Replicate Model Classifier

An HTTP API to classify Replicate models into Hugging Face tasks using a language model.

Cloudflare Workers for hosting the HTTP API
Hono for authoring the HTTP API
Anthropic Claude 3.7 Sonnet for model classification
Replicate API for model metadata
Hugging Face Tasks for model task metadata
Cloudflare D1 for caching classifications

Repository: https://github.com/zeke/replicate-model-classifier

API

Base URL: https://replicate-model-classifier.ziki.workers.dev/

Classify a model

GET /api/models/:owner/:model

Returns a JSON object with the model classification. Responses are cached.

Cache Busting

To force a fresh classification and bypass the cache, add the bust=1 query parameter:

GET /api/models/:owner/:model?bust=1

This will skip the cache and trigger a new classification, updating the cache with the latest result.

Example:

/api/models/salesforce/blip?bust=1

Output Format

The response includes a classification object with the following fields:

summary: A short summary of what the model does in 10 words or less. Not a sales pitch.
inputTypes: An array of the types of inputs the model accepts, like "text", "image", "audio", etc.
outputTypes: An array of the types of outputs the model returns, like "text", "image", "audio", etc.
task: The Hugging Face task name the model performs.
categories: An array of 5 categories the model belongs to (e.g., "image", "audio", "diffusion").
useCases: An array of 10 use cases for the model. Each is a single sentence of 8 words or less.

Example response:

{
  "model": "salesforce/blip",
  "classification": {
    "summary": "Generate image captions and answer questions about images",
    "inputTypes": ["image", "text"],
    "outputTypes": ["text"],
    "task": "visual-question-answering",
    "categories": ["image", "captioning", "vision", "qa", "multimodal"],
    "useCases": [
      "Generate image captions for social media",
      "Answer questions about medical images",
      "Create alt text for accessibility",
      "Analyze security camera footage",
      "Describe artwork for museums",
      "Generate product descriptions",
      "Answer questions about diagrams",
      "Create image-based quizzes",
      "Analyze satellite imagery",
      "Describe scenes in videos"
    ]
  }
}

The response includes cache headers:

X-Cache: Either "HIT" or "MISS" to indicate if the response came from cache
Cache-Control: "public, max-age=315360000" (10 years)

Examples

View the prompt

To get a pretty-printed view of the prompt that was used to classify the model, add the prompt query parameter:

GET /api/models/:owner/:model?prompt=1

Examples

View everything

It can be helpful to see all the data that goes into the model classification. You can see all the data by adding the debug query parameter:

GET /api/models/:owner/:model?debug=1

Examples:

View all classifications

Get all cached model classifications:

GET /api/classifications

See /api/classifications

Returns a JSON object with all cached classifications:

{
  "classifications": [
    {
      "model": "salesforce/blip",
      "classification": {
        "summary": "Generate image captions and answer questions about images",
        "inputTypes": ["image", "text"],
        "outputTypes": ["text"],
        "task": "visual-question-answering",
        "taskSummary": "Visual Question Answering is the task of answering open-ended questions based on an image. They output natural language responses to natural language questions.",
        "useCases": [
          "Generate image captions for social media",
          "Answer questions about medical images",
          "Create alt text for accessibility",
          "Analyze security camera footage",
          "Describe artwork for museums",
          "Generate product descriptions",
          "Answer questions about diagrams",
          "Create image-based quizzes",
          "Analyze satellite imagery",
          "Describe scenes in videos"
        ]
      },
      "createdAt": "2024-04-29T04:00:00.000Z",
      "updatedAt": "2024-04-29T04:00:00.000Z"
    }
  ]
}

View cache statistics

Get statistics about the classification cache:

GET /api/cache/stats

Returns a JSON object with cache statistics:

{
  "total": 100,
  "oldest": {
    "model": "salesforce/blip",
    "createdAt": "2024-04-29T04:00:00.000Z"
  },
  "newest": {
    "model": "meta/meta-llama-3-8b-instruct",
    "createdAt": "2024-04-29T05:00:00.000Z"
  }
}

View Hugging Face task data

GET /api/tasks

See /api/tasks

View Hugging Face task names

GET /api/taskNames

See /api/taskNames

Use Cases

Each model classification includes a useCases array that provides 10 specific use cases for the model. These use cases are generated by the language model based on the model's capabilities, input/output types, and task classification. Each use case is a concise, single-sentence description of a practical application of the model.

The use cases are designed to be:

Specific and actionable
Diverse in their applications
Relevant to the model's capabilities
Concise (8 words or less)
Practical and real-world focused

This feature helps users quickly understand the potential applications of each model and find models that match their specific needs.