aligntune 0.2.0__py3-none-any.whl

aligntune/__init__.py

@@ -0,0 +1,368 @@
+"""
+AlignTune: A comprehensive fine-tuning library supporting both SFT and RL training methods.
+"""
+
+import logging
+import os
+import sys
+
+# Import colored logging utilities
+try:
+    from .utils.colored_logging import (
+        init_aligntune_logging,
+        print_aligntune_banner,
+        print_section_banner,
+        print_subsection,
+        aligntune_info,
+        aligntune_warning,
+        aligntune_error,
+        aligntune_success,
+        aligntune_step,
+        setup_colored_logging,
+    )
+    COLORED_LOGGING_AVAILABLE = True
+except ImportError:
+    COLORED_LOGGING_AVAILABLE = False
+    # Fallback to basic logging
+    if not logging.getLogger().handlers:
+        logging.basicConfig(level=logging.INFO)
+    logger = logging.getLogger("aligntune")
+    logger.setLevel(logging.WARNING)
+
+# Initialize colored logging if available
+if COLORED_LOGGING_AVAILABLE:
+    logger = setup_colored_logging("aligntune", logging.WARNING)
+else:
+    # Configure basic logging as fallback
+    if not logging.getLogger().handlers:
+        logging.basicConfig(level=logging.INFO)
+    logger = logging.getLogger("aligntune")
+    logger.setLevel(logging.WARNING)
+
+# -----------------------------------------------------------------------------
+# VERSION & METADATA
+# -----------------------------------------------------------------------------
+try:
+    from importlib.metadata import version as _pkg_version, PackageNotFoundError as _PkgNotFound
+    __version__ = _pkg_version("aligntune")
+except Exception:  # Fallback during editable installs before metadata exists
+    __version__ = "0.0.0"
+__author__ = "Your Name"
+__email__ = "your.email@example.com"
+
+# -----------------------------------------------------------------------------
+# CORE IMPORTS
+# -----------------------------------------------------------------------------
+# Import all components from the centralized import management system
+from ._imports import (
+    # Core availability flags
+    UNSLOTH_AVAILABLE,
+    UNSLOTH_ERROR_INFO,
+    TRL_AVAILABLE,
+    UNIFIED_RL_AVAILABLE,
+    UNIFIED_SFT_AVAILABLE,
+    EVAL_AVAILABLE,
+    REWARDS_AVAILABLE,
+    BACKEND_FACTORY_AVAILABLE,
+    UNIFIED_AVAILABLE,
+    CLI_AVAILABLE,
+    
+    # Unified system components
+    UnifiedConfig,
+    AlgorithmType,
+    PrecisionType,
+    BackendType,
+    UnifiedModelConfig,
+    UnifiedDatasetConfig,
+    RewardConfig,
+    UnifiedTrainingConfig,
+    DistributedConfig,
+    UnifiedLoggingConfig,
+    ConfigLoader,
+    TrainerBase,
+    TrainingState,
+    TrainerFactory,
+    DatasetRegistry,
+    RewardRegistry,
+    TaskRegistry,
+    UnifiedLogger,
+    UnifiedEvaluator,
+    PolicyModel,
+    ReferenceModel,
+    ValueModel,
+    RolloutEngine,
+    # Optimization components
+    OptimizerRegistry,
+    SchedulerRegistry,
+    OptimizerType,
+    SchedulerType,
+    get_optimizer_for_config,
+    get_scheduler_for_config,
+    validate_optimizer_availability,
+    validate_scheduler_availability,
+    # Recipe system
+    RecipeRegistry,
+    RecipeTypeEnum,
+    ModelFamily,
+    RecipeMetadata,
+    Recipe,
+    load_recipe_from_yaml,
+    load_builtin_recipes,
+    # Validation and diagnostics
+    ConfigValidator,
+    validate_config,
+    TrainingDiagnostics,
+    TrainingMonitor,
+    DiagnosticsCollector,
+    generate_training_report,
+    run_config_validation,
+    run_comprehensive_diagnostics,
+    # Error handling and UX
+    AlignTuneError,
+    ConfigurationError,
+    TrainingError,
+    EnvironmentError,
+    ValidationError,
+    handle_error,
+    create_progress_display,
+    HealthMonitor,
