FLANG-SL1000: Sequence Labeling for Financial KPI Extraction

Model Description

FLANG-SL1000 is a FLANG-based sequence labeling model fine-tuned on the HiFi-KPI dataset for extracting financial key performance indicators (KPIs) from English SEC earnings filings (10-K and 10-Q reports).

The model starts from SALT-NLP/FLANG-BERT, a financial-domain BERT model. It is the FLANG-based counterpart of BERT-SL1000 in the HiFi-KPI model collection. The dataset and extraction tasks are described in HiFi-KPI: A Dataset for Hierarchical KPI Extraction from Earnings Filings.

It predicts the 1,000 most frequent original iXBRL entity labels from HiFi-KPI, together with the outside label O.

• Base model: SALT-NLP/FLANG-BERT
• Model architecture: BertForTokenClassification
• Task: Token classification (sequence labeling)
• Language: English
• Dataset: AAU-NLP/HiFi-KPI

Use Cases

• Extracting financial KPI mentions from SEC 10-K and 10-Q filings.
• Financial document parsing with iXBRL-based entity recognition.
• Mapping financial mentions to their original iXBRL labels.

Label Space

The checkpoint has 1,001 output labels: 1,000 financial entity labels and O. The exact label-to-ID mapping is stored in config.json.

The saved labels use plain entity names rather than BIO prefixes. Predictions are token-level labels; this checkpoint does not itself return structured records containing financial values, dates, or currencies.

Usage

pip install -U transformers torch

from transformers import pipeline

extract_kpis = pipeline(
    "token-classification",
    model="AAU-NLP/FLANG-SL1000",
    tokenizer="AAU-NLP/FLANG-SL1000",
    aggregation_strategy="none",
)

predictions = extract_kpis("Revenue was $12.5 million for the quarter.")
print(predictions)

The example returns subword-token predictions with labels, scores, and character offsets. Long filings should be split into paragraphs or chunks that fit the model's input window.

Dataset & Code

• Paper: HiFi-KPI: A Dataset for Hierarchical KPI Extraction from Earnings Filings
• Dataset: HiFi-KPI on Hugging Face
• Code: HiFi-KPI GitHub Repository
• Base model: SALT-NLP/FLANG-BERT
• BERT counterpart: BERT-SL1000

Citation

If you use this model or the dataset, please cite:

@inproceedings{aavang-etal-2026-hifi,
    title = "{H}i{F}i-{KPI}: A Dataset for Hierarchical {KPI} Extraction from Earnings Filings",
    author = "Aavang, Rasmus T.  and
      Rizzi, Giovanni  and
      Tjalk-B{\o}ggild, Rasmus  and
      Iolov, Alexandre  and
      Zhang, Mike  and
      Bjerva, Johannes",
    editor = "Piperidis, Stelios  and
      Bel, N{\'u}ria  and
      van den Heuvel, Henk  and
      Ide, Nancy  and
      Krek, Simon  and
      Toral, Antonio",
    booktitle = "Proceedings of the Fifteenth Language Resources and Evaluation Conference",
    month = may,
    year = "2026",
    address = "Palma de Mallorca, Spain",
    publisher = "ELRA Language Resource Association",
    url = "https://aclanthology.org/2026.lrec-1.30/",
    doi = "10.63317/2nbsp7zzfb3g",
    pages = "441--455",
}

License

This model is licensed under the Apache License, Version 2.0.