OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Infrastructure & Serving

LLM Tpu

LLM Tpu is an open-source developer repository for deploying generative language and multimodal models on SOPHGO TPU chips. It includes model demos, a command-line quick start and a conversion workflow that uses TPU-MLIR to turn Hugging Face weights into bmodels.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

LLM Tpu — OpenViST project profile in AI Infrastructure & Serving.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use LLM Tpu when deploying one of its documented models on a supported SOPHGO TPU, or when preparing model weights for that hardware through the TPU-MLIR conversion workflow.

Who it is for

  • Developers deploying language models on SOPHGO BM1684X, BM1688 or CV186X TPUs.
  • Teams evaluating documented multimodal model demos for supported SOPHGO hardware.
  • Engineers preparing model weights for TPU inference with TPU-MLIR.

Documented uses

  • Select a documented model demo and invoke it with a command-line model option.
  • Convert Hugging Face weights to a bmodel with the llm_convert.py and TPU-MLIR workflow.
  • Use supplied Python or C++ reference demos where listed for a model.
Before you choose

Requirements & limits

The repository is public and identified as open source; its README displays an Apache-2.0 license badge.

  • The named hardware targets are BM1684X, BM1688 and CV186X.
  • The README lists x86 and aarch64 platforms and Python 3.7 or later.
  • TPU-MLIR is required for compiling weights; the repository also points to precompiled bmodels in model demos.
  • Supported chips, compilation availability and demo languages differ by model, so check the individual model entry.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Prepare a TPU-MLIR environment for model conversion

Set up the documented conversion environment needed to export Hugging Face weights to a bmodel.

Before you start

  • A SOPHGO TPU target such as BM1684X, BM1688 or CV186X
  • TPU-MLIR, using either Docker or a source build

Steps

  1. Obtain the LLM-TPU repository.

  2. Prepare TPU-MLIR using Docker or a source build.

  3. Use the repository's llm_convert.py workflow to export Hugging Face weights to a bmodel.

Expected outcome

A bmodel produced through the repository's documented conversion path.

Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
18 October 2026
Original website title
sophgo/LLM-TPU · Official homepage title · retained in its source language
How we review sourcesSuggest a correction