Skip to content

api_reference.rst

Latest commit

 

History

History
651 lines (466 loc) · 14.7 KB

api_reference.rst

File metadata and controls

651 lines (466 loc) · 14.7 KB
Raw

API Reference

Complete API reference for the LLM-LoRA framework modules, classes, and functions. This reference provides detailed documentation for all public interfaces and utilities.

Core Training Modules

training_model.one_file_train

Main training orchestrator providing the complete training pipeline.

.. automodule:: training_model.one_file_train
   :members:
   :undoc-members:
   :show-inheritance:

Key Functions

.. autofunction:: training_model.one_file_train.train

.. autofunction:: training_model.one_file_train.main

.. autofunction:: training_model.one_file_train.setup_logging

.. autofunction:: training_model.one_file_train.load_model_and_tokenizer

training_model.dpo_train

Direct Preference Optimization training implementation.

.. automodule:: training_model.dpo_train
   :members:
   :undoc-members:
   :show-inheritance:

Key Classes

.. autoclass:: training_model.dpo_train.DPOTrainer
   :members:
   :undoc-members:


.. autoclass:: training_model.dpo_train.DPOConfig
   :members:
   :undoc-members:

Key Functions

.. autofunction:: training_model.dpo_train.dpo_train

.. autofunction:: training_model.dpo_train.compute_dpo_loss

.. autofunction:: training_model.dpo_train.prepare_preference_data

training_model.grpo_train

Group Relative Policy Optimization training implementation.

.. automodule:: training_model.grpo_train
   :members:
   :undoc-members:
   :show-inheritance:

Key Classes

.. autoclass:: training_model.grpo_train.GRPOTrainer
   :members:
   :undoc-members:


.. autoclass:: training_model.grpo_train.GRPOConfig
   :members:
   :undoc-members:

Key Functions

.. autofunction:: training_model.grpo_train.grpo_train

.. autofunction:: training_model.grpo_train.compute_grpo_loss

.. autofunction:: training_model.grpo_train.prepare_group_data

Utility Modules

training_model.auth_utils

Authentication and logging utility functions.

.. automodule:: training_model.auth_utils
   :members:
   :undoc-members:
   :show-inheritance:

Authentication Functions

.. autofunction:: training_model.auth_utils.tokens_init

training_model.data_preparation

Data preparation utilities for different dataset formats.

.. automodule:: training_model.data_preparation
   :members:
   :undoc-members:
   :show-inheritance:

Data Processing Functions

.. autofunction:: training_model.data_preparation.dataset_to_json

.. autofunction:: training_model.data_preparation.get_user_prompt

Configuration and Logging

training_model.logging_config

Logging configuration and setup utilities.

.. automodule:: training_model.logging_config
   :members:
   :undoc-members:
   :show-inheritance:

Key Functions

.. autofunction:: training_model.logging_config.setup_logging

.. autofunction:: training_model.logging_config.configure_transformers_logging

.. autofunction:: training_model.logging_config.setup_wandb_logging

training_model.optuna

Hyperparameter optimization using Optuna.

.. automodule:: training_model.optuna
   :members:
   :undoc-members:
   :show-inheritance:

Key Classes

.. autoclass:: training_model.optuna.OptunaOptimizer
   :members:
   :undoc-members:

Key Functions

.. autofunction:: training_model.optuna.optimize_hyperparameters

.. autofunction:: training_model.optuna.create_study

.. autofunction:: training_model.optuna.objective_function

Evaluation Modules

evaluation.model_evaluation

General model evaluation utilities and metrics.

.. automodule:: evaluation.model_evaluation
   :members:
   :undoc-members:
   :show-inheritance:

Key Classes

.. autoclass:: evaluation.model_evaluation.ModelEvaluator
   :members:
   :undoc-members:


.. autoclass:: evaluation.model_evaluation.EvaluationConfig
   :members:
   :undoc-members:

Key Functions

.. autofunction:: evaluation.model_evaluation.evaluate_model

.. autofunction:: evaluation.model_evaluation.compute_perplexity

.. autofunction:: evaluation.model_evaluation.compute_bleu_score

.. autofunction:: evaluation.model_evaluation.compute_rouge_score

evaluation.deepeval_integration

DeepEval framework integration for advanced evaluation.

