Sign In Start Free →
📋 Changelog

What's New

Latest updates, features, and improvements to the GradientPond platform.

v2.4.1 December 10, 2024

Distributed Training at 10K+ GPU Scale

  • Support for distributed training across 10,000+ GPUs with automatic sharding
  • 40% faster checkpoint recovery with zero-copy memory-mapped saves
  • New elastic training mode — automatically scale workers up/down
  • Improved fault tolerance with automatic worker replacement
  • Fixed memory leak in long-running distributed jobs
v2.4.0 November 22, 2024

Population-Based Training & HPO Improvements

  • New: Population-Based Training (PBT) for hyperparameter optimization
  • Bayesian optimization now supports multi-objective search
  • Early stopping with configurable patience and delta thresholds
  • HPO dashboard with real-time parallel coordinate plots
  • 30% reduction in SDK initialization time
v2.3.2 October 30, 2024

Model Registry Enhancements

  • Model comparison view with side-by-side metric diffs
  • Automated model validation pipelines on promotion
  • Custom metadata tags and searchable model annotations
  • Webhook notifications for model stage transitions
  • Fixed: Model download timeout for artifacts > 50GB
v2.3.1 October 12, 2024

JAX & Flax Native Support

  • First-class JAX integration with automatic XLA compilation tracking
  • Flax training state serialization and checkpoint management
  • TPU pod training support via GCP integration
  • New: `gp.jax.log_params()` for Flax parameter logging
  • Performance: 50% less overhead for high-frequency metric logging
v2.3.0 September 25, 2024

Dataset Versioning v2

  • Complete rewrite of dataset versioning with Git-like semantics
  • Branching and merging for collaborative dataset curation
  • Diff viewer for dataset changes between versions
  • Lazy loading for datasets exceeding available memory
  • S3, GCS, and Azure Blob as backend storage options
v2.2.3 September 8, 2024

Collaboration & Notifications

  • Real-time experiment annotations and comments
  • Slack and Microsoft Teams notification integrations
  • @mention teammates on specific runs or metrics
  • Shared experiment collections with team visibility controls
  • Fixed: Notification delivery delays for webhook endpoints
v2.2.2 August 20, 2024

CLI Overhaul & Developer Experience

  • Completely redesigned CLI with interactive prompts and autocomplete
  • New: `gp compare` command for terminal-based run comparison
  • New: `gp deploy` for one-command model deployment
  • Shell completions for bash, zsh, and fish
  • Offline mode: queue metrics locally when disconnected
v2.2.1 August 5, 2024

Performance & Stability

  • Dashboard load time reduced by 60% with virtualized rendering
  • API response times improved by 40% across all endpoints
  • Fixed: Race condition in concurrent metric writes
  • Fixed: Memory usage spike when loading large experiment histories
  • Upgraded infrastructure to support 5x more concurrent connections