README.md

metadata

license: mit
task_categories:
  - text-generation
  - token-classification
language:
  - en
tags:
  - finance
  - xbrl
  - numeric-tagging
  - datatype-extraction
  - fintagging
size_categories:
  - 1K<n<10K
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train.parquet
      - split: test
        path: data/test.parquet

FinTagging Value-Type Extraction SFT Data

This dataset is derived from the sampled FinTagging 800/200 split. It defines an LLM instruction-following subtask where the model input is a prompt plus the financial table/text, and the model output is a JSON list of numeric entity/datatype pairs.

Task Format

• query: full model input, including the instruction template and financial table/text.
• answer: JSON string target. It is a list of objects with numeric_entity and datatype.
• context: raw financial table/text without the instruction prompt.
• output_entities: structured copy of answer for inspection.

Example answer:

[
  {"numeric_entity": "62", "datatype": "monetaryItemType"},
  {"numeric_entity": "171", "datatype": "monetaryItemType"}
]

Splits

Split	Samples	Output entries	Unique datatypes
train	800	11,687	5
test	200	2,546	5

Datatype Counts

Datatype	Train count	Test count
monetaryItemType	10,448	2,285
percentItemType	574	116
sharesItemType	375	70
perShareItemType	219	58
integerItemType	71	17

Columns

• source_sample_idx: original source row index.
• context_id: original context identifier.
• split: split label.
• query: prompt plus financial table/text.
• answer: JSON list target string.
• context: raw table/text.
• output_entities: structured target list.