An Elixir client for Replicate. It lets you run models from your Elixir code, and everything else you can do with Replicate's HTTP API.
Install by adding replicate
to your list of dependencies in mix.exs
:
def deps do
[
{:replicate, "~> 1.3.0"}
]
end
Want to jump right in to building your own apps with Elixir and Replicate? Check out 🔮 Conjurer, a simple demo app we built with the Elixir client.
After installation, you need to set your Replicate API token in your environment.
Grab your token from replicate.com/account and set it as an environment variable:
export REPLICATE_API_TOKEN=<your token>
And run source .env
.
Then, add the config to your config.exs
:
config :replicate,
replicate_api_token: System.get_env("REPLICATE_API_TOKEN")
Now you can use Replicate
to do cool machine learny stuff.
🚨 We recommend not adding the token directly to your source code, because you don't want to put your credentials in source control. If anyone used your API key, their usage would be charged to your account.
You can run a model synchronously:
iex> Replicate.run("stability-ai/stable-diffusion:db21e45d3f7023abc2a46ee38a23973f6dce16bb082a930b0c49861f96d1e5bf", prompt: "a watercolor of the babadook by Picasso")
["https://replicate.delivery/pbxt/LgZ5BLmMWzqODp4rzSERXDNglStBTltHpj0533i9385qSgQE/out-0.png"]
You can start a model and run it in the background with Replicate.Predictions.create/5
:
iex> model = Replicate.Models.get!("stability-ai/stable-diffusion")
iex> version = Replicate.Models.get_version!(model, "db21e45d3f7023abc2a46ee38a23973f6dce16bb082a930b0c49861f96d1e5bf")
iex> {:ok, prediction} = Replicate.Predictions.create(version, %{prompt: "a 19th century portrait of a wombat gentleman"})
iex> prediction.status
"starting"
You can also load local files as input, e.g. Read File (binary for non-text files) -> Base64 Encode -> Create Data URI:
iex> binary_content = File.read!("./audio.wav")
<<79, 103, 103, 83, 0, 2, 0, 0, 0, 0, 0, 0, 0, 0, ...>>
iex> base64_content = Base.encode64(binary_content)
...
iex> data_uri = "data:audio/wav;base64," <> base64_content
...
iex> Replicate.run(
...> "vaibhavs10/incredibly-fast-whisper:3ab86df6c8f54c11309d4d1f930ac292bad43ace52d10c80d87eb258b3c9f79c",
...> audio: data_uri,
...> task: "transcribe",
...> language: "None",
... timestamp: "chunk",
... batch_size: 64,
... diarize: false
... )
...
We can take a look at the prediction:
iex> prediction
%Replicate.Predictions.Prediction{
id: "krdjxq6rw5bx3dopem52ohezca",
error: nil,
input: %{"prompt" => "a 19th century portrait of a wombat gentleman"},
logs: "",
output: nil,
status: "starting",
version: "db21e45d3f7023abc2a46ee38a23973f6dce16bb082a930b0c49861f96d1e5bf",
started_at: nil,
created_at: "2023-06-02T20:43:55.720751299Z",
completed_at: nil
}
Get the latest status of the prediction with Replicate.Predictions.get/1
:
iex> {:ok, prediction} = Replicate.Predictions.get(prediction.id)
iex> prediction.logs |> String.split("\n")
["Using seed: 54144", "input_shape: torch.Size([1, 77])",
" 0%| | 0/50 [00:00<?, ?it/s]",
" 6%|▌ | 3/50 [00:00<00:02, 21.14it/s]",
" 12%|█▏ | 6/50 [00:00<00:02, 21.18it/s]",
" 18%|█▊ | 9/50 [00:00<00:01, 21.20it/s]",
" 24%|██▍ | 12/50 [00:00<00:01, 21.25it/s]",
" 30%|███ | 15/50 [00:00<00:01, 21.21it/s]",
" 36%|███▌ | 18/50 [00:00<00:01, 21.24it/s]",
" 42%|████▏ | 21/50 [00:00<00:01, 21.23it/s]",
" 48%|████▊ | 24/50 [00:01<00:01, 21.19it/s]",
" 54%|█████▍ | 27/50 [00:01<00:01, 21.16it/s]",
" 60%|██████ | 30/50 [00:01<00:00, 21.17it/s]",
" 66%|██████▌ | 33/50 [00:01<00:00, 21.15it/s]",
" 72%|███████▏ | 36/50 [00:01<00:00, 21.17it/s]",
" 78%|███████▊ | 39/50 [00:01<00:00, 21.16it/s]",
" 84%|████████▍ | 42/50 [00:01<00:00, 21.16it/s]",
" 90%|█████████ | 45/50 [00:02<00:00, 21.16it/s]",
" 96%|█████████▌| 48/50 [00:02<00:00, 20.99it/s]",
"100%|██████████| 50/50 [00:02<00:00, 21.14it/s]",
""]
And wait for completion with Replicate.Predictions.wait/1
:
iex> {:ok, prediction} = Replicate.Predictions.wait(prediction)
iex> prediction.status
"succeeded"
iex> prediction.output
["https://replicate.delivery/pbxt/xbT582euzFwqCiupR5PVyMUFemfgZHbPyAm5kenezBS3RDQIC/out-0.png"]
You can run a model and get a webhook when it completes, instead of waiting for it to finish:
Replicate.Predictions.create(version, %{prompt: "a 19th century portrait of a wombat gentleman"}, "https://example.com/webhook", ["completed"])
If you want to see a demo of how to use webhooks in production, check out 🔮 Conjurer.
