ms-swift

ms-swift is a framework for large language models and multi-mode models covering education, inference, evaluation and distribution processes. It supports PEFT and full-parameter fine-tuning methods, while Megatron is equipped with parallelism techniques and various reinforcement learning algorithms. The structure, which can work on a large number of hardware architectures, allows for large-scale model training with memory optimization and quantization technologies.

MS

modelscope · Local AI / Self-Hosted LLM

STARS16.0K
FORKS1.8K
COMMITS3.7K
CONTRIBUTORS213
VERSIONv4.5.3
UPDATEDyesterday
PythonLLM Fine-tuningMultimodalModel DağıtımıApache-2.0

Common commercial products in this category: ChatGPT, Claude, Gemini

What can it do?
  • You can train 600+ text-based and 400+ multi-mode models with your own data sets.
  • Using tools like vLLM, SGLang, and LMDeploy, you can quickly distribute your models for inference.
  • You can train models in hardware resources limited to LoRA, QLoRA and various quantization techniques.
Heads up
  • • High number of open issues (602)

My note

Alternative pages

More in: Local AI / Self-Hosted LLM