Hủy
Cloud Insights

All documents

Model Fine-Tuning

FPT Monitoring
Billing
Billing
AI Marketplace
AI Studio
FPT AI Inference
AI Inference
AI Infrastructure
FPT Security
FPT Cloud Server
FPT DevSecOps Services
FPT Integration
FPT Database Engine
FPT Cloud Backup & DR
FPT Storage
FPT Network
FPT Container
    Model Metrics
    Model Metrics
    Updated on 23 Sep 2025
    Print: Export: PDF

    Notice: Model metrics are enabled only when the execution pipeline is in the Running status and at the Training stage.

    Alt text

    Model metrics are collected to track the training performance of AI models during and after the fine-tuning process. These metrics help detect training anomalies, guide hyperparameter adjustments and improve model performance.

    Training metrics:

    Metric What it evaluates
    loss Measures how well the model is learning. A high loss means poor prediction; a low loss means the model is fitting the data well.
    learning_rate Controls how fast the model learns. A learning rate that’s too high can cause instability; too low can slow down training.
    grad_norms Indicates the magnitude of gradients. Helps detect issues like vanishing or exploding gradients, which affect learning.
    epoch Tracks how many full passes the model has made over the training data. Useful for monitoring learning progress over time.

    Evaluation metrics:

    Notice: Only shown when evaluation data is used.

    Metric What it evaluates
    eval_runtime Measures how long the evaluation process takes. Useful for performance benchmarking.
    eval_samples_per_second Indicates evaluation throughput. Higher is better for faster model validation.
    eval_steps_per_second Measures how many evaluation steps are completed per second. Reflects evaluation efficiency.
    eval_loss Measures how well the model generalizes to unseen data. Helps detect overfitting or underfitting.

    Training performance metrics:

    Metric What it evaluates
    train_runtime Total time spent training. Useful for estimating training cost and efficiency.
    train_samples_per_second Measures training throughput. Higher values indicate faster training.
    train_steps_per_second Indicates how many training steps are completed per second. Reflects training speed.
    total_flos Total floating point operations used. Helps estimate computational cost and model complexity.
    train_loss Measures how well the model fits the training data. Should decrease over time if training is effective.
    © 2025 FPT Cloud. All Rights Reserved.