.. automodule:: evaluation.deepeval_integration
   :members:
   :undoc-members:
   :show-inheritance:

Key Functions

.. autofunction:: evaluation.deepeval_integration.test_mention_number_of_values

.. autofunction:: evaluation.deepeval_integration.test_from_dataset

.. autofunction:: evaluation.deepeval_integration.set_local_model_via_cli

testing_model.models

Custom model implementations for testing and evaluation.

.. automodule:: testing_model.models
   :members:
   :undoc-members:
   :show-inheritance:

Key Classes

.. autoclass:: testing_model.models.CustomLocalModel
   :members:
   :undoc-members:


.. autoclass:: testing_model.models.CustomMistralModel
   :members:
   :undoc-members:

evaluation.game_evaluation

Specialized evaluation for game-based conversational AI.

.. automodule:: evaluation.game_evaluation
   :members:
   :undoc-members:
   :show-inheritance:

Key Classes

.. autoclass:: evaluation.game_evaluation.GameEvaluator
   :members:
   :undoc-members:


.. autoclass:: evaluation.game_evaluation.DialogueCoherenceMetric
   :members:
   :undoc-members:

Key Functions

.. autofunction:: evaluation.game_evaluation.evaluate_game_model

.. autofunction:: evaluation.game_evaluation.compute_coherence_score

.. autofunction:: evaluation.game_evaluation.evaluate_character_consistency

Fire CLI Interface

Main CLI Controller

.. automodule:: training_model
   :members:
   :undoc-members:
   :show-inheritance:

Key Classes

.. autoclass:: training_model.LLMLoRaCLI
   :members:
   :undoc-members:

CLI Methods

.. automethod:: main.LLMLoRAFramework.train_model

.. automethod:: main.LLMLoRAFramework.evaluate_model

.. automethod:: training_model.LLMLoRaCLI.convert

.. automethod:: training_model.LLMLoRaCLI.convert

.. automethod:: main.LLMLoRAFramework.list_models

.. automethod:: main.LLMLoRAFramework.clean_checkpoints

Configuration Classes

Model Configuration

.. autoclass:: training_model.config.ModelConfig
   :members:
   :undoc-members:

Training Configuration

.. autoclass:: training_model.config.TrainingConfig
   :members:
   :undoc-members:

Paths Configuration

.. autoclass:: training_model.config.PathsConfig
   :members:
   :undoc-members:

Conversion Configuration

.. autoclass:: training_model.config.ConversionConfig
   :members:
   :undoc-members:

Data Classes and Types

Training Data Types

.. autoclass:: training_model.types.InstructionData
   :members:
   :undoc-members:


.. autoclass:: training_model.types.PreferenceData
   :members:
   :undoc-members:


.. autoclass:: training_model.types.GroupPreferenceData
   :members:
   :undoc-members:

Evaluation Data Types

.. autoclass:: evaluation.types.EvaluationResult
   :members:
   :undoc-members:


.. autoclass:: evaluation.types.MetricResult
   :members:
   :undoc-members:


.. autoclass:: evaluation.types.TestCase
   :members:
   :undoc-members:

Model Types

.. autoclass:: training_model.types.ModelOutput
   :members:
   :undoc-members:


.. autoclass:: training_model.types.TrainingMetrics
   :members:
   :undoc-members:

Constants and Enums

Training Methods Enum

.. autoclass:: training_model.enums.TrainingMethod
   :members:
   :undoc-members:


.. autoclass:: training_model.enums.QuantizationType
   :members:
   :undoc-members:


.. autoclass:: training_model.enums.ModelFormat
   :members:
   :undoc-members:

Evaluation Enums

.. autoclass:: evaluation.enums.EvaluationMetric
   :members:
   :undoc-members:


.. autoclass:: evaluation.enums.EvaluationType
   :members:
   :undoc-members:

Exception Classes

Training Exceptions

.. autoexception:: training_model.exceptions.TrainingError

.. autoexception:: training_model.exceptions.ModelLoadError

.. autoexception:: training_model.exceptions.DataPreparationError

.. autoexception:: training_model.exceptions.ConversionError

Evaluation Exceptions

.. autoexception:: evaluation.exceptions.EvaluationError

.. autoexception:: evaluation.exceptions.MetricComputationError

.. autoexception:: evaluation.exceptions.TestCaseError

Configuration Exceptions

.. autoexception:: training_model.exceptions.ConfigurationError

.. autoexception:: training_model.exceptions.InvalidParameterError

Helper Functions

Version Information

Framework Version

.. autodata:: training_model.__version__

   Current version of the LLM-LoRA framework.


