Hugging Face's logo

alea-institute
/
kl3m-doc-pico-contracts-001

Fill-Mask
Transformers
Safetensors
English
roberta
kl3m
legal
financial
contracts
mlm
Model card Files Community
kl3m-doc-pico-contracts-001 / README.md
alea-institute's picture
alea-institute
Update README and config files - README.md
9b2d083 verified
preview code
|
raw
history blame contribute delete
13 kB
metadata
language:
  - en
library_name: transformers
license: cc-by-4.0
tags:
  - kl3m
  - legal
  - financial
  - contracts
  - mlm
  - roberta
pipeline_tag: fill-mask
widget:
  - text: <|cls|> This<|mask|> Credit Agreement is hereby <|sep|>
  - example: >-
      <|cls|> The<|mask|> Agreement contains the entire understanding between
      <|sep|>
  - temperature: 0.7
  - do_sample: true
date: '2025-02-04T00:00:00.000Z'

kl3m-doc-pico-contracts-001

kl3m-doc-pico-contracts-001 is a specialized masked language model (MLM) based on the RoBERTa architecture, specifically fine-tuned for contract analysis. Built upon the foundation of kl3m-doc-pico-001, this model was trained for an additional 500,000 steps on contract data, making it particularly effective for contract-related terminology and structures.

Model Details

Use Cases

This model is particularly useful for:

  • Contract classification and analysis
  • Contract clause identification and extraction
  • Understanding contract-specific terminology
  • Filling in missing terms in contracts and agreements
  • Entity recognition in legal contracts
  • Feature extraction for downstream contract analysis tasks

The model demonstrates strong knowledge of contract-specific terminology and common contractual structures, as shown in the examples below.

Usage

You can use this model for masked language modeling with the simple pipeline API:

from transformers import pipeline

# Load the fill-mask pipeline with the model
fill_mask = pipeline('fill-mask', model="alea-institute/kl3m-doc-pico-contracts-001")

# Example: Contract clause heading
# Note the mask token placement - directly adjacent to "AND" without space
text = "<|cls|> 8. REPRESENTATIONS AND<|mask|>. Each party hereby represents and warrants to the other party as of the date hereof as follows: <|sep|>"
results = fill_mask(text)

# Display predictions
print("Top predictions:")
for i, result in enumerate(results[:5]):
    print(f"{i+1}. {result['token_str']} (score: {result['score']:.4f})")

# Output:
# Top predictions:
# 1. WARRANTIES (score: 0.7282)
# 2. Warranties (score: 0.1674)
# 3. REPRESENTATIONS (score: 0.0377)
# 4. warrants (score: 0.0127)
# 5. WARR (score: 0.0074)

You can try additional examples:

# Example: Defined term
text2 = "<|cls|> \"Effective<|mask|>\" means the date on which all conditions precedent set forth in Article V are satisfied or waived by the Administrative Agent. <|sep|>"
results2 = fill_mask(text2)

# Display predictions
print("\nDefined Term Example - Top 5 predictions:")
for i, result in enumerate(results2[:5]):
    print(f"{i+1}. {result['token_str']} (score: {result['score']:.4f})")

# Output:
# Defined Term Example - Top 5 predictions:
# 1. Date (score: 0.9735)
# 2. Time (score: 0.0075)
# 3. date (score: 0.0053)
# 4. Dates (score: 0.0024)
# 5. Agreement (score: 0.0011)

# Example: Regulatory context
text3 = "<|cls|> All transactions shall comply with the requirements set forth in the Truth in<|mask|> Act and its implementing Regulation Z. <|sep|>"
results3 = fill_mask(text3)

# Display predictions
print("\nRegulatory Example - Top 5 predictions:")
for i, result in enumerate(results3[:5]):
    print(f"{i+1}. {result['token_str']} (score: {result['score']:.4f})")

# Output:
# Regulatory Example - Top 5 predictions:
# 1. Lending (score: 0.2906)
# 2. Control (score: 0.1233)
# 3. ment (score: 0.0519)
# 4. Credit (score: 0.0454)
# 5. Disabilities (score: 0.0451)

Note how this model strongly outperforms the more general model on contract terminology, with much higher confidence in its predictions for terms like "WARRANTIES" and "Date".

