Tw Legal Synthetic Qa

A synthetic dialogue dataset for supervised fine-tuning (SFT) of language models on legal topics. It was created by lianghsun using the THUDM/chatglm3-6b-32k model and the lianghsun/tw-processed-judgments dataset, with prompts designed to generate question-answer pairs. The dataset is in Traditional Chinese and was last updated on August 1, 2024.

Use Cases

• Supervised fine-tuning (SFT) of language models based on the described legal dialogue structure.
• Training models to answer legal questions in Traditional Chinese based on the described user-assistant format.
• Benchmarking model performance on synthetic legal question-answering tasks.
• Studying the generation of domain-specific synthetic training data for law.

Strengths

• Dataset is specifically designed for supervised fine-tuning (SFT), as stated in the description.
• Data instances follow a structured user-assistant dialogue format for legal questions and answers.
• Content is in Traditional Chinese, targeting a specific linguistic and legal jurisdiction.

Limitations

• Row count, file formats, and column-level documentation are unknown, limiting suitability assessment.
• Description metadata is limited; actual data quality and structure require manual inspection after download.
• Data is synthetic, generated by a model, which may introduce artifacts or biases not present in real legal dialogues.

Provenance

Source

lianghsun on Hugging Face, derived from lianghsun/tw-processed-judgments and generated using THUDM/chatglm3-6b-32k.

Collection Method

Synthetically generated via prompted language model based on a processed judgments dataset.

Freshness

Last updated 2024-08-01 08:10:08; freshness should be verified.

Geography

Likely Taiwan, given the use of Traditional Chinese and reference to 'tw' in the title and source data.