LLaMA-Factory
ms-swift
VeRL
Reuse your GPUs and models
Install it on your own servers, beside your GPUs, and connect the model services you already run.
Products
Micius Ink
A model training and deployment platform for private enterprise use, the intelligent foundation for enterprise data and models.
How it works
From your documents to a model in production.
01 · Data preparation
Turn SOPs and 8D reports into cited Q&A
Skip writing training data by hand. Upload PDF, Word, PowerPoint or structured data, and Micius drafts question-and-answer pairs tied to their sources.
02 · Annotation and review
Your experts fix the answers, no code needed
Before training, a domain expert checks every answer against its source, edits and labels it, and approves or returns it, singly or in batches.
03 · Training
Build your own model on Micius-V1.0-Flash
Start from Micius-V1.0-Flash or your own model and train it for your task, such as routing quality anomalies to the right team, with LLaMA-Factory, ms-swift or VeRL.
04 · Evaluation
See whether training actually helped
Before you deploy, see how far each checkpoint beats the base model on your own test set, and which cases it still gets wrong.
05 · Deployment
Deploy it once, use it across your apps
Deploy the trained model on your own servers. Pick it in Lumen or Micius Code, or call it from your own systems over the API.
Enterprise
Runs on your servers. Every model traces back to its data.
Trace any model to its data
If a deployed model goes wrong, trace it back to the data version, training run and evaluation behind it. Roles are set per project and every action is logged.
Put what your engineers know into a model.
Tell us about your data and what the model should do. We will plan the deployment with you.