Skip to main content

One-Click Fine-Tuning

What is One-Click Fine-Tuning?​

The OpenCSG Community offers high-performance GPU-accelerated managed fine-tuning instances. Supporting mainstream frameworks (such as LLaMA-Factory and MS-Swift), users simply select their desired compute resources and datasets to quickly initiate large model fine-tuning.

Core Features​

  • Flexible Dedicated Compute: Select from various GPU specifications on-demand.
  • Rich Framework Integration: Built-in support for popular fine-tuning frameworks like LLaMA-Factory (with GUI) and MS-Swift.
  • Full-Lifecycle Management: Covers instance creation, metric monitoring, and model exporting.
  • One-Click Evaluation: Provides model evaluation services to help quickly verify performance improvements after fine-tuning.

Demonstration​

Please refer to the following video for a walkthrough of the fine-tuning process:

Process Overview​

The complete workflow for fine-tuning is as follows:

  1. Create Fine-Tuning Instance: Select the base model and compute resources.
  2. Use and Monitor: Configure the dataset and observe training in real-time.
  3. Export Model: Save the fine-tuned weights as a new, independent repository.