< img height="1" width="1" style="display:none;" alt="" src="https://px.ads.linkedin.com/collect/?pid=3131724&fmt=gif" />
Last updated:

    Model Fine-tuning Overview

    What is Model Fine-tuning

    The platform provides GPU-accelerated fine-tuning instance hosting services, supporting mainstream fine-tuning frameworks (LLaMA-Factory, MS-Swift). Users only need to select compute resources and a dataset to quickly perform customized training on large models — no complex training code required.

    Supported Fine-tuning Frameworks

    Framework Features
    LLaMA-Factory Provides a visual web training UI (LlamaBoard); supports multiple fine-tuning methods (LoRA, QLoRA, full-parameter fine-tuning, etc.); easy to get started
    MS-Swift From ModelScope community; supports a wide range of domestic and international mainstream models; provides rich fine-tuning algorithms and quantization options

    Core Features

    • One-Click Launch: Click the fine-tuning button on the model details page to automatically allocate compute and mount the model environment.
    • Full Lifecycle Management: Covers fine-tuning creation, training metric monitoring, and model export.
    • Visual Framework UI: LLaMA-Factory includes a visual training panel — configure training parameters without the command line.
    • Flexible Compute: Select different GPU resource specifications on demand.
    • Model Export: After fine-tuning, export the trained weights as an independent model repository.

    Fine-tuning Workflow

    Select base model → Create fine-tuning instance → Configure dataset and training parameters → Start training → Monitor training metrics → Export fine-tuned model