xTuring
Build, personalize and control your own LLMs. From data pre-processing to fine-tuning, xTuring provides an easy way to personalize open-source LLMs. Join our di
What is xTuring?
Build, personalize and control your own LLMs. From data pre-processing to fine-tuning, xTuring provides an easy way to personalize open-source LLMs. Join our discord community: https://discord.gg/TgHXuSJEk6
Overview of xTuring
What it is
xTuring is an open-source library for fine-tuning, evaluating, and running private, personalized large language models (LLMs). It supports open-source models such as GPT-OSS, LLaMA/LLaMA 2, Qwen3, MiniMax M2, GPT-J, GPT-2, DistilGPT-2, and Mamba. The library is designed to work locally or in a private cloud, with a focus on simplicity, speed, and cost efficiency.
Key features
- Simple API for data preparation, training, and inference
- Private by default: runs locally or in your VPC
- Efficient fine-tuning using LoRA and low-precision (INT8/INT4) to reduce costs
- Scales from CPU/laptop to multi-GPU setups
- Built-in evaluation metrics, including perplexity
- Supports multiple model architectures and variants (e.g., GPT-OSS 120B/20B, LLaMA 2, Qwen3 0.6B)
- CLI and UI playground for interactive use
Use cases
- Fine-tuning open-source LLMs on custom instruction datasets (e.g., Alpaca format)
- Evaluating model performance with perplexity on any causal LM dataset
- Running inference on CPUs using Intel® Extension for Transformers with weight-only quantization
- Generating instruction datasets locally using Qwen3-Omni multimodal checkpoints
- Deploying private, personalized LLMs without sending data to external servers
Getting started
bashpip install xturing``````pythonfrom xturing.datasets import InstructionDatasetfrom xturing.models import BaseModeldataset = InstructionDataset('./examples/models/llama/alpaca_data')model = BaseModel.create('qwen3_0_6b_lora')model.finetune(dataset=dataset)output = model.generate(texts=['Explain quantum computing for beginners.'])print(f'Model output: {output}')
License and pricing
Apache-2.0 license, open source.
Based on the project's public README and repository metadata.