You can cancel a running prediction by passing an id or
%Replicate.Predictions.Prediction{}
to Replicate.Predictions.cancel/1
:
iex> model = Replicate.Models.get!("stability-ai/stable-diffusion")
iex> version = Replicate.Models.get_version!(model, "db21e45d3f7023abc2a46ee38a23973f6dce16bb082a930b0c49861f96d1e5bf")
iex> {:ok, prediction} = Replicate.Predictions.create(version, %{prompt: "Watercolor painting of the Babadook"})
iex> prediction.status
"starting"
iex> {:ok, prediction} = Replicate.Predictions.cancel(prediction.id)
iex> prediction.status
"canceled"
iex> model = Replicate.Models.get!("stability-ai/stable-diffusion")
iex> version = Replicate.Models.get_version!(model, "db21e45d3f7023abc2a46ee38a23973f6dce16bb082a930b0c49861f96d1e5bf")
iex> {:ok, prediction} = Replicate.Predictions.create(version, %{prompt: "a 19th century portrait of a wombat gentleman"})
iex> {:ok, prediction} = Replicate.Predictions.cancel(prediction)
iex> prediction.status
"canceled"
You can list all the predictions you've run:
iex> Replicate.Predictions.list()
[%Prediction{
id: "1234",
status: "starting",
input: %{"prompt" => "a 19th century portrait of a wombat gentleman"},
version: "27b93a2413e7f36cd83da926f3656280b2931564ff050bf9575f1fdf9bcd7478",
output: ["https://replicate.com/api/models/stability-ai/stable-diffusion/files/50fcac81-865d-499e-81ac-49de0cb79264/out-0.png"]
},
%Prediction{
id: "1235",
status: "starting",
input: %{"prompt" => "a 19th century portrait of a wombat gentleman"},
version: "27b93a2413e7f36cd83da926f3656280b2931564ff050bf9575f1fdf9bcd7478",
output: ["https://replicate.com/api/models/stability-ai/stable-diffusion/files/50fcac81-865d-499e-81ac-49de0cb79264/out-0.png"]}
]
You can list all the versions of a model:
iex> model = Replicate.Models.get!("stability-ai/stable-diffusion")
iex> versions = Replicate.Models.list_versions(model)
iex> Replicate.Models.list_versions(model) |> Enum.map(& &1.id) |> Enum.slice(0..5)
["db21e45d3f7023abc2a46ee38a23973f6dce16bb082a930b0c49861f96d1e5bf",
"328bd9692d29d6781034e3acab8cf3fcb122161e6f5afb896a4ca9fd57090577",
"f178fa7a1ae43a9a9af01b833b9d2ecf97b1bcb0acfd2dc5dd04895e042863f1",
"0827b64897df7b6e8c04625167bbb275b9db0f14ab09e2454b9824141963c966",
"27b93a2413e7f36cd83da926f3656280b2931564ff050bf9575f1fdf9bcd7478",
"8abccf52e7cba9f6e82317253f4a3549082e966db5584e92c808ece132037776"]
Get a paginated list of all public models.
iex> %{next: next, previous: previous, results: results} = Replicate.Models.list()
iex> next
"https://api.replicate.com/v1/trainings?cursor=cD0yMDIyLTAxLTIxKzIzJTNBMTglM0EyNC41MzAzNTclMkIwMCUzQTAw"
iex> previous
nil
iex> results |> length()
25
iex> results |> Enum.at(0)
%Replicate.Models.Model{
url: "https://replicate.com/replicate/hello-world",
owner: "replicate",
name: "hello-world",
description: "A tiny model that says hello",
visibility: "public",
github_url: "https://github.com/replicate/cog-examples",
paper_url: nil,
license_url: nil,
run_count: 12345,
cover_image_url: nil,
default_example: nil,
latest_version: %{
"cog_version" => "0.3.0",
"created_at" => "2022-03-21T13:01:04.418669Z",
"id" => "v2",
"openapi_schema" => %{}
}
}
Output files are returned as HTTPS URLs. Here's one way to load files without any dependencies:
iex> [url | _rest] = Replicate.run("stability-ai/stable-diffusion:db21e45d3f7023abc2a46ee38a23973f6dce16bb082a930b0c49861f96d1e5bf", prompt: "a watercolor of the babadook by Picasso")
iex> url
"https://replicate.delivery/pbxt/LgZ5BLmMWzqODp4rzSERXDNglStBTltHpj0533i9385qSgQE/out-0.png"
iex> {:ok, resp} = :httpc.request(:get, {url, []}, [], [body_format: :binary])
iex> {{_, 200, 'OK'}, _headers, body} = resp
iex> File.write!("babadook_watercolor.jpg", body)
You can create a model for a user or organization with a given name, visibility, and hardware SKU:
iex> {:ok, model} =
Replicate.Models.create(
owner: "your-username",
name: "my-model",
visibility: "public",
hardware: "gpu-a40-large"
)
Deployments allow you to control the configuration of a model with a private, fixed API endpoint. You can control the version of the model, the hardware it runs on, and how it scales.
Once you create a deployment on Replicate, you can make predictions like this:
iex> {:ok, deployment} = Replicate.Deployments.get("test/model")
iex> {:ok, prediction} = Replicate.Deployments.create_prediction(deployment, %{prompt: "a 19th century portrait of a wombat gentleman"})
Paginates through results provided by the endpoint_func
function.
Returns a stream of results.
iex> stream = Replicate.paginate(&Replicate.Models.list/0)
iex> first_batch = stream |> Enum.at(0)
iex> first_batch |> length()
25
iex> %Replicate.Models.Model{name: name} = first_batch |> Enum.at(0)
iex> name
"hello-world"