Puffin: 3,000 Multi-Turn GPT-4 Conversations

Exactly 3,000 examples comprise this dataset, with each response created using GPT-4. It contains over 2,000 multi-turn conversations between GPT-4 and real humans. The dataset was created by author LDJnr and last updated on June 7, 2024.

Use Cases

• Fine-tuning language models for multi-turn dialogue based on the described conversational structure.
• Benchmarking model performance on long-context tasks based on the average context length of over 1,000 tokens.
• Studying human-AI interaction patterns based on the real human conversations with GPT-4.
• Training models for instruction following based on the dataset's nature as a collection of GPT-4 responses.

Strengths

• Contains exactly 3,000 examples, providing a defined scale.
• All responses were created using GPT-4, indicating a consistent, high-quality source.
• Comprised of over 2,000 multi-turn conversations, offering complex interaction data.
• Average context length per conversation is over 1,000 tokens, suggesting substantive exchanges.

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.
• The author notes a newer version called Pure-Dove is recommended, making this version potentially outdated.

Provenance

Source

huggingface

Collection Method

Conversations between GPT-4 and real humans, curated by LDJnr.

Freshness

Last updated 2024-06-07 20:15:15; freshness should be verified.