Skip to content

TeleChat3

A large language model trained on domestic hardware

Visit official site

What TeleChat3 is

TeleChat3 is described as a high-performance large language model from the AI research institute of China Telecom, trained on domestic computing hardware and showing performance on knowledge, creation, instruction and code.

What you can do with it

  • Run inference on a single card or across several cards
  • Fine-tune the model and merge weights with LLaMA-Factory
  • Serve requests quickly through vLLM and SGLang

Application scenarios

  • Draft prose for a writing assistant tool
  • Answer knowledge questions in an education setting
  • Build a code helper on the code generation ability

Getting started

  • Download the model and the matching libraries
  • Import libraries such as transformers and torch
  • Load the model and tokenizer and set the inference device
  • Format input data into the structure the model expects

Pros & cons

✓ What we like

  • Evaluation covers knowledge, math and creation
  • Fast serving works with vLLM and SGLang

! What to watch out for

  • The described training relies on domestic computing hardware
  • Serving is tied to the named inference frameworks

FAQ

Who developed the model?

The material names the AI research institute of China Telecom.

How is fine-tuning done?

The material points to LLaMA-Factory for fine-tuning and weight merging.

Last reviewed: 2026-11-15

More AI coding tools tools

View all →

How we review