xTuring is a Python tool to fine-tune, evaluate, and run private, personalized LLMs with LoRA and low-precision support, aiming for local/private cloud usage.
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (1 so far).
What it is
xTuring is a Python library to fine-tune open-source LLMs on user data, with private-by-default operation (local or VPC), and support for LoRA and INT8/INT4 precision.
How it works
The project provides a simple API for data preparation, training, and inference, with built-in evaluation metrics (e.g., perplexity) and multiple model backends (e.g., GPT-OSS, LLaMA, Qwen3, Mistral, Mamba). It supports loading pre-finetuned checkpoints and various precision and LoRA configurations through model wrappers like BaseModel, GenericLoraKbitModel, and Llama2.
Getting started
Installation:
pip install xturing
Quickstart example shows loading a toy dataset and a small CPU-friendly model, then finetuning and generating:
from xturing.datasets import InstructionDataset
from xturing.models import BaseModel
dataset = 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}")
Recent releases
Latest releases include v0.1.8 (2023-09-07): Stability and documentation; v0.1.7 (2023-07-26): Evaluation and CPU support; v0.1.6 (2023-07-12): INT4 models; v0.1.5 (2023-06-14): BaseModel/LoRA changes; v0.1.4 (2023-06-13): New models and logging.
Traction
Stars: 2673. Forks: 211. Open issues: 14. Language: Python. License: Apache-2.0.
Behind the repo
No linked startup/company information provided in the facts.
Caveats
License: Apache-2.0. Created: 2023-03-19. Last push: 2026-03-04. Open issues: 14.