For feature extraction and document similarity analysis, the pipeline API is also recommended:

from transformers import pipeline
import numpy as np
from sklearn.metrics.pairwise import cosine_similarity

# Load the feature-extraction pipeline
extractor = pipeline('feature-extraction', model="alea-institute/kl3m-doc-pico-contracts-001", return_tensors=True)

# Example legal documents
texts = [
    # Court Complaint
    "<|cls|> IN THE UNITED STATES DISTRICT COURT FOR THE EASTERN DISTRICT OF PENNSYLVANIA\n\nJOHN DOE,\nPlaintiff,\n\nvs.\n\nACME CORPORATION,\nDefendant.\n\nCIVIL ACTION NO. 21-12345\n\nCOMPLAINT\n\nPlaintiff John Doe, by and through his undersigned counsel, hereby files this Complaint against Defendant Acme Corporation, and in support thereof, alleges as follows: <|sep|>",
    
    # Consumer Terms
    "<|cls|> TERMS AND CONDITIONS\n\nLast Updated: April 10, 2025\n\nThese Terms and Conditions (\"Terms\") govern your access to and use of the Service. By accessing or using the Service, you agree to be bound by these Terms. If you do not agree to these Terms, you may not access or use the Service. These Terms constitute a legally binding agreement between you and the Company. <|sep|>",
    
    # Credit Agreement
    "<|cls|> CREDIT AGREEMENT\n\nDated as of April 10, 2025\n\nAmong\n\nACME BORROWER INC.,\nas the Borrower,\n\nBANK OF FINANCE,\nas Administrative Agent,\n\nand\n\nTHE LENDERS PARTY HERETO\n\nThis CREDIT AGREEMENT (\"Agreement\") is entered into as of April 10, 2025, among ACME BORROWER INC., a Delaware corporation (the \"Borrower\"), each lender from time to time party hereto (collectively, the \"Lenders\"), and BANK OF FINANCE, as Administrative Agent. <|sep|>"
]

# Strategy 1: CLS token embeddings
cls_embeddings = []
for text in texts:
    features = extractor(text)
    # Get the CLS token (first token) embedding
    features_array = features[0].numpy() if hasattr(features[0], 'numpy') else features[0]
    cls_embedding = features_array[0]
    cls_embeddings.append(cls_embedding)

# Calculate similarity between documents using CLS tokens
cls_similarity = cosine_similarity(np.vstack(cls_embeddings))
print("\nDocument similarity (CLS token):")
print(np.round(cls_similarity, 3))
# Output:
# [[1.    0.660 0.626]
#  [0.660 1.    0.798]
#  [0.626 0.798 1.   ]]

# Strategy 2: Mean pooling
mean_embeddings = []
for text in texts:
    features = extractor(text)
    # Average over all tokens
    features_array = features[0].numpy() if hasattr(features[0], 'numpy') else features[0]
    mean_embedding = np.mean(features_array, axis=0)
    mean_embeddings.append(mean_embedding)

# Calculate similarity using mean pooling
mean_similarity = cosine_similarity(np.vstack(mean_embeddings))
print("\nDocument similarity (Mean pooling):")
print(np.round(mean_similarity, 3))
# Output:
# [[1.    0.733 0.825]
#  [0.733 1.    0.678]
#  [0.825 0.678 1.   ]]

Training

The model was initially trained on a diverse corpus of legal and financial documents as the kl3m-doc-pico-001 model, then further fine-tuned for an additional 500,000 steps specifically on contract data. This specialized training enables the model to better understand contract structures, terminology, and semantics.

It leverages the KL3M tokenizer which provides 9-17% more efficient tokenization for domain-specific content than general-purpose tokenizers.

Special Tokens

This model includes the following special tokens:

  • CLS token: <|cls|> (ID: 5) - Used for the beginning of input text
  • MASK token: <|mask|> (ID: 6) - Used to mark tokens for prediction
  • SEP token: <|sep|> (ID: 4) - Used for the end of input text
  • PAD token: <|pad|> (ID: 2) - Used for padding sequences to a uniform length
  • BOS token: <|start|> (ID: 0) - Beginning of sequence
  • EOS token: <|end|> (ID: 1) - End of sequence
  • UNK token: <|unk|> (ID: 3) - Unknown token

Important usage notes:

When using the MASK token for predictions, be aware that this model uses a space-prefixed BPE tokenizer. The <|mask|> token should be placed IMMEDIATELY after the previous token with NO space, because most tokens in this tokenizer have an initial space encoded within them. For example: "word<|mask|>" rather than "word <|mask|>".

