Homicide Alpaca: 11,103 Question-Answer Pairs for Instruction Tuning

Homicide Alpaca is an Alpaca-styled export of the homicide-investigation dataset, created by 'theprint' and last updated on August 23, 2024. It contains 11,103 entries of question-answer pairs, with an average question length of 91 characters and an average answer length of 775 characters. The dataset includes confidence scores for the entries, with a top score of 0.74 and a median of 0.46.

Use Cases

• Fine-tuning language models for question-answering based on the structured Q&A format.
• Training models to generate long-form text responses based on the provided answer length statistics.
• Evaluating model confidence calibration using the provided confidence scores for each entry.
• Instruction-tuning for investigative or domain-specific dialogue systems based on the homicide investigation theme.

Strengths

• Contains 11,103 instruction-response pairs, providing substantial volume for training.
• Includes confidence scores (average 0.48, median 0.46) for assessing answer quality.
• Features detailed answers with an average length of 775 characters, suitable for long-form generation.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Description metadata is limited; actual data quality requires manual inspection after download.
• The source 'homicide-investigation' dataset is referenced but not described, limiting context.

Provenance

Source

theprint/homicide-investigation dataset

Collection Method

Alpaca-styled export of an original dataset.

Freshness

Last updated 2024-08-23 16:15:47; freshness should be verified.