README.md

metadata

language:
  - en
license: apache-2.0
tags:
  - sentence-transformers
  - sentence-similarity
  - feature-extraction
  - generated_from_trainer
  - dataset_size:6300
  - loss:MatryoshkaLoss
  - loss:MultipleNegativesRankingLoss
base_model: nomic-ai/modernbert-embed-base
widget:
  - source_sentence: >-
      HP reviews goodwill for impairment by initially performing a qualitative
      assessment to see if the fair value of a reporting unit is likely less
      than its carrying amount. If more likely, a quantitative assessment
      follows.
    sentences:
      - >-
        What percentage did the Communications segment account for of the 2023
        total segment income?
      - How does HP determine whether goodwill impairment exists?
      - >-
        What was the primary reason for the actuarial gain during the year ended
        December 31, 2022?
  - source_sentence: >-
      The consolidated financial statements and accompanying notes are listed in
      Part IV, Item 15(a)(1).
    sentences:
      - What does Item 8 in the Annual Report on Form 10-K detail?
      - >-
        In which part of the Annual Report on Form 10-K are the consolidated
        financial statements and accompanying notes listed?
      - What is the estimated redemption rate for Chipotle gift cards?
  - source_sentence: >-
      American Express maintains direct relationships with Card Members and
      merchants, which provides it with direct access to information at both
      ends of the transaction, distinguishing its integrated payments platform
      from the bankcard networks.
    sentences:
      - >-
        How does American Express's integrated payments platform differentiate
        itself from bankcard networks?
      - How are contingent consideration liabilities valued?
      - >-
        How does Chipotle calculate revenue recognition for redeemed Chipotle
        Rewards?
  - source_sentence: >-
      Open Value agreements are a simple, cost-effective way to acquire the
      latest Microsoft technology. These agreements are designed for small and
      medium organizations that want to license cloud services and on-premises
      software over a three-year period. Under Open Value agreements,
      organizations can elect to purchase perpetual licenses or subscribe to
      licenses and SA is included.
    sentences:
      - >-
        How are unpaid losses and loss expenses calculated in the financial
        statements of an insurance and reinsurance company?
      - >-
        What type of financial documents are included in Part IV, Item 15(a)(1)
        of the Annual Report on Form 10-K?
      - >-
        What type of organizations is the Open Value agreements designed for and
        what licenses does it include?
  - source_sentence: >-
      The company's financial report indicates that the pre-tax amounts of gains
      (losses) from foreign currency forward exchange contracts designated as
      cash flow hedges were gains of $82 million in 2021, gains of $103 million
      in 2022, and losses of $2 million in 2023.
    sentences:
      - >-
        What were the pre-tax amounts of (gains) losses from foreign currency
        forward exchange contracts designated as cash flow hedges for the years
        ended December 31 from 2021 to 2023?
      - >-
        What is the projected change in income before income taxes if the 2023
        discount rate for the U.S. defined benefit pension and retiree health
        benefit plans changes by a quarter percentage point?
      - >-
        What sources contribute to Ford Credit’s liquidity as of December 31,
        2023, and what was their total value?
pipeline_tag: sentence-similarity
library_name: sentence-transformers
metrics:
  - cosine_accuracy@1
  - cosine_accuracy@3
  - cosine_accuracy@5
  - cosine_accuracy@10
  - cosine_precision@1
  - cosine_precision@3
  - cosine_precision@5
  - cosine_precision@10
  - cosine_recall@1
  - cosine_recall@3
  - cosine_recall@5
  - cosine_recall@10
  - cosine_ndcg@10
  - cosine_mrr@10
  - cosine_map@100
model-index:
  - name: BGE base Financial Matryoshka
    results:
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: dim 768
          type: dim_768
        metrics:
          - type: cosine_accuracy@1
            value: 0.6914285714285714
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.8171428571428572
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.87
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.9128571428571428
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.6914285714285714
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.2723809523809524
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.174
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.09128571428571428
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.6914285714285714
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.8171428571428572
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.87
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.9128571428571428
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.8015002951126636
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.7659410430839002
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.76947397245476
            name: Cosine Map@100
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: dim 256
          type: dim_256
        metrics:
          - type: cosine_accuracy@1
            value: 0.6642857142857143
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.81
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.8557142857142858
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.8971428571428571
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.6642857142857143
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.27
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.17114285714285712
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.0897142857142857
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.6642857142857143
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.81
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.8557142857142858
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.8971428571428571
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.7834209531598721
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.7467698412698411
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.7514515853623652
            name: Cosine Map@100
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: dim 64
          type: dim_64
        metrics:
          - type: cosine_accuracy@1
            value: 0.62
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.7671428571428571
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.8171428571428572
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.8742857142857143
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.62
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.2557142857142857
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.1634285714285714
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.08742857142857142
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.62
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.7671428571428571
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.8171428571428572
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.8742857142857143
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.7453405840762105
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.7042613378684806
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.70911408987056
            name: Cosine Map@100