This space-aware placement is crucial for getting accurate predictions, as demonstrated in our test examples.

Contract-Specific Capabilities

The model shows particularly strong performance in identifying contract-specific terms and their relationships. For example, when given a masked token in the context of different types of agreements, the model correctly predicts relevant contract-specific terms:

  • In credit agreements: "Revolving," "Credit," "Loan," "Term"
  • In general agreements: "Confidentiality," "Purchase," "Subscription"
  • In supplier contexts: "Purchase," "License," "Services," "Supply"

This demonstrates the model's specialized knowledge of contract structures and terminology following the additional training steps on contract data.

Standard Test Examples

Using our standardized test examples for comparing embedding models:

Fill-Mask Results

  1. Contract Clause Heading:
    "<|cls|> 8. REPRESENTATIONS AND<|mask|>. Each party hereby represents and warrants to the other party as of the date hereof as follows: <|sep|>"

    Top 5 predictions:

    1. WARRANTIES (0.7282)
    2. Warranties (0.1674)
    3. REPRESENTATIONS (0.0377)
    4. warrants (0.0127)
    5. WARR (0.0074)

    Note: This model shows extremely strong performance on contract-specific language compared to the more general kl3m-doc-pico-001 model, correctly predicting "WARRANTIES" with high confidence.

  2. Defined Term Example:
    "<|cls|> \"Effective<|mask|>\" means the date on which all conditions precedent set forth in Article V are satisfied or waived by the Administrative Agent. <|sep|>"

    Top 5 predictions:

    1. Date (0.9735)
    2. Time (0.0075)
    3. date (0.0053)
    4. Dates (0.0024)
    5. Agreement (0.0011)

  3. Regulation Example:
    "<|cls|> All transactions shall comply with the requirements set forth in the Truth in<|mask|> Act and its implementing Regulation Z. <|sep|>"

    Top 5 predictions:

    1. Lending (0.2906)
    2. Control (0.1233)
    3. ment (0.0519)
    4. Credit (0.0454)
    5. Disabilities (0.0451)

Document Similarity Results

Using the standardized document examples for embeddings:

Document Pair Cosine Similarity (CLS token) Cosine Similarity (Mean pooling)
Court Complaint vs. Consumer Terms 0.660 0.733
Court Complaint vs. Credit Agreement 0.626 0.825
Consumer Terms vs. Credit Agreement 0.798 0.678

The contract-specialized model shows balanced similarity measurements that effectively capture document relationships. With CLS token embeddings, it identifies greatest similarity between Consumer Terms and Credit Agreement (0.798), while with mean pooling it shows highest similarity between Court Complaint and Credit Agreement (0.825).

Limitations

While compact and specialized for contracts, this model has some limitations:

  • Limited world knowledge compared to billion-parameter models
  • Primarily focused on English legal and financial contracts
  • Best suited for contract-specific rather than general-purpose tasks
  • May not perform as well on highly specialized sub-domains outside its training data
  • Requires domain expertise to interpret results effectively

References

Citation

If you use this model in your research, please cite:

@misc{bommarito2025kl3m,
  title={KL3M Tokenizers: A Family of Domain-Specific and Character-Level Tokenizers for Legal, Financial, and Preprocessing Applications},
  author={Bommarito II, Michael J. and Katz, Daniel Martin and Bommarito, Jillian},
  year={2025},
  eprint={2503.17247},
  archivePrefix={arXiv},
  primaryClass={cs.CL}
}

@misc{bommarito2025kl3mdata,
  title={The KL3M Data Project: Copyright-Clean Training Resources for Large Language Models},
  author={Bommarito II, Michael J. and Bommarito, Jillian and Katz, Daniel Martin},
  year={2025},
  eprint={2504.07854},
  archivePrefix={arXiv},
  primaryClass={cs.CL}
}

License

This model is licensed under CC-BY 4.0.

Contact

The KL3M model family is maintained by the ALEA Institute. For technical support, collaboration opportunities, or general inquiries:

https://aleainstitute.ai