mbegerez/bge-base-financial-matryoshka

SentenceTransformer based on BAAI/bge-base-en-v1.5

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

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("mbegerez/bge-base-financial-matryoshka")
# Run inference
sentences = [
    "Belonging - Build a Walmart for everyone: a diverse, equitable and inclusive company, where associates' ideas and opinions matter. We are focused on having an inclusive culture where everyone feels they belong. We publish our diversity representation twice yearly, and hold ourselves accountable to providing recurring culture, diversity, equity, and inclusion updates to senior leadership, including our President and CEO, and members of the Board of Directors. Of the approximately 2.1 million associates employed worldwide, 52% identify as women. In the U.S., 50% of the approximately 1.6 million associates identify as people of color. We review our processes regarding our commitment to fair-pay practices.",
    'How does Walmart support diversity, equity, and inclusion within its workforce?',
    'What was the net sales of the company in fiscal 2022?',
]
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

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

  {
      "truncate_dim": 768
  }
  

Metric	Value
cosine_accuracy@1	0.7071
cosine_accuracy@3	0.8543
cosine_accuracy@5	0.89
cosine_accuracy@10	0.9286
cosine_precision@1	0.7071
cosine_precision@3	0.2848
cosine_precision@5	0.178
cosine_precision@10	0.0929
cosine_recall@1	0.7071
cosine_recall@3	0.8543
cosine_recall@5	0.89
cosine_recall@10	0.9286
cosine_ndcg@10	0.8212
cosine_mrr@10	0.7865
cosine_map@100	0.7895

Information Retrieval

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

  {
      "truncate_dim": 512
  }
  

Metric	Value
cosine_accuracy@1	0.7129
cosine_accuracy@3	0.8571
cosine_accuracy@5	0.8843
cosine_accuracy@10	0.9243
cosine_precision@1	0.7129
cosine_precision@3	0.2857
cosine_precision@5	0.1769
cosine_precision@10	0.0924
cosine_recall@1	0.7129
cosine_recall@3	0.8571
cosine_recall@5	0.8843
cosine_recall@10	0.9243
cosine_ndcg@10	0.8226
cosine_mrr@10	0.7896
cosine_map@100	0.793

Information Retrieval

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

  {
      "truncate_dim": 256
  }
  

Metric	Value
cosine_accuracy@1	0.71
cosine_accuracy@3	0.8486
cosine_accuracy@5	0.89
cosine_accuracy@10	0.9214
cosine_precision@1	0.71
cosine_precision@3	0.2829
cosine_precision@5	0.178
cosine_precision@10	0.0921
cosine_recall@1	0.71
cosine_recall@3	0.8486
cosine_recall@5	0.89
cosine_recall@10	0.9214
cosine_ndcg@10	0.8192
cosine_mrr@10	0.7859
cosine_map@100	0.7894

Information Retrieval

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

  {
      "truncate_dim": 128
  }
  

Metric	Value
cosine_accuracy@1	0.6829
cosine_accuracy@3	0.83
cosine_accuracy@5	0.8671
cosine_accuracy@10	0.9143
cosine_precision@1	0.6829
cosine_precision@3	0.2767
cosine_precision@5	0.1734
cosine_precision@10	0.0914
cosine_recall@1	0.6829
cosine_recall@3	0.83
cosine_recall@5	0.8671
cosine_recall@10	0.9143
cosine_ndcg@10	0.8009
cosine_mrr@10	0.7644
cosine_map@100	0.7677

Information Retrieval

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

  {
      "truncate_dim": 64
  }
  

Metric	Value
cosine_accuracy@1	0.65
cosine_accuracy@3	0.7929
cosine_accuracy@5	0.83
cosine_accuracy@10	0.8771
cosine_precision@1	0.65
cosine_precision@3	0.2643
cosine_precision@5	0.166
cosine_precision@10	0.0877
cosine_recall@1	0.65
cosine_recall@3	0.7929
cosine_recall@5	0.83
cosine_recall@10	0.8771
cosine_ndcg@10	0.765
cosine_mrr@10	0.729
cosine_map@100	0.7338

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: 6 tokens mean: 44.63 tokens max: 301 tokens	min: 7 tokens mean: 20.6 tokens max: 45 tokens

• Samples:

  positive	anchor
  Some key challenges and trends addressed in the 'Trends and Uncertainties' section of the MD&A include material events such as the decreasing but continuous supply chain constraints, uneven demands, new technology adoptions, and a conservative customer spending environment within a mixed macroeconomic context.	What are some of the key challenges and developments that Hewlett Packard Enterprise highlighted in the 'Trends and Uncertainties' section of their MD&A for fiscal 2022?
  Supply of Components Although most components essential to the Company’s business are generally available from multiple sources, certain components are currently obtained from single or limited sources.	From which sources does Apple obtain certain essential components?
  The document lists PPD reinstatement premiums as showing an unfavorable (2) million U.S. dollars adjustment.	What were the reinstatement premiums related to PPD expenses for the year shown in the document?

• 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: 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}
• tp_size: 0
• 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_768_cosine_ndcg@10	dim_512_cosine_ndcg@10	dim_256_cosine_ndcg@10	dim_128_cosine_ndcg@10	dim_64_cosine_ndcg@10
0.8122	10	25.1822	-	-	-	-	-
1.0	13	-	0.8089	0.8069	0.8069	0.7846	0.7410
1.5685	20	11.1815	-	-	-	-	-
2.0	26	-	0.8174	0.8224	0.8189	0.7988	0.7594
2.3249	30	7.2855	-	-	-	-	-
3.0	39	-	0.8205	0.8225	0.8195	0.7999	0.7636
3.0812	40	7.356	-	-	-	-	-
3.731	48	-	0.8212	0.8226	0.8192	0.8009	0.765

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.11.12
• Sentence Transformers: 4.1.0
• Transformers: 4.51.3
• PyTorch: 2.6.0+cu124
• Accelerate: 1.5.2
• Datasets: 3.5.0
• 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}
}