118
stars
8
commits
1
linked in READMEs
Aug 7, 2025
updated
README
Dataset summary
Multilingual-Thinking is a reasoning dataset where the chain-of-thought has been translated from English into one of 4 languages: Spanish, French, Italian, and German. The dataset was created by sampling 1k training samples from the SystemChat subset of SmolTalk2 and translating the reasoning traces with another language model.
This dataset was used in the OpenAI Cookbook to fine-tune the OpenAI gpt-oss models.
You can load the dataset using:
from datasets import load_dataset
ds = load_dataset("HuggingFaceH4/Multilingual-Thinking", split="train")
The gpt-oss models were trained on the Harmony response format for defining conversation structures, generating reasoning output and structuring function calls. The format is designed to mimic the OpenAI Responses API, and the table below summarizes the different message types used in the dataset:
developer | The developer message is used to provide custom instructions for the model (what we usually call the system role) |
|---|---|
user | The user message is used to provide the input to the model |
assistant | Output by the model which can either be a tool call or a message output. The output might also be associated with a particular “channel” identifying what the intent of the message is. |
analysis | These are messages that are being used by the model for its chain-of thought |
final | Messages tagged in the final channel are messages intended to be shown to the end-user and represent the responses from the model. |
messages | The list of messages that combine the content of the above to produce a full conversation. This is the input to the model. |
If you're familiar with OpenAI's messages format, you will recognise this as being quite similar, but with an important difference:
The
assistantturn contains two special fields: athinkingone which contains the model's reasoning process, and acontentone which contains the final response to the user.
Contributors
8 commits
Similar repos
118
stars
8
commits
1
linked in READMEs
Aug 7, 2025
updated
README
Dataset summary
Multilingual-Thinking is a reasoning dataset where the chain-of-thought has been translated from English into one of 4 languages: Spanish, French, Italian, and German. The dataset was created by sampling 1k training samples from the SystemChat subset of SmolTalk2 and translating the reasoning traces with another language model.
This dataset was used in the OpenAI Cookbook to fine-tune the OpenAI gpt-oss models.
You can load the dataset using:
from datasets import load_dataset
ds = load_dataset("HuggingFaceH4/Multilingual-Thinking", split="train")
The gpt-oss models were trained on the Harmony response format for defining conversation structures, generating reasoning output and structuring function calls. The format is designed to mimic the OpenAI Responses API, and the table below summarizes the different message types used in the dataset:
developer | The developer message is used to provide custom instructions for the model (what we usually call the system role) |
|---|---|
user | The user message is used to provide the input to the model |
assistant | Output by the model which can either be a tool call or a message output. The output might also be associated with a particular “channel” identifying what the intent of the message is. |
analysis | These are messages that are being used by the model for its chain-of thought |
final | Messages tagged in the final channel are messages intended to be shown to the end-user and represent the responses from the model. |
messages | The list of messages that combine the content of the above to produce a full conversation. This is the input to the model. |
If you're familiar with OpenAI's messages format, you will recognise this as being quite similar, but with an important difference:
The
assistantturn contains two special fields: athinkingone which contains the model's reasoning process, and acontentone which contains the final response to the user.
Similar repos
Linked in READMEs
Contributors
8 commits