Domofon-Cot-Conversations-700k: Synthetic XML Data for Language Model Training

700,000 synthetic XML conversations generated for training small language models. The dataset, created by user 'domofon', contains cleaned conversations from six families, including multi-turn factual dialogues and text-processing instructions. It was last updated on Hugging Face on May 31, -0026.

Use Cases

• Training models on chain-of-thought reasoning based on multi-turn factual conversations.
• Fine-tuning models for instruction following based on text-processing instructions.
• Teaching models XML formatting and structure based on the synthetic XML data.
• Training models on tool-use traces based on the described tool-call patterns.

Strengths

• Contains 700,000 conversation examples, providing substantial training material.
• Data is cleaned and organized into six distinct families, including 'conv' and 'instruct'.
• Last update timestamp is precisely recorded as 2026-05-31 18:23:04.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

Hugging Face user 'domofon'.

Collection Method

Synthetically generated data.

Time Range

Synthetic data; no real-world temporal coverage.

Freshness

Last updated 2026-05-31 18:23:04; freshness should be verified.

Geography

Synthetic data; no real-world spatial coverage.