BGE base Financial Matryoshka

This is a sentence-transformers model finetuned from BAAI/bge-base-en-v1.5. It maps sentences & paragraphs to a 768-dimensional dense vector space and can be used for semantic textual similarity, semantic search, paraphrase mining, text classification, clustering, and more.

Model Details

Model Description

• Model Type: Sentence Transformer
• Base model: BAAI/bge-base-en-v1.5
• Maximum Sequence Length: 512 tokens
• Output Dimensionality: 768 tokens
• Similarity Function: Cosine Similarity

• Language: en
• License: apache-2.0

Model Sources

• Documentation: Sentence Transformers Documentation
• Repository: Sentence Transformers on GitHub
• Hugging Face: Sentence Transformers on Hugging Face

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'max_seq_length': 512, 'do_lower_case': True}) with Transformer model: BertModel 
  (1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': True, 'pooling_mode_mean_tokens': False, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False, 'include_prompt': True})
  (2): Normalize()
)

Usage

Direct Usage (Sentence Transformers)

First install the Sentence Transformers library:

pip install -U sentence-transformers

Then you can load this model and run inference.

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("daniellgoncalves/bge-base-horizon-europe-matryoshka")
# Run inference
sentences = [
    'including  data. They are globally unique and long -lasting references to digital objects \r\n(such as data, publications and other research outputs)  or non -digital objects such as \r\nresearchers, research institutions, grants, etc. Frequently used persistent identifiers \r\ninclude digital object identifiers (DOIs), Handles, and others.  For further reading on PID \r\ntypes, please refer to https://www.dpconline.org/handbook/technical -solutions -and-\r\ntools/persistent -identifiers . \r\nTo enhance the findability of r esearch outputs, and their potential reuse, standardised \r\nmetadata frameworks are essential, ensuring that data and other research outputs \r\nare accompanied by rich metadata that provides them with context.  \r\nTo enhance the re -usability of research data, they must be licenced. For more \r\ninformation on the licences required for data under Horizon Europe, please refer to the \r\nAGA (article 17).  \r\nTrusted repositories  assume a central role in the Horizon Europe for the deposition of \r\nand access to publications and resea rch data. For a definition of trusted repositories in \r\nHorizon Europe please refer to the AGA (article 17). Proposers, with the help of data \r\nand research support staff (e.g. data stewards, data librarians, et c), should check \r\nwhether the repositories that th ey plan to deposit their data have the features of \r\ntrusted repositories, and justify this accordingly in their Data Management Plans.  \r\nData management plans (DMPs) are a cornerstone for responsible management of \r\nresearch outputs, notably data and are mandatory in Horizon Europe for projects \r\ngenerating  and/or reusing data (on requirements and the frequency of DMPs as \r\ndeliverables consult the AGA article 17). A template for a DMP  is provided under the \r\nreporting templates in the reference documents  of the Funding and Tenders portal of \r\nthe European Commission. Its use is recom mended but not mandatory.  DMPs are',
    'What is a cornerstone for responsible management of research outputs in Horizon Europe?\r\n',
    'How can a Beneficiary submit a Financial Statement?\r\n',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 768]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities.shape)
# [3, 3]

Training Details

Training Dataset

Unnamed Dataset

• Size: 45 training samples
• Columns: positive and anchor
• Approximate statistics based on the first 1000 samples:

  	positive	anchor
  type	string	string
  details	min: 54 tokensmean: 309.0 tokensmax: 512 tokens	min: 8 tokensmean: 18.09 tokensmax: 34 tokens

