Skip to content

Tele-FLM-1T

An open 1T multilingual model with shared training details

Visit official site
Category AI models

What Tele-FLM-1T is

Tele-FLM-1T is described as an open-source 1T multilingual language model with a GPT-style decoder-only Transformer design, released with its training details.

What you can do with it

  • Generate text across languages
  • Study the shared core design
  • Adapt the model to a task

Model notes

  • Growth technology uses three stages
  • RoPE, RMSNorm and SwiGLU are listed
  • Training uses 3D parallelism

Getting started

  • Find the model on Hugging Face
  • Read the card for details and limits
  • Download weights and related code
  • Adjust the model using shared practices

Pros & cons

✓ What we like

  • Compatibility with the Llama architecture is stated
  • Three-dimensional parallelism combines data, tensor and pipeline parallel work

! What to watch out for

  • The documented setup uses a large GPU cluster
  • Model card limits are read before downloading

FAQ

How many tokens were used?

The material states training on about 2T tokens.

What is shared besides weights?

Core design, engineering practice and training details are shared.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review