rya23/modernbert-embed-finance-matryoshka

ModernBERT Embed base Finance 10k 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

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, 'architecture': '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("rya23/modernbert-embed-finance-matryoshka")
# Run inference
sentences = [
    "What typical reimbursement methods are used in the company's contracts with hospitals for inpatient and outpatient services?",
    'We typically contract with hospitals on either (1) a per diem rate, which is an all-inclusive rate per day, (2) a case rate for diagnosis-related groups (DRG), which is an all-inclusive rate per admission, or (3) a discounted charge for inpatient hospital services. Outpatient hospital services generally are contracted at a flat rate by type of service, ambulatory payment classifications, or APCs, or at a discounted charge.',
    'In IBM’s 2023 Annual Report to Stockholders, the Financial Statements and Supplementary Data are detailed on pages 44 through 121.',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 768]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities)
# tensor([[1.0000, 0.6756, 0.0659],
#         [0.6756, 1.0000, 0.0087],
#         [0.0659, 0.0087, 1.0000]])

Evaluation

Metrics

Information Retrieval

• Dataset: dim_768
• Evaluated with InformationRetrievalEvaluator with these parameters:

  {
      "truncate_dim": 768
  }
  

Metric	Value
cosine_accuracy@1	0.7244
cosine_accuracy@3	0.8554
cosine_accuracy@5	0.8903
cosine_accuracy@10	0.9271
cosine_precision@1	0.7244
cosine_precision@3	0.2851
cosine_precision@5	0.1781
cosine_precision@10	0.0927
cosine_recall@1	0.7244
cosine_recall@3	0.8554
cosine_recall@5	0.8903
cosine_recall@10	0.9271
cosine_ndcg@10	0.8286
cosine_mrr@10	0.7968
cosine_map@100	0.7999

Information Retrieval

• Dataset: dim_512
• Evaluated with InformationRetrievalEvaluator with these parameters:

  {
      "truncate_dim": 512
  }
  

Metric	Value
cosine_accuracy@1	0.7239
cosine_accuracy@3	0.8533
cosine_accuracy@5	0.8874
cosine_accuracy@10	0.927
cosine_precision@1	0.7239
cosine_precision@3	0.2844
cosine_precision@5	0.1775
cosine_precision@10	0.0927
cosine_recall@1	0.7239
cosine_recall@3	0.8533
cosine_recall@5	0.8874
cosine_recall@10	0.927
cosine_ndcg@10	0.8273
cosine_mrr@10	0.7952
cosine_map@100	0.7983

Information Retrieval

• Dataset: dim_256
• Evaluated with InformationRetrievalEvaluator with these parameters:

  {
      "truncate_dim": 256
  }
  

Metric	Value
cosine_accuracy@1	0.7231
cosine_accuracy@3	0.8513
cosine_accuracy@5	0.886
cosine_accuracy@10	0.924
cosine_precision@1	0.7231
cosine_precision@3	0.2838
cosine_precision@5	0.1772
cosine_precision@10	0.0924
cosine_recall@1	0.7231
cosine_recall@3	0.8513
cosine_recall@5	0.886
cosine_recall@10	0.924
cosine_ndcg@10	0.8256
cosine_mrr@10	0.7938
cosine_map@100	0.7971

Information Retrieval

• Dataset: dim_128
• Evaluated with InformationRetrievalEvaluator with these parameters:

  {
      "truncate_dim": 128
  }
  

Metric	Value
cosine_accuracy@1	0.7017
cosine_accuracy@3	0.8363
cosine_accuracy@5	0.8726
cosine_accuracy@10	0.9166
cosine_precision@1	0.7017
cosine_precision@3	0.2788
cosine_precision@5	0.1745
cosine_precision@10	0.0917
cosine_recall@1	0.7017
cosine_recall@3	0.8363
cosine_recall@5	0.8726
cosine_recall@10	0.9166
cosine_ndcg@10	0.8108
cosine_mrr@10	0.7767
cosine_map@100	0.7803

Information Retrieval

• Dataset: dim_64
• Evaluated with InformationRetrievalEvaluator with these parameters:

  {
      "truncate_dim": 64
  }
  

Metric	Value
cosine_accuracy@1	0.6703
cosine_accuracy@3	0.8053
cosine_accuracy@5	0.8491
cosine_accuracy@10	0.8959
cosine_precision@1	0.6703
cosine_precision@3	0.2684
cosine_precision@5	0.1698
cosine_precision@10	0.0896
cosine_recall@1	0.6703
cosine_recall@3	0.8053
cosine_recall@5	0.8491
cosine_recall@10	0.8959
cosine_ndcg@10	0.7834
cosine_mrr@10	0.7474
cosine_map@100	0.7516

Training Details

Training Dataset

json

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

  	anchor	positive
  type	string	string
  details	min: 9 tokens mean: 20.55 tokens max: 43 tokens	min: 5 tokens mean: 46.3 tokens max: 243 tokens

