Provides a Python SDK that integrates with existing machine learning workflows to enable real-time debugging, diagnosis, and improvement of training data without requiring data migration. It helps identify inefficient samples, track dataset changes, and optimize model performance by linking per-sample metrics to specific dataset revisions and hyperparameter combinations.
Funding
Funding not disclosed
Founders
Product
Problem
Debugging, diagnosing, and improving machine learning training data is challenging, often requiring data migration and complex workflows. Identifying inefficient samples, tracking dataset changes, and optimizing model performance can be difficult and time-consuming.
Solution
3LC provides a Python SDK that integrates with existing machine learning workflows, enabling real-time debugging, diagnosis, and improvement of training data without requiring data migration. The SDK helps identify important or inefficient samples, understand model struggles, and improve model performance by weighting data. It also facilitates dataset versioning, allowing users to make sparse, non-destructive edits, maintain a lineage of changes, and restore previous revisions. Furthermore, 3LC offers experiment tracking with per-sample, per-epoch metrics, tying each training run to a specific dataset revision for full reproducibility and enabling the identification of optimal hyperparameter and dataset modification combinations.
Target Audience
The primary users are machine learning engineers and data scientists who need to efficiently debug, diagnose, and improve their training data.
Features
- Python SDK for integration with existing ML workflows
- Real-time debugging and diagnosis of training data
- Identification of important or inefficient samples
- Dataset versioning with sparse, non-destructive edits
- Experiment tracking with per-sample, per-epoch metrics
- Visualization tools for data analysis
- Seamless data editing capabilities
- Integration with existing ML tools
- No data migration required