Synthetic Data Generated from IETF mailing lists.

The generated synthetic data was used to train retrieval models.

The synthetic data was generated using https://github.com/cheop-byeon/synthetic-data-kit, a tool-kit derived from https://github.com/meta-llama/synthetic-data-kit.

The dataset could be downloaded using the provided script download_RFCAlign.py. The topic is the corresponding WG (working group) name.

RFCAlign Dataset

Basic Download

# Download entire RFCAlign dataset
python download_RFCAlign.py

Download Whole Repository with huggingface-cli

# Install CLI (if needed)
pip install -U "huggingface_hub[cli]"

# Download full RFCAlign repository to local folder
huggingface-cli download jiebi/RFCAlign --repo-type dataset --local-dir ./dataset/RFCAlign

Parameter Options

python download_RFCAlign.py [--split <value>] [--topic <value>] [--no-download]

--split

Top-level folder to download.

Allowed values:

• llama_non-verbose
• llama_verbose
• qwen_non-verbose
• qwen_verbose

--topic

Optional topic file name without .jsonl.

Rules:

• Must be used together with --split
• Downloads only one file: <split>/<topic>.jsonl

Examples:

• --topic ace
• --topic quic
• --topic tls

--no-download

Inspect remote repository structure and local folder tree only. No files are downloaded.

Usage Examples

# 1) Inspect only (no download)
python download_RFCAlign.py --no-download

# 2) Download full RFCAlign dataset
python download_RFCAlign.py

# 3) Download one split folder only
python download_RFCAlign.py --split qwen_verbose

# 4) Download one specific file only
python download_RFCAlign.py --split qwen_verbose --topic ace

# 5) Another one-file download example
python download_RFCAlign.py --split llama_non-verbose --topic tls

Notes

• Download target directory: ./dataset/RFCAlign/
• Full download: all available files under all split folders
• Split download: only files under selected split
• Split + topic download: only one .jsonl file

The data was used for retrieval tasks, for training (https://github.com/cheop-byeon/FlagEmbedding) and for evaluation (https://github.com/cheop-byeon/mteb-R2Gen).