ethan-ky/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. 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 tokens
• Similarity Function: Cosine Similarity

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("ethan-ky/bge-base-financial-matryoshka")
# Run inference
sentences = [
    "North America's total net revenues for the fiscal year ended October 1, 2023, were $26,569.6 million.",
    'What was the total net revenue for North America in fiscal 2023?',
    'What are the consequences of impermissible use or disclosure of PHI according to the HITECH Act?',
]
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

Metric	Value
cosine_accuracy@1	0.6171
cosine_accuracy@3	0.7457
cosine_accuracy@5	0.8114
cosine_accuracy@10	0.8586
cosine_precision@1	0.6171
cosine_precision@3	0.2486
cosine_precision@5	0.1623
cosine_precision@10	0.0859
cosine_recall@1	0.6171
cosine_recall@3	0.7457
cosine_recall@5	0.8114
cosine_recall@10	0.8586
cosine_ndcg@10	0.7357
cosine_mrr@10	0.6965
cosine_map@100	0.7016

Information Retrieval

• Dataset: dim_512
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.6214
cosine_accuracy@3	0.74
cosine_accuracy@5	0.8
cosine_accuracy@10	0.8643
cosine_precision@1	0.6214
cosine_precision@3	0.2467
cosine_precision@5	0.16
cosine_precision@10	0.0864
cosine_recall@1	0.6214
cosine_recall@3	0.74
cosine_recall@5	0.8
cosine_recall@10	0.8643
cosine_ndcg@10	0.7382
cosine_mrr@10	0.6983
cosine_map@100	0.7028

Information Retrieval

• Dataset: dim_256
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.6
cosine_accuracy@3	0.7271
cosine_accuracy@5	0.7929
cosine_accuracy@10	0.8443
cosine_precision@1	0.6
cosine_precision@3	0.2424
cosine_precision@5	0.1586
cosine_precision@10	0.0844
cosine_recall@1	0.6
cosine_recall@3	0.7271
cosine_recall@5	0.7929
cosine_recall@10	0.8443
cosine_ndcg@10	0.7182
cosine_mrr@10	0.6783
cosine_map@100	0.6836

Information Retrieval

• Dataset: dim_128
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.5729
cosine_accuracy@3	0.7014
cosine_accuracy@5	0.7557
cosine_accuracy@10	0.8157
cosine_precision@1	0.5729
cosine_precision@3	0.2338
cosine_precision@5	0.1511
cosine_precision@10	0.0816
cosine_recall@1	0.5729
cosine_recall@3	0.7014
cosine_recall@5	0.7557
cosine_recall@10	0.8157
cosine_ndcg@10	0.6915
cosine_mrr@10	0.6522
cosine_map@100	0.658

Information Retrieval

• Dataset: dim_64
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.5143
cosine_accuracy@3	0.6371
cosine_accuracy@5	0.6729
cosine_accuracy@10	0.7357
cosine_precision@1	0.5143
cosine_precision@3	0.2124
cosine_precision@5	0.1346
cosine_precision@10	0.0736
cosine_recall@1	0.5143
cosine_recall@3	0.6371
cosine_recall@5	0.6729
cosine_recall@10	0.7357
cosine_ndcg@10	0.6197
cosine_mrr@10	0.5832
cosine_map@100	0.5907

Training Details

Training Dataset

Unnamed Dataset

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

  	positive	anchor
  type	string	string
  details	min: 2 tokens mean: 45.35 tokens max: 512 tokens	min: 2 tokens mean: 20.67 tokens max: 46 tokens

• Samples:

  positive	anchor
  Our ability to develop and operate units at the right locations and to deliver a customer-centric omni-channel experience largely determines our competitive position within the retail industry. We believe price leadership is a critical part of our business model and we continue to focus on moving our markets towards an EDLP approach. Additionally, our ability to operate food departments effectively has a significant impact on our competitive position in the markets where we operate.	What factors contribute to Walmart International's competitive position?
  tax annual aggregate losses incurred in any year from U.S. hurricane events could be in excess of $3,827 million (or 6.4 percent of total Chubb shareholders’ equity at December 31, 2023).	What is the expected maximum potential loss from hurricane events for Chubb as of the end of 2023?
  The 'Glossary of Terms and Acronyms’ is included on pages 315-321.	What is included on pages 315 to 321 of 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
• 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}
• 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
• 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
• batch_sampler: no_duplicates
• multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	Training Loss	dim_128_cosine_map@100	dim_256_cosine_map@100	dim_512_cosine_map@100	dim_64_cosine_map@100	dim_768_cosine_map@100
0.8122	10	1.3939	-	-	-	-	-
0.9746	12	-	0.658	0.6836	0.7028	0.5907	0.7016
1.6244	20	1.3574	-	-	-	-	-
1.9492	24	-	0.6580	0.6836	0.7028	0.5907	0.7016
2.4365	30	1.3485	-	-	-	-	-
2.9239	36	-	0.6580	0.6836	0.7028	0.5907	0.7016
3.2487	40	1.3606	-	-	-	-	-
3.8985	48	-	0.6580	0.6836	0.7028	0.5907	0.7016

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.9.19
• Sentence Transformers: 3.0.1
• Transformers: 4.41.2
• PyTorch: 2.1.2+cu121
• Accelerate: 0.33.0
• Datasets: 2.19.1
• Tokenizers: 0.19.1

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}
}