+    config_error,
+    training_error,
+    env_error,
+    validation_error,
+    PPOTrainer,
+    DPOTrainer,
+    GRPOTrainer,
+    GSPOTrainer,
+    
+    # SFT system components
+    SFTConfig,
+    SFTTaskType,
+    SFTModelConfig,
+    SFTDatasetConfig,
+    SFTTrainingConfig,
+    SFTLoggingConfig,
+    SFTConfigLoader,
+    SFTTrainerFactory,
+    InstructionTrainer,
+    ClassificationTrainer,
+    ChatTrainer,
+    
+    # Evaluation system
+    EvalType,
+    TaskCategory,
+    EvalConfig,
+    EvalTask,
+    EvalResult,
+    EvalRunner,
+    LMEvalConfig,
+    LMEvalRunner,
+    get_available_lm_eval_tasks,
+    run_standard_benchmark,
+    
+    # Rewards system
+    RewardType,
+    RewardConfig as RewardsRewardConfig,
+    RewardFunction,
+    RewardFunctionFactory,
+    CompositeReward,
+    rewards_registry,
+    
+    # Backend factory
+    BackendFactory,
+    TrainingType,
+    BackendFactoryType as BackendType,
+    FactoryBackendType,
+    RLAlgorithm,
+    BackendConfig,
+    create_sft_trainer,
+    create_rl_trainer,
+    list_backends,
+    
+    # CLI components
+    cli_main,
+    
+    # Helper functions
+    get_available_trainers,
+    print_available_trainers,
+    check_dependencies,
+    get_missing_dependencies,
+)
+
+# Import fallback functions from dedicated module
+from ._fallbacks import (
+    train_dpo_from_yaml,
+    train_ppo_from_yaml,
+    train_grpo_from_yaml,
+    train_grpo_from_config,
+    create_sample_dpo_config,
+    create_minimal_dpo_config,
+    create_sample_ppo_config,
+    create_minimal_ppo_config,
+    create_specialized_ppo_configs,
+    load_ppo_config_from_yaml,
+    show_ppo_configuration_menu,
+    select_ppo_dataset_size,
+    create_grpo_configurations,
+    select_grpo_config_interactively,
+    evaluate_grpo_model,
+)
+
+# =============================================================================
+# EXPORTS
+# =============================================================================
+
+# Core exports that are always available
+__all__ = [
+    # Version info
+    "__version__",
+    "__author__",
+    "__email__",
+    
+    # Core availability flags
+    "UNSLOTH_AVAILABLE",
+    "TRL_AVAILABLE",
+    "UNIFIED_RL_AVAILABLE",
+    "UNIFIED_SFT_AVAILABLE",
+    "EVAL_AVAILABLE",
+    "REWARDS_AVAILABLE",
+    "BACKEND_FACTORY_AVAILABLE",
+    "UNIFIED_AVAILABLE",
+    "CLI_AVAILABLE",
+    
+    # Helper functions
+    "get_available_trainers",
+    "print_available_trainers",
+    "check_dependencies",
+    "get_missing_dependencies",
+]
+
+# Add unified system components if available
+if UNIFIED_AVAILABLE:
+    __all__.extend([
+        "UnifiedConfig",
+        "AlgorithmType",
+        "PrecisionType",
+        "BackendType",
+        "UnifiedModelConfig",
+        "UnifiedDatasetConfig",
+        "RewardConfig",
+        "UnifiedTrainingConfig",
+        "DistributedConfig",
+        "UnifiedLoggingConfig",
+        "ConfigLoader",
+        "TrainerBase",
+        "TrainingState",
+        "TrainerFactory",
+        "DatasetRegistry",
+        "RewardRegistry",
+        "TaskRegistry",
+        "UnifiedLogger",
+        "UnifiedEvaluator",
+        "PolicyModel",
+        "ReferenceModel",
+        "ValueModel",
+        "RolloutEngine",
+        # Optimization components
+        "OptimizerRegistry",
+        "SchedulerRegistry",
+        "OptimizerType",
+        "SchedulerType",
+        "get_optimizer_for_config",
+        "get_scheduler_for_config",
+        "validate_optimizer_availability",
+        "validate_scheduler_availability",
+        # Recipe system
+        "RecipeRegistry",
+        "RecipeTypeEnum",
+        "ModelFamily",
+        "RecipeMetadata",
+        "Recipe",
