iammayur/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, 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
• Supported Modality: Text
• 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({'transformer_task': 'feature-extraction', 'modality_config': {'text': {'method': 'forward', 'method_output_name': 'last_hidden_state'}}, 'module_output_name': 'token_embeddings', 'architecture': 'BertModel'})
  (1): Pooling({'embedding_dimension': 768, 'pooling_mode': 'cls', '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("iammayur/bge-base-financial-matryoshka")
# Run inference
queries = [
    'Cost of sales for the company was $5,920.5 million in 2022, up from $4,922.7 million in 2021, which represents a 20.3% increase. This included $767.7 million of unfavorable costs driven by higher sales volume and increased supply chain inflation costs, including logistics and labor.',
]
documents = [
    'What were the main components of the increased cost of sales in 2022 compared to 2021?',
    'How much is the service fee on client cash deposits held at the TD Depository Institutions under the 2023 IDA agreement?',
    'What is the primary method by which the company manages its cash, cash equivalents, and marketable securities?',
]
query_embeddings = model.encode_query(queries)
document_embeddings = model.encode_document(documents)
print(query_embeddings.shape, document_embeddings.shape)
# [1, 768] [3, 768]

# Get the similarity scores for the embeddings
similarities = model.similarity(query_embeddings, document_embeddings)
print(similarities)
# tensor([[0.8170, 0.1130, 0.1414]])

Evaluation

Metrics

Information Retrieval

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

  {
      "truncate_dim": 768
  }
  

Metric	Value
cosine_accuracy@1	0.7257
cosine_accuracy@3	0.8357
cosine_accuracy@5	0.8771
cosine_accuracy@10	0.9014
cosine_precision@1	0.7257
cosine_precision@3	0.2786
cosine_precision@5	0.1754
cosine_precision@10	0.0901
cosine_recall@1	0.7257
cosine_recall@3	0.8357
cosine_recall@5	0.8771
cosine_recall@10	0.9014
cosine_ndcg@10	0.8168
cosine_mrr@10	0.7892
cosine_map@100	0.7928

Information Retrieval

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

  {
      "truncate_dim": 512
  }
  

Metric	Value
cosine_accuracy@1	0.7229
cosine_accuracy@3	0.8357
cosine_accuracy@5	0.8729
cosine_accuracy@10	0.9029
cosine_precision@1	0.7229
cosine_precision@3	0.2786
cosine_precision@5	0.1746
cosine_precision@10	0.0903
cosine_recall@1	0.7229
cosine_recall@3	0.8357
cosine_recall@5	0.8729
cosine_recall@10	0.9029
cosine_ndcg@10	0.8158
cosine_mrr@10	0.7876
cosine_map@100	0.7912

Information Retrieval

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

  {
      "truncate_dim": 256
  }
  

Metric	Value
cosine_accuracy@1	0.7186
cosine_accuracy@3	0.8329
cosine_accuracy@5	0.8743
cosine_accuracy@10	0.9057
cosine_precision@1	0.7186
cosine_precision@3	0.2776
cosine_precision@5	0.1749
cosine_precision@10	0.0906
cosine_recall@1	0.7186
cosine_recall@3	0.8329
cosine_recall@5	0.8743
cosine_recall@10	0.9057
cosine_ndcg@10	0.8137
cosine_mrr@10	0.7839
cosine_map@100	0.7872

Information Retrieval

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

  {
      "truncate_dim": 128
  }
  

Metric	Value
cosine_accuracy@1	0.7043
cosine_accuracy@3	0.8257
cosine_accuracy@5	0.8586
cosine_accuracy@10	0.8986
cosine_precision@1	0.7043
cosine_precision@3	0.2752
cosine_precision@5	0.1717
cosine_precision@10	0.0899
cosine_recall@1	0.7043
cosine_recall@3	0.8257
cosine_recall@5	0.8586
cosine_recall@10	0.8986
cosine_ndcg@10	0.8034
cosine_mrr@10	0.7726
cosine_map@100	0.7761

Information Retrieval

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

  {
      "truncate_dim": 64
  }
  

Metric	Value
cosine_accuracy@1	0.6643
cosine_accuracy@3	0.7829
cosine_accuracy@5	0.8271
cosine_accuracy@10	0.8743
cosine_precision@1	0.6643
cosine_precision@3	0.261
cosine_precision@5	0.1654
cosine_precision@10	0.0874
cosine_recall@1	0.6643
cosine_recall@3	0.7829
cosine_recall@5	0.8271
cosine_recall@10	0.8743
cosine_ndcg@10	0.7682
cosine_mrr@10	0.7343
cosine_map@100	0.7385

Training Details

Training Dataset

json

• Dataset: json
• Size: 6,300 training samples
• Columns: positive and anchor
• Approximate statistics based on the first 100 samples:

  	positive	anchor
  type	string	string
  modality	text	text
  details	min: 14 tokens mean: 42.72 tokens max: 122 tokens	min: 10 tokens mean: 20.15 tokens max: 40 tokens

• Samples:

  positive	anchor
  Alphabet is a collection of businesses, the largest of which is Google. Alphabet reports Google in two segments, Google Services and Google Cloud; all non-Google businesses are collectively reported as Other Bets.	What are Alphabet's primary business segments and how are they reported?
  The company has the option to redeem the Notes for cash between specific dates if the sale price of their common stock exceeds a set threshold relative to the conversion price over a specified number of trading days, including on the day immediately before the notice of redemption is sent.	What are the conditions under which the company may redeem the Notes for cash?
  Net earnings attributable to Hasbro, Inc. declined in 2022 to $203.5 million, compared to $428.7 million in 2021.	How much did Hasbro's net earnings attributable to Hasbro, Inc. decline in 2022 compared to 2021?

