MistyDragon/bge-small-financial-matryoshka

BGE base Financial Matryoshka

This is a sentence-transformers model finetuned from BAAI/bge-small-en-v1.5 on the json dataset. It maps sentences & paragraphs to a 384-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-small-en-v1.5
• Maximum Sequence Length: 512 tokens
• Output Dimensionality: 384 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': 384, '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("MistyDragon/bge-small-financial-matryoshka")
# Run inference
sentences = [
    'Caterpillar Insurance Co. Ltd. is registered as a Class 2 (General Business) and Class B (Long-Term) insurer with the Bermuda Monetary Authority.',
    'What types of insurance licenses does Caterpillar Insurance Co. Ltd. hold in Bermuda?',
    "What is indicated by 'Item 8' in a financial document?",
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 384]

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

Evaluation

Metrics

Information Retrieval

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

  {
      "truncate_dim": 256
  }
  

Metric	Value
cosine_accuracy@1	0.6986
cosine_accuracy@3	0.8314
cosine_accuracy@5	0.8729
cosine_accuracy@10	0.9171
cosine_precision@1	0.6986
cosine_precision@3	0.2771
cosine_precision@5	0.1746
cosine_precision@10	0.0917
cosine_recall@1	0.6986
cosine_recall@3	0.8314
cosine_recall@5	0.8729
cosine_recall@10	0.9171
cosine_ndcg@10	0.8091
cosine_mrr@10	0.7745
cosine_map@100	0.7781

Information Retrieval

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

  {
      "truncate_dim": 128
  }
  

Metric	Value
cosine_accuracy@1	0.6771
cosine_accuracy@3	0.8171
cosine_accuracy@5	0.8643
cosine_accuracy@10	0.9171
cosine_precision@1	0.6771
cosine_precision@3	0.2724
cosine_precision@5	0.1729
cosine_precision@10	0.0917
cosine_recall@1	0.6771
cosine_recall@3	0.8171
cosine_recall@5	0.8643
cosine_recall@10	0.9171
cosine_ndcg@10	0.7978
cosine_mrr@10	0.7596
cosine_map@100	0.7626

Information Retrieval

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

  {
      "truncate_dim": 64
  }
  

Metric	Value
cosine_accuracy@1	0.66
cosine_accuracy@3	0.8014
cosine_accuracy@5	0.8543
cosine_accuracy@10	0.9029
cosine_precision@1	0.66
cosine_precision@3	0.2671
cosine_precision@5	0.1709
cosine_precision@10	0.0903
cosine_recall@1	0.66
cosine_recall@3	0.8014
cosine_recall@5	0.8543
cosine_recall@10	0.9029
cosine_ndcg@10	0.7797
cosine_mrr@10	0.7405
cosine_map@100	0.7439

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: 47.77 tokens max: 439 tokens	min: 8 tokens mean: 20.48 tokens max: 45 tokens

• Samples:

  positive	anchor
  Return on investment (ROI)	12.7
  According to the terms of the Senior Credit Facilities, cash amounts exceeding $175 million can be deducted from the total debt in the leverage ratio calculation, though this is subject to certain restrictions.	How does the Senior Credit Facilities' treatment of cash affect the calculation of the leverage ratio?
  In 2023, approximately 67% of the total U.S. dialysis patient service revenues were generated from government-based programs.	What percentage of the total U.S. dialysis patient service revenues were generated from government-based programs in 2023?

• Loss: MatryoshkaLoss with these parameters:

  {
      "loss": "MultipleNegativesRankingLoss",
      "matryoshka_dims": [
          256,
          128,
          64
      ],
      "matryoshka_weights": [
          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
• 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_256_cosine_ndcg@10	dim_128_cosine_ndcg@10	dim_64_cosine_ndcg@10
0.1015	10	4.9287	-	-	-
0.2030	20	3.7753	-	-	-
0.3046	30	2.7807	-	-	-
0.4061	40	2.6642	-	-	-
0.5076	50	1.8158	-	-	-
0.6091	60	1.2895	-	-	-
0.7107	70	1.356	-	-	-
0.8122	80	1.2217	-	-	-
0.9137	90	1.2548	-	-	-
1.0	99	-	0.7949	0.7853	0.7609
1.0102	100	1.1693	-	-	-
1.1117	110	1.0828	-	-	-
1.2132	120	0.9545	-	-	-
1.3147	130	1.1774	-	-	-
1.4162	140	0.55	-	-	-
1.5178	150	0.891	-	-	-
1.6193	160	0.9661	-	-	-
1.7208	170	0.9355	-	-	-
1.8223	180	0.9888	-	-	-
1.9239	190	1.0157	-	-	-
2.0	198	-	0.8067	0.7945	0.7742
2.0203	200	0.7944	-	-	-
2.1218	210	0.5637	-	-	-
2.2234	220	0.3895	-	-	-
2.3249	230	1.0888	-	-	-
2.4264	240	0.8784	-	-	-
2.5279	250	0.5746	-	-	-
2.6294	260	1.064	-	-	-
2.7310	270	0.8036	-	-	-
2.8325	280	0.6005	-	-	-
2.9340	290	0.7571	-	-	-
3.0	297	-	0.81	0.7982	0.7785
3.0305	300	0.6178	-	-	-
3.1320	310	0.5013	-	-	-
3.2335	320	0.7171	-	-	-
3.3350	330	0.5717	-	-	-
3.4365	340	0.7031	-	-	-
3.5381	350	0.8601	-	-	-
3.6396	360	0.597	-	-	-
3.7411	370	0.4611	-	-	-
3.8426	380	0.6503	-	-	-
3.9442	390	0.3176	-	-	-
4.0	396	-	0.8091	0.7978	0.7797

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.11.13
• Sentence Transformers: 4.1.0
• Transformers: 4.52.4
• PyTorch: 2.6.0+cu124
• Accelerate: 1.8.1
• Datasets: 3.6.0
• 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}
}