+        "load_recipe_from_yaml",
+        "load_builtin_recipes",
+        # Validation and diagnostics
+        "ConfigValidator",
+        "validate_config",
+        "TrainingDiagnostics",
+        "TrainingMonitor",
+        "DiagnosticsCollector",
+        "generate_training_report",
+        "run_config_validation",
+        "run_comprehensive_diagnostics",
+        # Error handling and UX
+        "AlignTuneError",
+        "ConfigurationError",
+        "TrainingError",
+        "EnvironmentError",
+        "ValidationError",
+        "handle_error",
+        "create_progress_display",
+        "HealthMonitor",
+        "config_error",
+        "training_error",
+        "env_error",
+        "validation_error",
+        "PPOTrainer",
+        "DPOTrainer",
+        "GRPOTrainer",
+        "GSPOTrainer",
+    ])
+
+# Add SFT system components if available
+if UNIFIED_SFT_AVAILABLE:
+    __all__.extend([
+        "SFTConfig",
+        "SFTTaskType",
+        "SFTModelConfig",
+        "SFTDatasetConfig",
+        "SFTTrainingConfig",
+        "SFTLoggingConfig",
+        "SFTConfigLoader",
+        "SFTTrainerFactory",
+        "InstructionTrainer",
+        "ClassificationTrainer",
+        "ChatTrainer",
+    ])
+
+# Add evaluation components if available
+if EVAL_AVAILABLE:
+    __all__.extend([
+        "EvalType",
+        "TaskCategory",
+        "EvalConfig",
+        "EvalTask",
+        "EvalResult",
+        "EvalRunner",
+        "LMEvalConfig",
+        "LMEvalRunner",
+        "get_available_lm_eval_tasks",
+        "run_standard_benchmark",
+    ])
+
+# Add rewards components if available
+if REWARDS_AVAILABLE:
+    __all__.extend([
+        "RewardType",
+        "RewardsRewardConfig",
+        "RewardFunction",
+        "RewardFunctionFactory",
+        "CompositeReward",
+        "rewards_registry",
+    ])
+
+    # Add backend factory components if available
+if BACKEND_FACTORY_AVAILABLE:
+    __all__.extend([
+        "BackendFactory",
+        "TrainingType",
+        "BackendType",
+        "FactoryBackendType",
+        "RLAlgorithm",
+        "BackendConfig",
+        "create_sft_trainer",
+        "create_rl_trainer",
+        "list_backends",
+        
+        # CLI components
+        "cli_main",
+    ])

aligntune/__main__.py

@@ -0,0 +1,11 @@
+"""
+AlignTune CLI Entry Point
+
+This module provides the main entry point for the AlignTune CLI.
+
+"""
+
+from .cli import main
+
+if __name__ == "__main__":
+    main()

aligntune/_fallbacks.py

@@ -0,0 +1,132 @@
+"""
+<<<<<<< HEAD
+=======
+Fallback handler module for AlignTune.
+
+This module provides fallback functions when dependencies are missing.
+All functions provide clear error messages with installation instructions.
+"""
+
+def train_dpo_from_yaml(*args, **kwargs):
+    """Fallback for DPO training when TRL is not available."""
+    raise ImportError(
+        "DPO training requires TRL. Install with: pip install trl\n"
+        "For more information, visit: https://github.com/huggingface/trl"
+    )
+
+def train_ppo_from_yaml(*args, **kwargs):
+    """Fallback for PPO training when TRL is not available."""
+    raise ImportError(
+        "PPO training requires TRL. Install with: pip install trl\n"
+        "For more information, visit: https://github.com/huggingface/trl"
+    )
+
+def train_grpo_from_yaml(*args, **kwargs):
+    """Fallback for GRPO training when TRL is not available."""
+    raise ImportError(
+        "GRPO training requires TRL. Install with: pip install trl\n"
+        "For more information, visit: https://github.com/huggingface/trl"
+    )
+
+def train_grpo_from_config(*args, **kwargs):
+    """Fallback for GRPO training from config when TRL is not available."""
