom-tasks/{task_id}
Content-Type: application/json

{
  "name": "Updated Task Name",
  "description": "Updated description",
  "tags": ["updated", "tags"]
}

Delete Custom Task

DELETE /custom-tasks/{task_id}

Task Management

Export Tasks

POST /custom-tasks/export
Content-Type: application/json

{
  "task_ids": ["task1_id", "task2_id"]
}

Import Tasks

POST /custom-tasks/import
Content-Type: multipart/form-data

file: tasks.json

Get Statistics

GET /custom-tasks/stats

Health Check

API Health

GET /health

Database Health

GET /health/db

Error Responses

All error responses follow this format:

{
  "error": "Error message",
  "detail": "Detailed error information",
  "status_code": 400
}

Rate Limiting

  • 100 requests per minute per IP
  • 1000 requests per hour per authenticated user
  • Custom tasks limited to 10 concurrent executions

Response Formats

Successful Response

{
  "success": true,
  "data": { ... },
  "message": "Operation completed successfully"
}

Error Response

{
  "success": false,
  "error": "Error description",
  "status_code": 400
}

WebSocket Endpoints

Real-time Task Updates

ws://localhost:8000/ws/tasks/{task_id}

Model Download Progress

ws://localhost:8000/ws/download/{download_id}