
Test
Create rigorous data & model tests in just a few lines of code.

Visualize
Explore your models and actually “see” what causes a test to fail.

Debug
Detect and fix bugs in your code effortlessly in seconds.
Explore
Interactive Debugger
Explore the entire computational graph of your model including all data, parameters and gradients.

import efemarai as ef for data, target in dataset: optimizer.zero_grad() # Pause execution & visualize computational graph with ef.scan(): output = model(data) loss = F.cross_entropy(output, target) loss.backward() optimizer.step()
Runtime Assertions
Use the provided assertions or write your own tensor and function assertions to catch bugs effortlessly.

import efemarai as ef # Enable all runtime assertions ef.full_check() for data, target in data_loader: optimizer.zero_grad() # Detect & visualize failed assertions with ef.scan(): output = model(data) loss = F.cross_entropy(output, target) loss.backward() optimizer.step()
Tensor Inspection
Visualize up to 7D tensors and inspect the value of any element in just a couple of clicks.

import efemarai as ef # Inspect the raw values of any tensor ef.inspect(tensor) # Inspect tensors containing various types of data, e.g. ef.inspect(tensor, view=ef.View.Image) ef.inspect(tensor, view=ef.View.PointCloud)
Testing Framework
Create extensive test suits both for your models and data with just a few lines of code.

import efemarai as ef # Create a data report with various health checks ef.summarize(data) # Test the quality of your data, e.g. ef.expect(data).blurriness.toBeLessThan(0.1) ef.expect(data).text.illegal_characters.toHaveLength(0) # Test both your model and data, catch regressions, e.g. ef.expect(model).accuracy.on(data).toBeGreaterThan(0.85) # Easily test non-trivial model properties, e.g. ef.expect(model).robustness.toBeGreaterThan(0.95)
Efemarai in action
Why Efemarai?
Reduce time for developing & debugging ML models by up to
50%
Boost the reliability and performance of ML models

Follow well established quality assurance principles

Our users so far have run
Debug Sessions
1.5k+
Visualizations
22k+
Assertions
107k+