+    raise ImportError(
+        "GRPO training requires TRL. Install with: pip install trl\n"
+        "For more information, visit: https://github.com/huggingface/trl"
+    )
+
+def create_sample_dpo_config(*args, **kwargs):
+    """Fallback for DPO config creation when TRL is not available."""
+    raise ImportError(
+        "DPO config creation requires TRL. Install with: pip install trl\n"
+        "For more information, visit: https://github.com/huggingface/trl"
+    )
+
+def create_minimal_dpo_config(*args, **kwargs):
+    """Fallback for minimal DPO config when TRL is not available."""
+    raise ImportError(
+        "DPO config creation requires TRL. Install with: pip install trl\n"
+        "For more information, visit: https://github.com/huggingface/trl"
+    )
+
+def create_sample_ppo_config(*args, **kwargs):
+    """Fallback for PPO config creation when TRL is not available."""
+    raise ImportError(
+        "PPO config creation requires TRL. Install with: pip install trl\n"
+        "For more information, visit: https://github.com/huggingface/trl"
+    )
+
+def create_minimal_ppo_config(*args, **kwargs):
+    """Fallback for minimal PPO config when TRL is not available."""
+    raise ImportError(
+        "PPO config creation requires TRL. Install with: pip install trl\n"
+        "For more information, visit: https://github.com/huggingface/trl"
+    )
+
+def create_specialized_ppo_configs(*args, **kwargs):
+    """Fallback for specialized PPO configs when TRL is not available."""
+    raise ImportError(
+        "PPO config creation requires TRL. Install with: pip install trl\n"
+        "For more information, visit: https://github.com/huggingface/trl"
+    )
+
+def load_ppo_config_from_yaml(*args, **kwargs):
+    """Fallback for PPO config loading when TRL is not available."""
+    raise ImportError(
+        "PPO config loading requires TRL. Install with: pip install trl\n"
+        "For more information, visit: https://github.com/huggingface/trl"
+    )
+
+def show_ppo_configuration_menu(*args, **kwargs):
+    """Fallback for PPO config menu when TRL is not available."""
+    raise ImportError(
+        "PPO config menu requires TRL. Install with: pip install trl\n"
+        "For more information, visit: https://github.com/huggingface/trl"
+    )
+
+def select_ppo_dataset_size(*args, **kwargs):
+    """Fallback for PPO dataset selection when TRL is not available."""
+    raise ImportError(
+        "PPO dataset selection requires TRL. Install with: pip install trl\n"
+        "For more information, visit: https://github.com/huggingface/trl"
+    )
+
+def create_grpo_configurations(*args, **kwargs):
+    """Fallback for GRPO config creation when TRL is not available."""
+    raise ImportError(
+        "GRPO config creation requires TRL. Install with: pip install trl\n"
+        "For more information, visit: https://github.com/huggingface/trl"
+    )
+
+def select_grpo_config_interactively(*args, **kwargs):
+    """Fallback for GRPO config selection when TRL is not available."""
+    raise ImportError(
+        "GRPO config selection requires TRL. Install with: pip install trl\n"
+        "For more information, visit: https://github.com/huggingface/trl"
+    )
+
+def evaluate_grpo_model(*args, **kwargs):
+    """Fallback for GRPO evaluation when TRL is not available."""
+    raise ImportError(
+        "GRPO evaluation requires TRL. Install with: pip install trl\n"
+        "For more information, visit: https://github.com/huggingface/trl"
+    )
+
+# Fallback classes for when trainers are not available
+class FallbackSFTTrainer:
+    """Fallback SFT trainer when trainer is not available."""
+    def __init__(self, *args, **kwargs):
+        raise ImportError(
+            "SFT trainer not available. Check your AlignTune installation.\n"
+            "Install with: pip install -e .\n"
+            "For more information, visit: https://github.com/yourusername/aligntune"
+        )
+
+class FallbackClassificationTrainer:
+    """Fallback classification trainer when trainer is not available."""
+    def __init__(self, *args, **kwargs):
+        raise ImportError(
+            "Classification trainer not available. Check your AlignTune installation.\n"
+            "Install with: pip install -e .\n"
+            "For more information, visit: https://github.com/yourusername/aligntune"
+        )