Enofirm-Temp/extra-addons/AI/llm_training

LLM Training Module

Manage LLM fine-tuning jobs across different providers directly from Odoo.

Module Type: 🔌 Extension (Model Fine-tuning)

Architecture

┌───────────────────────────────────────────────────────────────┐
│                       User Interface                          │
│                 ┌─────────────────────────┐                   │
│                 │  LLM > Training > Jobs  │                   │
│                 └───────────┬─────────────┘                   │
└─────────────────────────────┼─────────────────────────────────┘
                              │
                              ▼
              ┌───────────────────────────────────────────┐
              │      ★ llm_training (This Module) ★       │
              │          Fine-tuning Management           │
              │  📊 Datasets │ Jobs │ Model Training      │
              └─────────────────────┬─────────────────────┘
                                    │
                        ┌───────────┴───────────┐
                        ▼                       ▼
    ┌───────────────────────────┐   ┌───────────────────────────┐
    │           llm             │   │   Provider APIs           │
    │    (Core Base Module)     │   │  (OpenAI Fine-tune API)   │
    └───────────────────────────┘   └───────────────────────────┘

Installation

What to Install

For model fine-tuning:

odoo-bin -d your_db -i llm_training

Auto-Installed Dependencies

• llm (core infrastructure)

Why Use Training?

Feature	llm_training
Management	📊 Manage jobs from Odoo
Datasets	📁 Dataset organization
Monitoring	📈 Track job progress
Multi-provider	🔄 Works with OpenAI, etc.

Common Setups

I want to...	Install
Fine-tune models	llm_training + llm_openai
Full LLM workflow	llm_assistant + llm_openai + llm_training

This module provides functionality for managing LLM fine-tuning jobs across different providers. It allows users to create, monitor, and manage fine-tuning jobs directly from the Odoo interface.

Features

• Create and manage fine-tuning jobs for LLMs
• Track job status and metrics
• Support for multiple LLM providers (OpenAI, etc.)
• Integration with dataset management
• Job status monitoring and notifications

Usage

Accessing the Module

1. Navigate to LLM > Training > Jobs in the main menu
2. This will display the list of all training jobs with their current status

Creating a New Training Job

1. Click the Create button on the training jobs list view
2. Fill in the required fields:
   • Name: A descriptive name for the job
   • Provider: Select the LLM provider (e.g., OpenAI)
   • Base Model: Select the base model to fine-tune
   • Datasets: Add one or more datasets to use for training
3. Optionally configure:
   • Description: Add details about the job
   • Hyperparameters: Adjust training parameters in JSON format
4. Click Save to create the job in draft state
5. Click Submit to start the training process

Monitoring Job Status

1. The job will progress through several states:

   • Draft: Initial state
   • Validating: Validating datasets
   • Preparing: Preparing data for training
   • Queued: Job is queued with the provider
   • Training: Training is in progress
   • Completed: Training completed successfully
   • Failed: Training failed
   • Cancelled: Training was cancelled

2. For jobs in progress, click the Check Status button to update the status

Viewing Results

1. Once a job is completed, go to the Results tab
2. Here you can find:
   • The resulting fine-tuned model
   • Training metrics
   • Training logs

Using Fine-tuned Models

After successful training, the fine-tuned model will be automatically added to your available models and can be used in:

• LLM Threads
• LLM Agents
• Any other component that uses LLM models

Security

The module follows the standard two-tier security model:

• Regular users (base.group_user) have read-only access
• LLM Managers (llm.group_llm_manager) have full CRUD access

Technical Information

Model Structure

• llm.training.job: Main model for training jobs
• llm.training.dataset: Model for managing training datasets

Provider Integration

The module uses a standardized interface for provider integration:

• start_training_job: Initiates the training process
• check_training_job_status: Checks the current status
• cancel_training_job: Cancels an ongoing job

Providers implement these methods to handle provider-specific logic.

Troubleshooting

Common Issues

1. Job stuck in validating state:

   • Check that datasets are properly formatted
   • Verify provider credentials

2. Failed jobs:

   • Check the Results tab for error details
   • Verify hyperparameters are correctly formatted

3. Missing models after completion:

   • Use the Check Status button to refresh the job status
   • Verify provider access to the fine-tuned model

Contributing

Please follow the standard Odoo development guidelines when contributing to this module.