Skip to main content
Here’s a step-by-step walkthrough to get started with Zebrid for machine learning development. Perfect for engineers looking to test models, debug code, and scale jobs on GPU-backed infrastructure.

​
1. Register & Access Workspace

  • 1.1. Sign up with email address at: https://www.zebrid.com/login
  • 1.2. Enter your workspace by clicking Start Working on the homepage or via the Workspace in navigation bar.

​
2. Upload Assets

  • 2.1. (Pre-trained) Models: Go to Workspace > Model > New Model. Start with smaller models (few GBs) due to current upload limitations.
  • 2.2. Datasets: Add data via Workspace > Dataset > New Dataset.
  • 2.3. Codebase: Upload scripts to Workspace > Codebase > New Codebase. Note: Access data/models via ZebridLib APIs (see code examples below).

​
3. Create & Configure a Project

  • 3.1. New Project: Navigate to Workspace > Project > Create Project.
  • 3.2. Link your codebase, model, and dataset with the project. In the project’s Linked Resources, you’ll find relative paths of your assets on Zebrid storage.

​
4. Debug Code Interactively

  • 4.1. From your Project Card, Click to Start Live Processing.
— 4.1.1. Allow browser pop-ups to launch a JupyterLab VM with pre-installed ML packages. — 4.1.2. Your code syncs to a cloud directory (e.g., codebase/x/y/). Use commands like below to debug your code 
%cd codebase/x/y/  # Navigate to your code directory 
%run main.py --input XXX --output YYY  # Run your script
  • 4.2. Sync fixes to Codebase: Use ZebridLib to save changes:
from zebridlib import storage 
storage.write_file("./main.py", "codebase/x/y/main.py")
(Alternatively, download/reattach debugged code—we’re improving this workflow!) #5. Submit GPU Jobs
  • 5.1. In your project, go to Job > Submit Job.
  • 5.1.1. Select CPU/GPU specs, runtime image, parameters, etc.
  • 5.1.2. Upload requirements.txt to your codebase for dependencies.
  • 5.1.3. Note: No live logs yet — reach out (support@zebrid.com) with your task name if issues arise, we will assist you right away.

​
Key Code Snippets

​
Accessing Assets via ZebridLib

from zebridlib import storage 
import torch

​
Load model from cloud storage

def load_pretrained_model(model, model_path, device="cpu"): 
 local_dir = "./model/" 
 if storage.list_by_folder(model_path): 
 for f in storage.list_by_folder(model_path): 
 byte_content = storage.read_file(f) 
 with open(os.path.join(local_dir, os.path.basename(f)), "wb") as f_out: 
 f_out.write(byte_content) 
 model.load_state_dict(torch.load(local_dir+"best_model.pth", map_location=device)) 
 return model

​
Save trained model to cloud

def save_trained_model(model, save_path): 
 torch.save(model.state_dict(), "./model/best_model.pth") 
 storage.write_file("./model/best_model.pth", save_path+"best_model.pth")

​
6. Reference & Feedback

  • 6.1. Documentation: Access the full technical guide here: Zebrid Technical Documentation
  • 6.2. We’d love your feedback! We’re continuously improving, and your input makes a difference. Known limitations under improvement include large-file uploads and live logging. If you give it a try, let us know what works — and what doesn’t.
Know someone who might benefit from this? Tag a data scientist or ML engineer to help us spread the word. Have suggestions or want to share your experience? We’re all ears at support@zebrid.com. 👇#MachineLearning #AI #CloudComputing #TechTools Let us know if you could try it — or have any suggestions for improvements! 😊