Sieve: ML tools for software teams.

ML tools for software teams.

Pre-built solutions available as an API & a full featured framework to scale ML apps.

Object Tracking

Detect objects in a video and track them across frames.

Accelerate your AI R&D

Explore pre-built apps and functions, or work with the Sieve team to tackle your exact problem.

Scale to production

Minimize infrastructure setup while maintaining full flexibility on how you run your workloads at scale.

Collaborate with your entire team

Enable PMs, data scientists, and software engineers to interact with ML apps collaboratively.

a fully featured ML runtimeDeploy your own apps

Define some code

Don't worry about GPU configuration, CUDA versions, or Docker. Take regular Python code and add a simple decorator to it.

@sieve.Model(
  name="my_model",
  gpu = True,
  python_packages=["torch==1.8.1"]
)
class Model:
  def __setup__(self):
      import torch
      self.model = torch.load("weights.pth")

  def __predict__(self, img: sieve.Image) -> sieve.Image:
      output = self.model(img.array)
      return sieve.Image(array=output)

@sieve.Model(
  name="my_model",
  gpu = True,
  python_packages=["torch==1.8.1"]
)
class Model:
  def __setup__(self):
      import torch
      self.model = torch.load("weights.pth")

  def __predict__(self, img: sieve.Image) -> sieve.Image:
      output = self.model(img.array)
      return sieve.Image(array=output)

Deploy instantly, call from anywhere

Deploy code to Sieve in a single command. Call it via API, Python SDK, or the dashboard.

> sieve deploy

> sieve deploy

API & SDK

curl "https://mango.sievedata.com/v1/push" \
  -X POST \
  -H "X-API-Key: $YOUR_KEY" \
  -d '{
    "workflow_name": "my_model",
    "inputs": {
      "img": {
        "url": "s3://sievedata-public/mango.jpg"
      }
    }
  }'

curl "https://mango.sievedata.com/v1/push" \
  -X POST \
  -H "X-API-Key: $YOUR_KEY" \
  -d '{
    "workflow_name": "my_model",
    "inputs": {
      "img": {
        "url": "s3://sievedata-public/mango.jpg"
      }
    }
  }'

{
  "id": "15a9c38e-4874-47e8-b6a2-0c59dcb6b1f9",
  "description": "Pushed 15a9c38e-4874-47e8-b6a2-0c59dcb6b1f9 to queue"
}

{
  "id": "15a9c38e-4874-47e8-b6a2-0c59dcb6b1f9",
  "description": "Pushed 15a9c38e-4874-47e8-b6a2-0c59dcb6b1f9 to queue"
}

Web UI

img

URL

an image to pass into my_model

Call pre-existing functions and models

Leverage the library of functions and models built by the Sieve community. Call them from your own code, the Sieve dashboard, or use them in your custom workflows.

Build complex workflows

Most apps require many functions and models chained together in non-trivial ways. Use a combination of existing building blocks and your own code to build the solution you want.

@sieve.workflow(name="talking_head_generator")
def talking_head(driving_video: sieve.Video, script: str, avatar: sieve.Image) -> sieve.Video:
    audio = sieve.reference('sieve/custom_tortoise_tts')(script)
    frames = sieve.reference('sieve/video-splitter')(driving_video)
    face_boxes = sieve.reference('sieve/face-detector')(frames)
    faces_tracked = sieve.reference('sieve/sort')(face_boxes)
    video = sieve.reference('sieve/wav2lip')(driving_video, audio, faces_tracked)
    talking_head = sieve.reference('sieve/thinplate-mm')(video, avatar)
    return talking_head

@sieve.workflow(name="talking_head_generator")
def talking_head(driving_video: sieve.Video, script: str, avatar: sieve.Image) -> sieve.Video:
    audio = sieve.reference('sieve/custom_tortoise_tts')(script)
    frames = sieve.reference('sieve/video-splitter')(driving_video)
    face_boxes = sieve.reference('sieve/face-detector')(frames)
    faces_tracked = sieve.reference('sieve/sort')(face_boxes)
    video = sieve.reference('sieve/wav2lip')(driving_video, audio, faces_tracked)
    talking_head = sieve.reference('sieve/thinplate-mm')(video, avatar)
    return talking_head

Built to scaleForget about infrastructure

Autoscaling

Sieve scales to meet your demand while giving you granular control over replica counts and other useful metrics.

Fast data transfer

Quick processing across clouds due to specialized rich data transfer via Sieve data types.

Stream and batch processing

Use Sieve's iterator support to stream results from jobs, or launch massively parallel jobs to process large datasets.

Full featured observability

Seamless viewing of logs, traffic, and latency by jobs, apps, models, or functions.

Built for teamsFor engineers and non-engineers alike

Organization Support

Add your team members to a shared workspace of Sieve apps.

Single source of truth

All your functions and models in one place: reusable, version controlled, and shared across applications.

Cross-functional collaboration

Trigger apps, visualize results, and share with your entire team using Sieve's web-based dashboard or CLI.

Built for your use case

Work with the experts

We're a team of machine learning experts from places like Berkeley, NVIDIA, Apple, Snapchat, Microsoft, and Scale AI. Work with us to support your custom use case and integrate Sieve into your existing pipeline.