Datasets:

Asharox
/

summarization-sft-100k

language:
  - en
license: mit
task_categories:
  - text-classification
tags:
  - sentiment
  - alpaca
  - sft
  - instruction-tuning
size_categories:
  - 100K<n<1M
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*.parquet

summarization-sft-100k

Sentiment classification dataset in Alpaca instruction format for supervised fine-tuning (SFT).

100,000 examples sourced and normalised from:

mteb/mteb-human-tweet-sentiment-classification
stanfordnlp/sst2
SetFit/sst5 (5-class collapsed to 3)
jbeno/sentiment_merged

Label space

All labels normalised to 3 classes: negative, neutral, positive

Format

Column	Description
`instruction`	Task prompt with label options
`input`	The text to classify
`output`	One of: negative, neutral, positive

Example

{
  "instruction": "Classify the sentiment of the following text. Respond with one of: negative, neutral, positive.",
  "input": "This was an absolutely wonderful experience.",
  "output": "positive"
}