BGE base Financial Matryoshka

This is a sentence-transformers model finetuned from nomic-ai/modernbert-embed-base on the json dataset. 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: nomic-ai/modernbert-embed-base
• Maximum Sequence Length: 8192 tokens
• Output Dimensionality: 768 dimensions
• Similarity Function: Cosine Similarity
• Training Dataset:
  • json
• 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': 8192, 'do_lower_case': False}) with Transformer model: ModernBertModel 
  (1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, '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("Sorour/modernbert-financial-matryoshka")
# Run inference
sentences = [
    "The company's financial report indicates that the pre-tax amounts of gains (losses) from foreign currency forward exchange contracts designated as cash flow hedges were gains of $82 million in 2021, gains of $103 million in 2022, and losses of $2 million in 2023.",
    'What were the pre-tax amounts of (gains) losses from foreign currency forward exchange contracts designated as cash flow hedges for the years ended December 31 from 2021 to 2023?',
    'What is the projected change in income before income taxes if the 2023 discount rate for the U.S. defined benefit pension and retiree health benefit plans changes by a quarter percentage point?',
]
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]

Evaluation

Metrics

Information Retrieval

• Datasets: dim_768, dim_256 and dim_64
• Evaluated with InformationRetrievalEvaluator

Metric	dim_768	dim_256	dim_64
cosine_accuracy@1	0.6914	0.6643	0.62
cosine_accuracy@3	0.8171	0.81	0.7671
cosine_accuracy@5	0.87	0.8557	0.8171
cosine_accuracy@10	0.9129	0.8971	0.8743
cosine_precision@1	0.6914	0.6643	0.62
cosine_precision@3	0.2724	0.27	0.2557
cosine_precision@5	0.174	0.1711	0.1634
cosine_precision@10	0.0913	0.0897	0.0874
cosine_recall@1	0.6914	0.6643	0.62
cosine_recall@3	0.8171	0.81	0.7671
cosine_recall@5	0.87	0.8557	0.8171
cosine_recall@10	0.9129	0.8971	0.8743
cosine_ndcg@10	0.8015	0.7834	0.7453
cosine_mrr@10	0.7659	0.7468	0.7043
cosine_map@100	0.7695	0.7515	0.7091

Training Details

Training Dataset

json

• Dataset: json
• Size: 6,300 training samples
• Columns: positive and anchor
• Approximate statistics based on the first 1000 samples:

  	positive	anchor
  type	string	string
  details	min: 9 tokens mean: 47.08 tokens max: 998 tokens	min: 9 tokens mean: 20.19 tokens max: 41 tokens

• Samples:

  positive	anchor
  Item 8 includes Financial Statements and Supplementary Data.	What type of data is found in Item 8 of detailed financial documentation?
  HP records revenue from the sale of equipment under sales-type leases as revenue at the commencement of the lease. This method is applied unless certain conditions such as customer acceptance remain uncertain or significant obligations to the customer remain unfulfilled.	How does HP recognize revenue from the sale of equipment under sales-type leases?
  The company maintains insurance coverage for general liability, property, business interruption, terrorism, and other risks with respect to their business for all of their owned and leased hotels.	What types of risks are usually covered by the company's insurance policies?

• Loss: MatryoshkaLoss with these parameters:

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

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: epoch
• per_device_train_batch_size: 32
• per_device_eval_batch_size: 16
• gradient_accumulation_steps: 16
• learning_rate: 2e-05
• num_train_epochs: 4
• lr_scheduler_type: cosine
• warmup_ratio: 0.1
• bf16: True
• tf32: True
• load_best_model_at_end: True
• optim: adamw_torch_fused
• batch_sampler: no_duplicates

All Hyperparameters

Click to expand

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: epoch
• prediction_loss_only: True
• per_device_train_batch_size: 32
• per_device_eval_batch_size: 16
• per_gpu_train_batch_size: None
• per_gpu_eval_batch_size: None
• gradient_accumulation_steps: 16
• eval_accumulation_steps: None
• torch_empty_cache_steps: None
• learning_rate: 2e-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: 4
• max_steps: -1
• lr_scheduler_type: cosine
• lr_scheduler_kwargs: {}
• warmup_ratio: 0.1
• 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: True
• fp16: False
• fp16_opt_level: O1
• half_precision_backend: auto
• bf16_full_eval: False
• fp16_full_eval: False
• tf32: True
• 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: True
• 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_fused
• 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: None
• hub_always_push: False
• gradient_checkpointing: False
• gradient_checkpointing_kwargs: None
• include_inputs_for_metrics: False
• include_for_metrics: []
• 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
• eval_on_start: False
• use_liger_kernel: False
• eval_use_gather_object: False
• average_tokens_across_devices: False
• prompts: None
• batch_sampler: no_duplicates
• multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	Training Loss	dim_768_cosine_ndcg@10	dim_256_cosine_ndcg@10	dim_64_cosine_ndcg@10
0.8122	10	9.4544	-	-	-
1.0	13	-	0.7799	0.7650	0.7097
0.8122	10	3.1908	-	-	-
1.0	13	-	0.7952	0.7769	0.7259
1.5685	20	1.8807	-	-	-
2.0	26	-	0.8001	0.7833	0.7409
2.3249	30	1.7141	-	-	-
3.0	39	-	0.8023	0.7819	0.7460
3.0812	40	1.3672	-	-	-
3.731	48	-	0.8015	0.7834	0.7453

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.11.11
• Sentence Transformers: 3.4.1
• Transformers: 4.48.3
• PyTorch: 2.5.1+cu124
• Accelerate: 1.3.0
• Datasets: 3.3.2
• Tokenizers: 0.21.0

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}
}