model
stringclasses
5 values
query_prefix
stringclasses
1 value
passage_prefix
stringclasses
1 value
embedding_size
int64
1.02k
1.54k
revision
stringclasses
5 values
model_type
stringclasses
2 values
torch_dtype
stringclasses
1 value
max_length
int64
512
32k
intfloat/multilingual-e5-large
query:
passage:
1,024
ab10c1a
xlm-roberta
float32
512
Snowflake/snowflake-arctic-embed-l
null
null
1,024
ecaabe9
xlm-roberta
float32
512
Alibaba-NLP/gte-Qwen2-1.5B-instruct
null
null
1,536
5652710
qwen2
float32
32,000
BAAI/bge-m3
null
null
1,024
5617a9f
xlm-roberta
float32
8,192
FacebookAI/xlm-roberta-large
null
null
1,024
c23d21b
xlm-roberta
float32
512

Reference models for integration into HF for Legal 🤗

This dataset comprises a collection of models aimed at streamlining and partially automating the embedding process. Each model entry within this dataset includes essential information such as model identifiers, embedding configurations, and specific parameters, ensuring that users can seamlessly integrate these models into their workflows with minimal setup and maximum efficiency.

Dataset Structure

Field Type Description
model str The identifier of the model, typically formatted as organization/model-name.
query_prefix str A prefix string added to query inputs to delineate them.
passage_prefix str A prefix string added to passage inputs to delineate them.
embedding_size int The dimensional size of the embedding vectors produced by the model.
revision str The specific revision identifier of the model to ensure consistency.
model_type str The architectural type of the model, such as xlm-roberta or qwen2.
torch_dtype str The data type utilized in PyTorch operations, such as float32.
max_length int The maximum input length the model can process, specified in tokens.

Organization architecture

In order to simplify the deployment of the organization's various tools, we propose a simple architecture in which datasets containing the various legal and contractual texts are doubled by datasets containing embeddings for different models, to enable simplified index creation for Spaces initialization and the provision of vector data for the GPU-poor. A simplified representation might look like this:

Citing & Authors

If you use this dataset in your research, please use the following BibTeX entry.

@misc{HFforLegal2024,
  author =       {Louis Brulé Naudet},
  title =        {Reference models for integration into HF for Legal},
  year =         {2024}
  howpublished = {\url{https://huggingface.co/datasets/HFforLegal/embedding-models}},
}

Feedback

If you have any feedback, please reach out at louisbrulenaudet@icloud.com.

Downloads last month
62