Tatsuya-Ichinose/CoDIT-Qwen3-8B

CoDIT-Qwen3-8B

CoDIT-Qwen3-8B is a synthetic conversation dataset derived from [LMSYS-Chat-1M](https://huggingface.co/datasets/lmsys/lmsys-chat-1m) [[Zhang+, ICLR24]](https://openreview.net/forum?id=BOfDKxfwt0).

• —250,333 user instructions sourced from LMSYS-Chat-1M
• —250,333 assistant responses automatically synthesized using CoDIT with Qwen/Qwen3-8B(No Thinking), generating one response per user instruction.

[!NOTE] This repository contains only the synthesized assistant responses provided as a delta to the original LMSYS-Chat-1M dataset. User instructions are not included and must be obtained separately from LMSYS-Chat-1M.

License Information

User Instructions

• —Governed by the [LMSYS-Chat-1M Dataset License Agreement](https://huggingface.co/datasets/lmsys/lmsys-chat-1m#lmsys-chat-1m-dataset-license-agreement)

Assistant Responses

• —Distributed under the Apache License, Version 2.0.

🔑 Usage and Access

To use this dataset, you must first obtain the original LMSYS-Chat-1M dataset by agreeing to its license.

• —📄 License agreement: https://huggingface.co/datasets/lmsys/lmsys-chat-1m#lmsys-chat-1m-dataset-license-agreement

🚀 Usage Example

Step 1: Obtain Access to LMSYS-Chat-1M

[!TIP] Make sure you have accepted the LMSYS-Chat-1M license before accessing the dataset.

Accept the license to enable access: https://huggingface.co/datasets/lmsys/lmsys-chat-1m

Step 2: Merge Instructions and Responses

Merge user instructions with synthesized responses using conversation_id:

from datasets import load_dataset

# Load original instructions
ds1 = load_dataset("lmsys/lmsys-chat-1m", split="train")

# Load synthesized responses
ds2 = load_dataset("Tatsuya-Ichinose/CoDIT-Qwen3-8B", split="train")
ds2_map = {x["conversation_id"]: x["response"] for x in ds2}

# Merge instructions and responses
dataset = (
    ds1
    .filter(lambda x: x["conversation_id"] in ds2_map)
    .map(
        lambda x: {
            "conversation_id": x["conversation_id"],
            "conversation": [
                {"role": "user", "content": x["conversation"][0]["content"]},
                {"role": "assistant", "content": ds2_map[x["conversation_id"]]},
            ],
        },
        remove_columns=ds1.column_names,
    )
)

Acknowledgments

This work is based on Qwen3. We acknowledge and thank the Qwen Team for their contributions and the release of these models.

End of document