finnish-instructions

Centralized repo for Finnish instruction data. License?

Format

instruction: mandatory. May include all the required context.

input: optional. Used as an optional context for instructions.

output: Generated response.

Examples:

instruction: "Laadi kysymys annetusta tekstistä."
input: "Teksti: Sandra siirtyi toimistoon. Sandra meni kylpyhuoneeseen. Mary meni makuuhuoneeseen. Daniel siirtyi eteiseen."
output: "Missä on Daniel?"

or

instruction: "Kumpi on parempaa suomenkieltä, vaihtoehto a) {a} vai b) {b}"
input: ""
output: "a"

Datasets:

TODOs

• Anthropic HH-RLHF

Ready / contains already usable material

1. Machine translated from the original English using DeepL

• dolly-fi: Finnish version of the databricks-dolly-15k instruction dataset
• oasst-fi Finnish version of OpenAssistant dataset v1
• natural-instructions-fi is an on-going process of machine translating manually selected tasks.
• Lima-fi

2. Synthetic datasets

TODOs

• OCR-correction
• Masked word prediction
• Recognize text from space-separated characters
• Recognize text from characters without space

Ready

3. Native Finnish datasets:

Ready

• Paraphrase
• Question Answering

Processing