moresearch/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("moresearch/bge-base-financial-matryoshka")
# Run inference
sentences = [
    'As of December 31, 2023, we employed about 113,200 full-time persons of whom approximately 62,400 were located outside the United States. In the United States, we employed approximately 50,800 full-time persons.',
    'How many full-time employees were employed by the company as of December 31, 2023, and how are they distributed geographically?',
    'What challenges do solar power system owners face with traditional string inverters?',
]
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_384, dim_256, dim_128, dim_64 and dim_32
• Evaluated with InformationRetrievalEvaluator

Metric	dim_384	dim_256	dim_128	dim_64	dim_32
cosine_accuracy@1	0.7157	0.7157	0.7057	0.6729	0.5957
cosine_accuracy@3	0.8486	0.8486	0.8371	0.8129	0.7414
cosine_accuracy@5	0.8743	0.8786	0.8643	0.85	0.7829
cosine_accuracy@10	0.9171	0.9143	0.9086	0.8814	0.8314
cosine_precision@1	0.7157	0.7157	0.7057	0.6729	0.5957
cosine_precision@3	0.2829	0.2829	0.279	0.271	0.2471
cosine_precision@5	0.1749	0.1757	0.1729	0.17	0.1566
cosine_precision@10	0.0917	0.0914	0.0909	0.0881	0.0831
cosine_recall@1	0.7157	0.7157	0.7057	0.6729	0.5957
cosine_recall@3	0.8486	0.8486	0.8371	0.8129	0.7414
cosine_recall@5	0.8743	0.8786	0.8643	0.85	0.7829
cosine_recall@10	0.9171	0.9143	0.9086	0.8814	0.8314
cosine_ndcg@10	0.8199	0.8188	0.809	0.7829	0.7159
cosine_mrr@10	0.7885	0.7878	0.7771	0.7508	0.6788
cosine_map@100	0.7918	0.7912	0.7807	0.7551	0.684

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: 7 tokens mean: 45.87 tokens max: 288 tokens	min: 2 tokens mean: 20.32 tokens max: 41 tokens

• Samples:

  positive	anchor
  The company maintains a revolving credit facility that, unless extended, terminates on July 6, 2026.	What is the expiration date of the company's revolving credit facility, unless extended?
  The management of Bank of America Corporation is responsible for establishing and maintaining adequate internal control over financial reporting. The Corporation’s internal control over financial reporting is designed to provide reasonable assurance about the reliability of financial reporting and the preparation of financial statements in accordance with accounting principles generally accepted in the United States of America. Management's responsibilities include maintaining records that, in reasonable detail, accurately and fairly reflect the transactions and dispositions of the assets of the Corporation; ensuring that transactions are recorded as necessary for the preparation of financial statements; and preventing or detecting unauthorized acquisition, use, or disposition of the Corporation’s assets that could have a material effect on the financial statements.	What is the role of Bank of America Corporation's management in relation to internal control over financial reporting?
  In 2020, Gilead implemented multiple programs to train managers on inclusion and diversity topics and created strategies and initiatives focused on attracting, developing and retaining diverse talent and driving an inclusive culture in our workplace.	What initiatives has Gilead undertaken to promote workplace diversity?

• 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_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: 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: 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: 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: False
• 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_384_cosine_ndcg@10	dim_256_cosine_ndcg@10	dim_128_cosine_ndcg@10	dim_64_cosine_ndcg@10	dim_32_cosine_ndcg@10
0.8122	10	1.5733	-	-	-	-	-
0.9746	12	-	0.8075	0.8045	0.7876	0.7643	0.6844
1.6244	20	0.6549	-	-	-	-	-
1.9492	24	-	0.8188	0.8169	0.8035	0.7789	0.7107
2.4365	30	0.4373	-	-	-	-	-
2.9239	36	-	0.8210	0.8183	0.8079	0.7835	0.7161
3.2487	40	0.3951	-	-	-	-	-
3.8985	48	-	0.8199	0.8188	0.809	0.7829	0.7159

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.10.12
• Sentence Transformers: 3.3.1
• Transformers: 4.46.3
• PyTorch: 2.5.1+cu124
• Accelerate: 1.1.1
• Datasets: 3.1.0
• Tokenizers: 0.20.3

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}
}