• Samples:

  positive	anchor
  and rules and procedures for sharing of data including licensing.
  Research outputs — Results to which access can be given in the form of scientific publications, data or other
  engineered results and processes such as software, algorithms, protocols, models, workflows
  and electronic notebooks.
  [Scope of the obligations
  For this section, references to ‘beneficiary’or ‘beneficiaries’do not include affiliated entities (if any). ]
  [Agreement on background — Background free from restrictions
  The beneficiaries must identify in a written agreement the background as needed for implementing the action
  or for exploiting its results.
  Where the call conditions restrict control due to strategic interests reasons, background that is subject to
  control or other restrictions by a country (or entity from a country) which is not one of the eligible countries
  or target countries set out in the call conditions and that impact the exploitation of the results (i.e. would
  make the exploitation of the results subject to control or restrictions) must not be used and must be explicitly
  excluded in the agreement on background —unless otherwise agreed with the granting authority. ]
  [Results free from restrictions
  Where the call conditions restrict control due to [strategic interests reasons ], the beneficiaries must ensure
  that the results of the action are not subject to control or other restrictions by a country (or entity from a
  country) which is not one of the eligible countries or target countries set out in the call conditions — unless
  otherwise agreed with the granting authority. ]
  Ownership of results
  Results are owned by the beneficiaries that generate them. [[…]]	Who owns the results of the action in the given context?
  EU Grants: HE Programme Guide : V4.0 – 15.10.2023
  60
  To a degree matching the type of research being proposed (from basic to
  precompetitiv e) and as appropriate, AI -based systems or techniques should be, or be
  developed to become (implicitly or explicitly contributing to one or several of the
  following objectives):
  • technically robust, accurate and reproducible, and able to deal with and infor m
  about possible failures, inaccuracies and errors, proportionate to the assessed
  risk posed by the AI -based system or technique
  • socially robust, in that they duly consider the context and environment in which
  they operate
  • reliable and to function as inte nded, minimising unintentional and unexpected
  harm, preventing unacceptable harm and safeguarding the physical and mental
  integrity of humans
  • able to provide a suitable explanation of its decision -making process, whenever
  an AI -based system can have a sign ificant impact on people’s lives.	What should AI-based systems be able to provide in terms of decision-making process?
  EU Funding & Tenders Portal: Online Manual: V1.1 – 15.09.2022
  15 • 1 Participant Contact (PaCo) per beneficiar y
  • 1 LEAR per organisation
  • 1 Project Legal Signatory (PLSIGN) per organisation
  • 1 Project Financial Signatory (PFSIGN) per organi sation .
  One person can have several roles at the same time.
  Organisations that participate as Affiliated Entities (or other type of participant — Associated
  Partner, Subcontractor, etc) do NOT need any access roles in the Portal, since they are not
  allowed to use it. All actions in the Portal are handled for them by the Coordinator /Beneficiary
  they are linked to .

• Loss: MatryoshkaLoss with these parameters:

  {
      "loss": "MultipleNegativesRankingLoss",
      "matryoshka_dims": [
          768,
          512,
          256,
          128,
          64
      ],
      "matryoshka_weights": [
          1,
          1,
          1,
          1,
          1
      ],
      "n_dims_per_step": -1
  }
  

