ethanteh/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("ethanteh/bge-base-financial-matryoshka")
# Run inference
sentences = [
    'Income Taxes We are subject to income taxes in the U.S. and in many foreign jurisdictions. Significant judgment is required in determining our provision for income taxes, our deferred tax assets and liabilities and any valuation allowance recorded against our net deferred tax assets that are not more likely than not to be realized.',
    'How does the company treat income taxes in its financial reports?',
    "How does YouTube contribute to users' experience according to the company's statement?",
]
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_768, dim_512, dim_256, dim_128 and dim_64
• Evaluated with InformationRetrievalEvaluator

Metric	dim_768	dim_512	dim_256	dim_128	dim_64
cosine_accuracy@1	0.7029	0.6986	0.6971	0.6871	0.6571
cosine_accuracy@3	0.8186	0.8157	0.8086	0.8029	0.7814
cosine_accuracy@5	0.85	0.8529	0.8429	0.8343	0.8114
cosine_accuracy@10	0.8914	0.8986	0.8886	0.8857	0.8586
cosine_precision@1	0.7029	0.6986	0.6971	0.6871	0.6571
cosine_precision@3	0.2729	0.2719	0.2695	0.2676	0.2605
cosine_precision@5	0.17	0.1706	0.1686	0.1669	0.1623
cosine_precision@10	0.0891	0.0899	0.0889	0.0886	0.0859
cosine_recall@1	0.7029	0.6986	0.6971	0.6871	0.6571
cosine_recall@3	0.8186	0.8157	0.8086	0.8029	0.7814
cosine_recall@5	0.85	0.8529	0.8429	0.8343	0.8114
cosine_recall@10	0.8914	0.8986	0.8886	0.8857	0.8586
cosine_ndcg@10	0.7983	0.7982	0.7918	0.7845	0.757
cosine_mrr@10	0.7685	0.7662	0.7611	0.7525	0.7245
cosine_map@100	0.7725	0.7696	0.7652	0.7566	0.7294

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: 8 tokens mean: 46.74 tokens max: 512 tokens	min: 8 tokens mean: 20.52 tokens max: 39 tokens

• Samples:

  positive	anchor
  During 2021, as a result of the enactment of a tax law and the closing of various acquisitions, the company concluded that it is no longer its intention to reinvest its undistributed earnings of its foreign TRSs indefinitely outside the United States.	What is the impact of a tax law change and acquisition closings on the company's intention regarding the reinvestment of undistributed earnings of its foreign TRSs?
  In the year ended December 31, 2023, EBIT-adjusted decreased primarily due to: (1) increased Cost primarily due to increased campaigns and other warranty-related costs of $2.0 billion, increased EV-related charges of $1.9 billion primarily due to $1.6 billion in inventory adjustments to reflect the net realizable value at period end.	What factors contributed to the decrease in GM North America's EBIT-adjusted in 2023?
  Peloton's e-commerce platform offers a range of products and services, including Peloton Bikes, Bike+, Tread, and Row products, along with one-on-one sales consultations.	What types of products and services does Peloton offer through its e-commerce platform?

• 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
• gradient_accumulation_steps: 16
• learning_rate: 2e-05
• num_train_epochs: 4
• lr_scheduler_type: cosine
• warmup_ratio: 0.1
• fp16: 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: 8
• 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: False
• fp16: True
• 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
• 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_768_cosine_ndcg@10	dim_512_cosine_ndcg@10	dim_256_cosine_ndcg@10	dim_128_cosine_ndcg@10	dim_64_cosine_ndcg@10
0.2030	10	9.6119	-	-	-	-	-
0.4061	20	6.108	-	-	-	-	-
0.6091	30	3.9303	-	-	-	-	-
0.8122	40	3.4657	-	-	-	-	-
1.0	50	3.6929	0.7928	0.7891	0.7849	0.7732	0.7465
1.2030	60	1.86	-	-	-	-	-
1.4061	70	1.3879	-	-	-	-	-
1.6091	80	1.4367	-	-	-	-	-
1.8122	90	1.1032	-	-	-	-	-
2.0	100	1.696	0.7996	0.7966	0.7899	0.7815	0.7563
2.2030	110	1.0769	-	-	-	-	-
2.4061	120	0.6618	-	-	-	-	-
2.6091	130	0.912	-	-	-	-	-
2.8122	140	0.6271	-	-	-	-	-
3.0	150	0.9949	0.7984	0.7973	0.7925	0.7835	0.7574
3.2030	160	0.5734	-	-	-	-	-
3.4061	170	0.4934	-	-	-	-	-
3.6091	180	0.6593	-	-	-	-	-
3.8122	190	0.5452	-	-	-	-	-
3.934	196	-	0.7983	0.7982	0.7918	0.7845	0.757

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.11.11
• Sentence Transformers: 3.3.1
• Transformers: 4.48.1
• PyTorch: 2.5.1+cu121
• Accelerate: 1.2.1
• Datasets: 2.19.1
• Tokenizers: 0.21.0

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}
}