• 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

• per_device_train_batch_size: 16
• num_train_epochs: 4
• learning_rate: 2e-05
• lr_scheduler_type: cosine
• warmup_steps: 0.1
• gradient_accumulation_steps: 16
• bf16: True
• per_device_eval_batch_size: 16
• load_best_model_at_end: True
• batch_sampler: no_duplicates

All Hyperparameters

Click to expand

• per_device_train_batch_size: 16
• num_train_epochs: 4
• max_steps: -1
• learning_rate: 2e-05
• lr_scheduler_type: cosine
• lr_scheduler_kwargs: None
• warmup_steps: 0.1
• optim: adamw_torch_fused
• optim_args: None
• weight_decay: 0.0
• adam_beta1: 0.9
• adam_beta2: 0.999
• adam_epsilon: 1e-08
• optim_target_modules: None
• gradient_accumulation_steps: 16
• average_tokens_across_devices: True
• max_grad_norm: 1.0
• label_smoothing_factor: 0.0
• bf16: True
• fp16: False
• bf16_full_eval: False
• fp16_full_eval: False
• tf32: None
• gradient_checkpointing: False
• gradient_checkpointing_kwargs: None
• torch_compile: False
• torch_compile_backend: None
• torch_compile_mode: None
• use_liger_kernel: False
• liger_kernel_config: None
• use_cache: False
• neftune_noise_alpha: None
• torch_empty_cache_steps: None
• auto_find_batch_size: False
• log_on_each_node: True
• logging_nan_inf_filter: True
• include_num_input_tokens_seen: no
• log_level: passive
• log_level_replica: warning
• disable_tqdm: False
• project: huggingface
• trackio_space_id: None
• trackio_bucket_id: None
• trackio_static_space_id: None
• per_device_eval_batch_size: 16
• prediction_loss_only: True
• eval_on_start: False
• eval_do_concat_batches: True
• eval_use_gather_object: False
• eval_accumulation_steps: None
• include_for_metrics: []
• batch_eval_metrics: False
• save_only_model: False
• save_on_each_node: False
• enable_jit_checkpoint: False
• push_to_hub: False
• hub_private_repo: None
• hub_model_id: None
• hub_strategy: every_save
• hub_always_push: False
• hub_revision: None
• load_best_model_at_end: True
• ignore_data_skip: False
• restore_callback_states_from_checkpoint: False
• full_determinism: False
• seed: 42
• data_seed: None
• use_cpu: False
• accelerator_config: {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}
• parallelism_config: None
• dataloader_drop_last: False
• dataloader_num_workers: 0
• dataloader_pin_memory: True
• dataloader_persistent_workers: False
• dataloader_prefetch_factor: None
• dataloader_multiprocessing_context: None
• dataloader_in_order: True
• remove_unused_columns: True
• label_names: None
• train_sampling_strategy: random
• length_column_name: length
• ddp_find_unused_parameters: None
• ddp_bucket_cap_mb: None
• ddp_broadcast_buffers: False
• ddp_static_graph: None
• ddp_backend: None
• ddp_timeout: 1800
• fsdp: None
• fsdp_config: None
• deepspeed: None
• debug: []
• skip_memory_metrics: True
• do_predict: False
• resume_from_checkpoint: None
• local_rank: -1
• prompts: None
• batch_sampler: no_duplicates
• multi_dataset_batch_sampler: proportional
• router_mapping: {}
• learning_rate_mapping: {}
• warmup_ratio: None

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
0.4061	10	1.0226	-	-	-	-	-
0.8122	20	0.4788	-	-	-	-	-
1.0	25	-	0.8082	0.8071	0.8053	0.7850	0.7460
1.2030	30	0.3296	-	-	-	-	-
1.6091	40	0.2549	-	-	-	-	-
2.0	50	0.2531	0.8140	0.8147	0.8125	0.8013	0.7630
2.4061	60	0.2018	-	-	-	-	-
2.8122	70	0.2165	-	-	-	-	-
3.0	75	-	0.8172	0.8165	0.8126	0.8019	0.7680
3.2030	80	0.1983	-	-	-	-	-
3.6091	90	0.1817	-	-	-	-	-
4.0	100	0.1899	0.8168	0.8158	0.8137	0.8034	0.7682

• The bold row denotes the saved checkpoint.

Training Time

• Training: 30.3 minutes

Framework Versions

• Python: 3.13.15
• Sentence Transformers: 5.7.0
• Transformers: 5.16.1
• PyTorch: 2.11.0+cu128
• Accelerate: 1.14.0
• Datasets: 4.8.5
• Tokenizers: 0.23.1

Additional Resources

• Training and Finetuning Embedding Models with Sentence Transformers: the end-to-end guide for training or finetuning Sentence Transformer models.
• Introduction to Matryoshka Embedding Models: variable-size embeddings that can be truncated with minimal quality loss.
• Binary and Scalar Embedding Quantization for Significantly Faster & Cheaper Retrieval: post-training compression of embedding vectors.
• Multimodal Embedding & Reranker Models with Sentence Transformers: use text, image, audio, and video models through the same API.
• Training and Finetuning Multimodal Embedding & Reranker Models with Sentence Transformers: train multimodal embedding models, with a Visual Document Retrieval walkthrough.

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{oord2019representationlearningcontrastivepredictive,
      title={Representation Learning with Contrastive Predictive Coding},
      author={Aaron van den Oord and Yazhe Li and Oriol Vinyals},
      year={2019},
      eprint={1807.03748},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/1807.03748},
}