AccelHydra

Contents:

  • Introduction
    • What is AccelHydra?
    • It IS
    • It IS NOT
    • Why AccelHydra?
      • Key Benefits
      • When to Use AccelHydra
      • When Not to Use AccelHydra
    • Installation
    • Quick Start
    • Citation
  • Getting Started
    • Project Structure
    • Step 1: Define Your Model
    • Step 2: Define Your Datasets
    • Step 3: Define Your Trainer
    • Step 4: Write Configuration YAML(s)
    • Step 5: Launch Training (Optional: Customize with TrainLauncher)
      • Option 1: Use the Default TrainLauncher (Recommended for Simple Cases)
      • Option 2: Create a Custom TrainLauncher (For Custom Requirements)
      • Command Line Overrides
    • Next Steps
  • Trainer
    • Overview
    • Training Loop Flow
    • Core Methods
      • Abstract Methods (Must Implement)
      • Hooks (Optional Override)
      • Example: Complete Trainer Implementation
    • Full Trainer API
  • Configuration Guide
    • Basic Concepts
      • _target_ Pattern
      • Parameter Passing with _convert_
    • Configuration Structure
      • Single File Configuration
      • Modular Configuration with defaults
      • Variable Interpolation
    • Command Line Overrides
      • Basic Overrides
      • Multiple Overrides
      • Adding New Fields
      • Removing Fields
      • Changing Configuration Groups
    • OmegaConf Resolvers
      • len() Resolver
      • multiply() Resolver
    • Advanced Patterns
      • Conditional Configuration
      • Config Composition
    • Limitations
    • Examples
  • API Reference
    • Models
      • LoadPretrainedMixin
        • LoadPretrainedMixin.process_state_dict()
        • LoadPretrainedMixin.load_pretrained()
      • CountParamsMixin
        • CountParamsMixin.count_params()
      • SaveTrainableParamsMixin
        • SaveTrainableParamsMixin.param_names_to_save
        • SaveTrainableParamsMixin.load_state_dict()
    • Trainer Components
    • Training Launcher
      • TrainLauncher
        • TrainLauncher.get_register_resolver_fn()
        • TrainLauncher.get_steps_for_lr_scheduler()
        • TrainLauncher.get_dataloaders()
        • TrainLauncher.run()
    • Utilities
      • Data Utilities
        • init_dataloader_from_config()
      • Configuration Utilities
        • multiply()
        • register_omegaconf_resolvers()
        • load_config_with_overrides()
        • load_config_from_cli()
        • parse_launch_args()
      • General Utilities
        • is_package_available()
        • read_jsonl_to_mapping()
        • setup_resume_cfg()
      • Learning Rate Scheduler Utilities
        • get_warmup_steps()
        • get_dataloader_one_pass_outside_steps()
        • get_total_training_steps()
        • get_dataloader_one_pass_steps_inside_accelerator()
        • get_steps_inside_accelerator_from_outside_steps()
        • lr_scheduler_param_adapter()
      • PyTorch Utilities
        • remove_key_prefix_factory()
        • merge_matched_keys()
        • load_pretrained_model()
        • create_mask_from_length()
        • loss_with_mask()
        • trim_or_pad_length()
        • concat_non_padding()
        • restore_from_concat()
        • contains_nan()
      • Accelerate Extensions
        • AcceleratorSaveTrainableParams
AccelHydra
  • Search

© Copyright 2025, Xuenan Xu, Yixuan Li, Jiahao Mei.

Built with Sphinx using a theme provided by Read the Docs.