Skip to main content

😍 Building a Thai Chat Conversation Dataset Together!

The collaborative website for building a Thai chat conversation dataset​

Collaborative website for building the Thai chat conversation datasettag.openthaigpt.aieat.or.th

To help the OpenThaiGPT team obtain a dataset for developing Thai InstructGPT and the Reward Model (RM) as quickly as possible, the team built a website for collectively creating this dataset by forking the LAION-AI/Open-Assistant website, which is open source under the Apache 2.0 license.

Q/A​

Q: Why did you choose to fork the Open Assistant website and host it yourselves? Why not build our own, or simply use Open Assistant directly?

A: There are several advantages to forking the Open Assistant website.

  1. No time spent on building our own system.
  2. We continuously receive updates and bug fixes from Open Assistant.
  3. OpenThaiGPT developers receive the raw dataset immediately whenever someone contributes on the website, without waiting for Open Assistant to publish the data.
  4. The dataset from OpenThaiGPT can easily be passed on to Open Assistant, since both are produced by the same system.
  5. We gain a world-class standard pipeline and guideline for dataset creation.
  6. Because we focus on Thai alone, we can release datasets more frequently (every Sunday).
  7. It can be extended to other data tagging tasks that OpenThaiGPT may need in the future, and the system can be adapted to support other Southeast Asian languages, such as CLMV, more quickly.

Q: Is the source code of the OpenThaiGPT dataset creation website open?

A: We have released the source code of the OpenThaiGPT dataset creation website under the Apache 2.0 license at OpenThaiGPT/Open-Assistant on GitHub