Haruhi Baize Role-Playing Conversations from Multiple AI Models

Haruhi-Zero's conversation training data for extending ChatHaruhi from few-shot to zero-shot. The dataset contains synthetic role-playing dialogues generated by interacting Chinese chatbots using various AI model combinations, including OpenAI, GLM, Claude, and Baichuan.

Use Cases

• Train a Chinese conversational model on the Baize-style dialogue patterns generated by different AI model combinations.
• Analyze stylistic differences in role-playing conversations between specific model pairings like Claude_openai.
• Use the recorded prompt field for id0 to study the influence of initial character context on dialogue generation.

Strengths

• Dialogues are generated by multiple advanced AI models (OpenAI, GLM, Claude, Baichuan).
• Data is structured for a specific research goal: extending a chatbot from few-shot to zero-shot capability.
• Character card information for ids is linked to an external, documented source.

Limitations

• Unknown sample size and column structure prevent assessment of scale and feature richness.
• Data is synthetic, generated by AI models, which may introduce artifacts or biases from those models.
• Focus is on Chinese language and specific character role-play, limiting generalizability.

Provenance

Source

Hugging Face user silk-road.

Collection Method

Synthetic data generated by having AI chatbots perform Baize-style mutual conversations based on Chinese character cards.

Time Range

null

Freshness

Last updated January 15, 2024.

Geography

null