.. autodata:: training_model.__author__

   Framework author information.


.. autodata:: training_model.__description__

   Framework description.

Dependency Versions

.. autofunction:: training_model.version.get_dependency_versions

   Get versions of key dependencies.


.. autofunction:: training_model.version.check_compatibility

   Check compatibility with current environment.

Usage Examples

Basic Training Example

from training_model.one_file_train import train
from omegaconf import OmegaConf

# Load configuration
config = OmegaConf.load("conf/config.yaml")

# Run training
results = train(config)

print(f"Training completed. Model saved to: {results.model_path}")

DPO Training Example

from training_model.dpo_train import dpo_train
from training_model.data_preparation import load_preference_dataset

# Load preference data
dataset = load_preference_dataset("data/preferences.json")

# Configure DPO training
config = {
    "model_name": "models/sft_base",
    "beta": 0.1,
    "learning_rate": 1e-5
}

# Run DPO training
model = dpo_train(config, dataset)

Model Evaluation Example

from evaluation.model_evaluation import ModelEvaluator

# Initialize evaluator
evaluator = ModelEvaluator(
    model_path="models/trained_model",
    metrics=["bleu", "rogue", "perplexity"]
)

# Run evaluation
results = evaluator.evaluate(test_dataset="data/test.json")

# Print results
for metric, score in results.items():
    print(f"{metric}: {score}")

Model Conversion Example

from training_model.one_file_train import convert_to_gguf, convert_to_rkllm

# Convert to GGUF
# Convert to GGUF using CLI
python main.py convert --gguf=True

# Convert to RKLLM using CLI
python main.py convert --rkllm=True --target_platform=rk3588

Fire CLI Usage

import fire
from main import LLMLoRAFramework

# Use Fire CLI programmatically
framework = LLMLoRAFramework()

# Train model
framework.train_model(config_name="custom_config")

# Evaluate model
results = framework.evaluate_model(model_path="models/latest")

# Convert model
# Convert model using CLI
python main.py convert --gguf=True --rkllm=True

Advanced Configuration

from omegaconf import OmegaConf
from training_model.config import ModelConfig, TrainingConfig

# Create advanced configuration
config = OmegaConf.create({
    "model": {
        "model_name": "Vikhrmodels/Vikhr-YandexGPT-5-Lite-8B-it",
        "lora_r": 64,
        "lora_alpha": 128,
        "quantization": True
    },
    "training": {
        "training_method": "grpo",
        "batch_size": 2,
        "learning_rate": 5e-6,
        "epochs": 1
    },
    "conversion": {
        "convert": {"gguf": True, "rkllm": True},
        "convert_to_rkllm": True,
        "quantization_type": "q4_1"
    }
})

# Run training with advanced config
results = train(config)

Custom Metrics Example

from evaluation.model_evaluation import ModelEvaluator
from evaluation.types import MetricResult

class CustomMetric:
    def compute(self, predictions, references):
        # Implement custom metric logic
        score = custom_metric_computation(predictions, references)
        return MetricResult(name="custom_metric", score=score)

# Use custom metric
evaluator = ModelEvaluator(model_path="models/trained_model")
evaluator.add_metric(CustomMetric())

results = evaluator.evaluate(test_dataset="data/test.json")

Troubleshooting Reference

Common Error Codes

Error Codes
Code Error Type Description
E001 Model Loading Error Failed to load base model or checkpoint
E002 Data Format Error Invalid dataset format or structure
E003 Configuration Error Invalid or missing configuration parameters
E004 Memory Error Insufficient GPU or system memory
E005 Conversion Error Model format conversion failure
E006 Evaluation Error Error during model evaluation
E007 Dependency Error Missing or incompatible dependencies

Performance Optimization

.. autofunction:: training_model.optimization.optimize_training_performance

.. autofunction:: training_model.optimization.optimize_inference_performance

.. autofunction:: training_model.optimization.auto_tune_hyperparameters

Debugging Utilities

.. autofunction:: training_model.debug.enable_debug_mode

.. autofunction:: training_model.debug.print_model_info

.. autofunction:: training_model.debug.trace_memory_usage

Migration Guide

Version Migration

.. autofunction:: training_model.migration.migrate_config_v1_to_v2

.. autofunction:: training_model.migration.update_model_format

.. autofunction:: training_model.migration.check_compatibility