jesse-adanac/bge-base-financial-matryoshka

BGE base Financial Matryoshka

This is a sentence-transformers model finetuned from BAAI/bge-base-en-v1.5 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: BAAI/bge-base-en-v1.5
• Maximum Sequence Length: 512 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': 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("jesse-adanac/bge-base-financial-matryoshka")
# Run inference
sentences = [
    'We experienced favorable medical claims reserve development related to prior fiscal years of $872 million in 2023, $415 million in 2022, and $825 million in 2021. The favorable development recognized in 2023 and 2021 primarily resulted from trend factors developing more favorably than originally expected as well as for 2021 completion factors developing faster than expected. The favorable development recognized in 2022 resulted primarily from completion factors remaining largely unchanged, resulting in lower overall development as compared to 2023 and 2021.',
    'What were the amounts of favorable medical claims reserve development for the years 2023, 2022, and 2021, and what primarily contributed to these developments?',
    'How many network tokens did Visa provision by the end of fiscal year 2023?',
]
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_512, dim_256, dim_128 and dim_64
• Evaluated with InformationRetrievalEvaluator

Metric	dim_768	dim_512	dim_256	dim_128	dim_64
cosine_accuracy@1	0.7343	0.7286	0.7286	0.71	0.6671
cosine_accuracy@3	0.8657	0.8657	0.8643	0.8543	0.8214
cosine_accuracy@5	0.89	0.8914	0.89	0.8757	0.8543
cosine_accuracy@10	0.9343	0.9343	0.9257	0.9157	0.8929
cosine_precision@1	0.7343	0.7286	0.7286	0.71	0.6671
cosine_precision@3	0.2886	0.2886	0.2881	0.2848	0.2738
cosine_precision@5	0.178	0.1783	0.178	0.1751	0.1709
cosine_precision@10	0.0934	0.0934	0.0926	0.0916	0.0893
cosine_recall@1	0.7343	0.7286	0.7286	0.71	0.6671
cosine_recall@3	0.8657	0.8657	0.8643	0.8543	0.8214
cosine_recall@5	0.89	0.8914	0.89	0.8757	0.8543
cosine_recall@10	0.9343	0.9343	0.9257	0.9157	0.8929
cosine_ndcg@10	0.8386	0.8363	0.8327	0.8181	0.7857
cosine_mrr@10	0.8076	0.8046	0.8024	0.7864	0.7508
cosine_map@100	0.8098	0.8067	0.8051	0.7896	0.7544

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: 10 tokens mean: 46.06 tokens max: 289 tokens	min: 2 tokens mean: 20.52 tokens max: 43 tokens

• Samples:

  positive	anchor
  Nonperforming loans and leases are generally those that have been placed on nonaccrual status, such as when they are 90 days past due or have confirmed cases of fraud or bankruptcy. Additionally, specific types of loans like consumer real estate-secured loans are classified as nonperforming at 90 days past due unless they are fully insured, and commercial loans and leases are classified as nonperforming when past due 90 days or more unless well-secured and in the process of collection.	What criteria are used to classify loans and leases as nonperforming according to the described credit policy?
  Changes in foreign exchange rates impacted cash and cash equivalents positively by $15 and $46 in 2023 and 2021, and negatively by $249 in 2022.	How has the change in foreign exchange rates affected cash and cash equivalents in 2023 and 2021?
  ITEM 8: FINANCIAL STATEMENTS AND SUPPLEMENTARY DATA	What is Item 8 about in the context of an annual report on Form 10-K?

• 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: 16
• gradient_accumulation_steps: 8
• learning_rate: 2e-05
• num_train_epochs: 4
• lr_scheduler_type: cosine
• warmup_ratio: 0.1
• bf16: True
• tf32: False
• 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: 8
• per_device_eval_batch_size: 16
• per_gpu_train_batch_size: None
• per_gpu_eval_batch_size: None
• gradient_accumulation_steps: 8
• 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: False
• 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_512_cosine_ndcg@10	dim_256_cosine_ndcg@10	dim_128_cosine_ndcg@10	dim_64_cosine_ndcg@10
0.1015	10	6.3316	-	-	-	-	-
0.2030	20	4.4603	-	-	-	-	-
0.3046	30	3.6545	-	-	-	-	-
0.4061	40	2.1196	-	-	-	-	-
0.5076	50	1.9986	-	-	-	-	-
0.6091	60	2.0175	-	-	-	-	-
0.7107	70	1.5044	-	-	-	-	-
0.8122	80	1.5722	-	-	-	-	-
0.9137	90	0.7737	-	-	-	-	-
1.0	99	-	0.8277	0.8278	0.8255	0.8086	0.7791
1.0102	100	1.3297	-	-	-	-	-
1.1117	110	1.2026	-	-	-	-	-
1.2132	120	1.1166	-	-	-	-	-
1.3147	130	0.963	-	-	-	-	-
1.4162	140	0.9185	-	-	-	-	-
1.5178	150	0.7528	-	-	-	-	-
1.6193	160	0.8351	-	-	-	-	-
1.7208	170	1.116	-	-	-	-	-
1.8223	180	0.5654	-	-	-	-	-
1.9239	190	0.6193	-	-	-	-	-
2.0	198	-	0.8342	0.8350	0.8310	0.8113	0.7805
2.0203	200	0.6482	-	-	-	-	-
2.1218	210	0.6604	-	-	-	-	-
2.2234	220	0.4969	-	-	-	-	-
2.3249	230	0.4502	-	-	-	-	-
2.4264	240	0.8084	-	-	-	-	-
2.5279	250	0.4882	-	-	-	-	-
2.6294	260	0.3821	-	-	-	-	-
2.7310	270	0.4308	-	-	-	-	-
2.8325	280	0.8484	-	-	-	-	-
2.9340	290	0.4867	-	-	-	-	-
3.0	297	-	0.8367	0.8359	0.8313	0.8166	0.7842
3.0305	300	0.807	-	-	-	-	-
3.1320	310	0.6478	-	-	-	-	-
3.2335	320	0.5532	-	-	-	-	-
3.3350	330	0.4459	-	-	-	-	-
3.4365	340	0.6112	-	-	-	-	-
3.5381	350	0.7304	-	-	-	-	-
3.6396	360	0.9029	-	-	-	-	-
3.7411	370	0.3999	-	-	-	-	-
3.8426	380	0.7569	-	-	-	-	-
3.9442	390	0.9483	-	-	-	-	-
3.9645	392	-	0.8386	0.8363	0.8327	0.8181	0.7857

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.11.12
• Sentence Transformers: 3.4.1
• Transformers: 4.49.0
• PyTorch: 2.7.0+cu126
• Accelerate: 1.6.0
• Datasets: 2.19.1
• Tokenizers: 0.21.1

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}
}