Training Hyperparameters

All Hyperparameters

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: no
• prediction_loss_only: True
• per_device_train_batch_size: 8
• per_device_eval_batch_size: 8
• per_gpu_train_batch_size: None
• per_gpu_eval_batch_size: None
• gradient_accumulation_steps: 1
• eval_accumulation_steps: None
• learning_rate: 5e-05
• weight_decay: 0.0
• adam_beta1: 0.9
• adam_beta2: 0.999
• adam_epsilon: 1e-08
• max_grad_norm: 1.0
• num_train_epochs: 3.0
• max_steps: -1
• lr_scheduler_type: linear
• lr_scheduler_kwargs: {}
• warmup_ratio: 0.0
• warmup_steps: 0
• log_level: passive
• log_level_replica: warning
• log_on_each_node: True
• logging_nan_inf_filter: True
• save_safetensors: True
• save_on_each_node: False
• save_only_model: False
• restore_callback_states_from_checkpoint: False
• no_cuda: False
• use_cpu: False
• use_mps_device: False
• seed: 42
• data_seed: None
• jit_mode_eval: False
• use_ipex: False
• bf16: False
• fp16: False
• fp16_opt_level: O1
• half_precision_backend: auto
• bf16_full_eval: False
• fp16_full_eval: False
• tf32: None
• local_rank: 0
• ddp_backend: None
• tpu_num_cores: None
• tpu_metrics_debug: False
• debug: []
• dataloader_drop_last: False
• dataloader_num_workers: 0
• dataloader_prefetch_factor: None
• past_index: -1
• disable_tqdm: False
• remove_unused_columns: True
• label_names: None
• load_best_model_at_end: False
• ignore_data_skip: False
• fsdp: []
• fsdp_min_num_params: 0
• fsdp_config: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}
• fsdp_transformer_layer_cls_to_wrap: None
• accelerator_config: {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}
• deepspeed: None
• label_smoothing_factor: 0.0
• optim: adamw_torch
• optim_args: None
• adafactor: False
• group_by_length: False
• length_column_name: length
• ddp_find_unused_parameters: None
• ddp_bucket_cap_mb: None
• ddp_broadcast_buffers: False
• dataloader_pin_memory: True
• dataloader_persistent_workers: False
• skip_memory_metrics: True
• use_legacy_prediction_loop: False
• push_to_hub: False
• resume_from_checkpoint: None
• hub_model_id: None
• hub_strategy: every_save
• hub_private_repo: False
• hub_always_push: False
• gradient_checkpointing: False
• gradient_checkpointing_kwargs: None
• include_inputs_for_metrics: False
• eval_do_concat_batches: True
• fp16_backend: auto
• push_to_hub_model_id: None
• push_to_hub_organization: None
• mp_parameters:
• auto_find_batch_size: False
• full_determinism: False
• torchdynamo: None
• ray_scope: last
• ddp_timeout: 1800
• torch_compile: False
• torch_compile_backend: None
• torch_compile_mode: None
• dispatch_batches: None
• split_batches: None
• include_tokens_per_second: False
• include_num_input_tokens_seen: False
• neftune_noise_alpha: None
• optim_target_modules: None
• batch_eval_metrics: False
• batch_sampler: batch_sampler
• multi_dataset_batch_sampler: proportional

Framework Versions

• Python: 3.10.12
• Sentence Transformers: 3.0.1
• Transformers: 4.41.2
• PyTorch: 2.1.2+cu121
• Accelerate: 0.32.1
• Datasets: 2.19.1
• Tokenizers: 0.19.1

Citation

BibTeX

Sentence Transformers

@inproceedings{reimers-2019-sentence-bert,
    title = "Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks",
    author = "Reimers, Nils and Gurevych, Iryna",
    booktitle = "Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing",
    month = "11",
    year = "2019",
    publisher = "Association for Computational Linguistics",
    url = "https://arxiv.org/abs/1908.10084",
}

MatryoshkaLoss

@misc{kusupati2024matryoshka,
    title={Matryoshka Representation Learning}, 
    author={Aditya Kusupati and Gantavya Bhatt and Aniket Rege and Matthew Wallingford and Aditya Sinha and Vivek Ramanujan and William Howard-Snyder and Kaifeng Chen and Sham Kakade and Prateek Jain and Ali Farhadi},
    year={2024},
    eprint={2205.13147},
    archivePrefix={arXiv},
    primaryClass={cs.LG}
}

MultipleNegativesRankingLoss

@misc{henderson2017efficient,
    title={Efficient Natural Language Response Suggestion for Smart Reply}, 
    author={Matthew Henderson and Rami Al-Rfou and Brian Strope and Yun-hsuan Sung and Laszlo Lukacs and Ruiqi Guo and Sanjiv Kumar and Balint Miklos and Ray Kurzweil},
    year={2017},
    eprint={1705.00652},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}