• Samples:

  anchor	positive
  How many shares of class A common stock were authorized for grant under Visa's Equity Incentive Compensation Plan?	Under the Company’s 2007 Amended and Restated Equity Incentive Compensation Plan (EIP), the compensation committee of the board of directors was authorized to grant up to 198 million shares of class A common stock to its employees and non-employee directors.
  What was Garmin Ltd.'s net income for the fiscal year ended December 30, 2023?	Garmin Ltd. reported a net income of $1,289,636 for the fiscal year ended December 30, 2023.
  Why are some device sales revenue at AT&T not immediately recognized upon the device sale?	AT&T recognizes revenue from device sales with promotions or installment payments differently. For promotional discounts, revenue is deferred and amortized over the contract term. Meanwhile, installment sales involve recognizing revenue upfront but deferring the cash receipt until payments are made, resulting in a recorded contract asset to be amortized over time.

• 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

Non-Default Hyperparameters

• eval_strategy: epoch
• per_device_eval_batch_size: 4
• gradient_accumulation_steps: 48
• learning_rate: 2e-05
• num_train_epochs: 4
• lr_scheduler_type: cosine
• warmup_ratio: 0.1
• warmup_steps: 0.1
• fp16: True
• load_best_model_at_end: True
• batch_sampler: no_duplicates

All Hyperparameters

Click to expand

• do_predict: False
• eval_strategy: epoch
• prediction_loss_only: True
• per_device_train_batch_size: 8
• per_device_eval_batch_size: 4
• gradient_accumulation_steps: 48
• 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: None
• warmup_ratio: 0.1
• warmup_steps: 0.1
• log_level: passive
• log_level_replica: warning
• log_on_each_node: True
• logging_nan_inf_filter: True
• enable_jit_checkpoint: False
• save_on_each_node: False
• save_only_model: False
• restore_callback_states_from_checkpoint: False
• use_cpu: False
• seed: 42
• data_seed: None
• bf16: False
• fp16: True
• bf16_full_eval: False
• fp16_full_eval: False
• tf32: None
• local_rank: -1
• ddp_backend: None
• debug: []
• dataloader_drop_last: False
• dataloader_num_workers: 0
• dataloader_prefetch_factor: None
• disable_tqdm: False
• remove_unused_columns: True
• label_names: None
• load_best_model_at_end: True
• ignore_data_skip: False
• fsdp: []
• fsdp_config: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}
• accelerator_config: {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}
• parallelism_config: None
• deepspeed: None
• label_smoothing_factor: 0.0
• optim: adamw_torch_fused
• optim_args: None
• group_by_length: False
• length_column_name: length
• project: huggingface
• trackio_space_id: trackio
• 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
• push_to_hub: False
• resume_from_checkpoint: None
• hub_model_id: None
• hub_strategy: every_save
• hub_private_repo: None
• hub_always_push: False
• hub_revision: None
• gradient_checkpointing: False
• gradient_checkpointing_kwargs: None
• include_for_metrics: []
• eval_do_concat_batches: True
• auto_find_batch_size: False
• full_determinism: False
• ddp_timeout: 1800
• torch_compile: False
• torch_compile_backend: None
• torch_compile_mode: None
• include_num_input_tokens_seen: no
• neftune_noise_alpha: None
• optim_target_modules: None
• batch_eval_metrics: False
• eval_on_start: False
• use_liger_kernel: False
• liger_kernel_config: None
• eval_use_gather_object: False
• average_tokens_across_devices: True
• use_cache: False
• prompts: None
• batch_sampler: no_duplicates
• multi_dataset_batch_sampler: proportional
• router_mapping: {}
• learning_rate_mapping: {}

Training Logs

Epoch	Step	Training Loss	dim_768_cosine_ndcg@10	dim_512_cosine_ndcg@10	dim_256_cosine_ndcg@10	dim_128_cosine_ndcg@10	dim_64_cosine_ndcg@10
0.6091	10	0.3092	-	-	-	-	-
1.0	17	-	0.8155	0.8138	0.8104	0.7948	0.7647
1.1827	20	0.0958	-	-	-	-	-
1.7919	30	0.0675	-	-	-	-	-
2.0	34	-	0.8257	0.8245	0.8219	0.8045	0.7757
2.3655	40	0.0458	-	-	-	-	-
2.9746	50	0.0505	-	-	-	-	-
3.0	51	-	0.8277	0.8259	0.8243	0.8087	0.7819
3.5482	60	0.0593	-	-	-	-	-
4.0	68	-	0.8286	0.8273	0.8256	0.8108	0.7834

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.12.12
• Sentence Transformers: 5.2.2
• Transformers: 5.0.0
• PyTorch: 2.9.0+cu128
• Accelerate: 1.12.0
• Datasets: 4.0.0
• Tokenizers: 0.22.2

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}
}