GLM-4-Instruct-4K-zh: 4,000 Chinese Instruction-Response Pairs for LLM Training

A collection of 4,000 high-quality Chinese instruction-response pairs for training large language models. The dataset was created by author 'rqq' by translating questions from the 'Sao10K/Claude-3-Opus-Instruct-5K' dataset into Chinese and generating responses using the GLM-4 model. It was last updated on May 6, 2024.

Use Cases

• Fine-tuning instruction-following models for Chinese based on the provided 4,000 question-answer pairs.
• Training or benchmarking conversational AI agents using the structured Alpaca format mentioned in the description.
• Studying the translation and adaptation of instruction datasets from one language to another.
• Developing or evaluating Chinese language models' ability to follow complex instructions.

Strengths

• Contains 4,000 high-quality responses generated by the GLM-4 model.
• Questions are derived from a noted source dataset, 'Sao10K/Claude-3-Opus-Instruct-5K', and translated into Chinese.
• Formatted in Alpaca format, which is directly compatible with popular training frameworks like LLaMA-Factory.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the companion translation dataset is unspecified.
• The dataset's license is unknown, which may restrict commercial or research use.

Provenance

Source

rqq on Hugging Face

Collection Method

Questions were translated from the 'Sao10K/Claude-3-Opus-Instruct-5K' dataset, and responses were generated by the GLM-4 model.

Freshness

Last updated 2024-05-06 07:23:43; freshness should be verified.