all-the-public-replicate-models

Metadata for all[^1] the public models on Replicate, bundled up into an npm package.

Installation

npm install all-the-public-replicate-models

Usage (as a library)

Full-bodied usage (all the metadata, ~17MB)

import models from 'all-the-public-replicate-models'

console.log(models)

Lite usage (just the basic metadata, ~375K):

import models from 'all-the-public-replicate-models/lite'

console.log(models)

Find the top 10 models by run count:

import models from 'all-the-public-replicate-models'
import {chain} from 'lodash-es'

const mostRun = chain(models).orderBy('run_count', 'desc').take(10).value()
console.log({mostRun})

Usage (as a CLI)

The CLI dumps the model metadata to standard output as a big JSON object:

$ npx all-the-public-replicate-models

The output will be:

[
  {...},
  {...},
  {...},
]

You can use jq to filter the output. Here's an example that finds all the whisper models and sorts them by run count:

npx all-the-public-replicate-models | jq -r 'map(select(.name | contains("whisper"))) | sort_by(.run_count) | reverse | .[] | "\(.url) \(.run_count)"'

• https://replicate.com/openai/whisper 3790120
• https://replicate.com/m1guelpf/whisper-subtitles 38020
• https://replicate.com/hnesk/whisper-wordtimestamps 28889
• https://replicate.com/alqasemy2020/whisper-jax 20296
• https://replicate.com/wglodell/cog-whisperx-withprompt 19326
• https://replicate.com/daanelson/whisperx 15528 ...

Or you can dump all the model data to a file:

npx all-the-public-replicate-models > models.json

[^1]: Technically it's not all the models, but every model that is public, has at least one published version, and has at least one example prediction.