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 <!-- at revision a5beb1e3e68b9ab74eb54cfd186867f64f240e1a -->
• 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("shivamsharma1967/bge-base-financial-matryoshka")
# Run inference
sentences = [
    'Table of Contents\nAMAZON.COM, INC.\nCONSOLIDATED STATEMENTS OF OPERATIONS\n(in millions, except per share data)\n \n \nYear Ended December 31,\n \n2015\n \n2016\n \n2017\nNet product sales\n$\n79,268 $\n94,665 $\n118,573\nNet service sales\n27,738 \n41,322 \n59,293\nTotal net sales\n107,006 \n135,987 \n177,866\nOperating expenses:\n \n \n \nCost of sales\n71,651 \n88,265 \n111,934\nFulfillment\n13,410 \n17,619 \n25,249\nMarketing\n5,254 \n7,233 \n10,069\nTechnology and content\n12,540 \n16,085 \n22,620\nGeneral and administrative\n1,747 \n2,432 \n3,674\nOther operating expense, net\n171 \n167 \n214\nTotal operating expenses\n104,773 \n131,801 \n173,760\nOperating income\n2,233 \n4,186 \n4,106\nInterest income\n50 \n100 \n202\nInterest expense\n(459) \n(484) \n(848)\nOther income (expense), net\n(256) \n90 \n346\nTotal non-operating income (expense)\n(665) \n(294) \n(300)\nIncome before income taxes\n1,568 \n3,892 \n3,806\nProvision for income taxes\n(950) \n(1,425) \n(769)\nEquity-method investment activity, net of tax\n(22) \n(96) \n(4)\nNet income\n$\n596 $\n2,371 $\n3,033\nBasic earnings per share\n$\n1.28 $\n5.01 $\n6.32\nDiluted earnings per share\n$\n1.25 $\n4.90 $\n6.15\nWeighted-average shares used in computation of earnings per share:\n \n \n \nBasic\n467 \n474 \n480\nDiluted\n477 \n484 \n493\nSee accompanying notes to consolidated financial statements.\n38\nTable of Contents\nAMAZON.COM, INC.\nCONSOLIDATED STATEMENTS OF OPERATIONS\n(in millions, except per share data)\n \n \nYear Ended December 31,\n \n2015\n \n2016\n \n2017\nNet product sales\n$\n79,268 $\n94,665 $\n118,573\nNet service sales\n27,738 \n41,322 \n59,293\nTotal net sales\n107,006 \n135,987 \n177,866\nOperating expenses:\n \n \n \nCost of sales\n71,651 \n88,265 \n111,934\nFulfillment\n13,410 \n17,619 \n25,249\nMarketing\n5,254 \n7,233 \n10,069\nTechnology and content\n12,540 \n16,085 \n22,620\nGeneral and administrative\n1,747 \n2,432 \n3,674\nOther operating expense, net\n171 \n167 \n214\nTotal operating expenses\n104,773 \n131,801 \n173,760\nOperating income\n2,233 \n4,186 \n4,106\nInterest income\n50 \n100 \n202\nInterest expense\n(459) \n(484) \n(848)\nOther income (expense), net\n(256) \n90 \n346\nTotal non-operating income (expense)\n(665) \n(294) \n(300)\nIncome before income taxes\n1,568 \n3,892 \n3,806\nProvision for income taxes\n(950) \n(1,425) \n(769)\nEquity-method investment activity, net of tax\n(22) \n(96) \n(4)\nNet income\n$\n596 $\n2,371 $\n3,033\nBasic earnings per share\n$\n1.28 $\n5.01 $\n6.32\nDiluted earnings per share\n$\n1.25 $\n4.90 $\n6.15\nWeighted-average shares used in computation of earnings per share:\n \n \n \nBasic\n467 \n474 \n480\nDiluted\n477 \n484 \n493\nSee accompanying notes to consolidated financial statements.\n38',
    "What is Amazon's year-over-year change in revenue from FY2016 to FY2017 (in units of percents and round to one decimal place)? Calculate what was asked by utilizing the line items clearly shown in the statement of income.",
    'What is the FY2018 - FY2020 3 year average of capex as a % of revenue for MGM Resorts? Answer in units of percents and round to one decimal place. Please utilize information provided primarily within the statement of cash flows and the statement of income.',
]
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]

<!-- ### Direct Usage (Transformers) <details><summary>Click to see the direct usage in Transformers</summary> </details> --> <!-- ### Downstream Usage (Sentence Transformers) You can finetune this model on your own dataset. <details><summary>Click to expand</summary> </details> --> <!-- ### Out-of-Scope Use *List how the model may foreseeably be misused and address what users ought not to do with the model.* -->

Evaluation

Metrics

Information Retrieval

• Datasets: dim_768, dim_512, dim_256, dim_128 and dim_64
• Evaluated with <code>InformationRetrievalEvaluator</code>

<!-- ## Bias, Risks and Limitations *What are the known or foreseeable issues stemming from this model? You could also flag here known failure cases or weaknesses of the model.* --> <!-- ### Recommendations *What are recommendations with respect to the foreseeable issues? For example, filtering explicit content.* -->

Training Details

Training Dataset

json

• Dataset: json
• Size: 135 training samples
• Columns: <code>positive</code> and <code>anchor</code>
• Approximate statistics based on the first 135 samples:

  	positive	anchor
  type	string	string
  details	<ul><li>min: 359 tokens</li><li>mean: 508.73 tokens</li><li>max: 512 tokens</li></ul>	<ul><li>min: 11 tokens</li><li>mean: 39.7 tokens</li><li>max: 175 tokens</li></ul>

• Samples:

  positive	anchor
  <code>Twelve Months Ended June 30, 2022<br>Twelve Months Ended June 30, 2023<br>($ million)<br>EBITDA<br>EBIT<br>Net <br>Income<br>EPS <br>(Diluted<br>US <br>cents)(1)<br>EBITDA<br>EBIT<br>Net <br>Income<br>EPS <br>(Diluted <br>US <br>cents)(1)<br>Net income attributable to Amcor<br> <br>805 <br> <br>805 <br> <br>805 <br> <br>52.9 <br> <br>1,048 <br> <br>1,048 <br> <br>1,048 <br> <br>70.5 <br>Net income attributable to non-controlling <br>interests<br> <br>10 <br> <br>10 <br> <br>10 <br> <br>10 <br>Tax expense<br> <br>300 <br> <br>300 <br> <br>193 <br> <br>193 <br>Interest expense, net<br> <br>135 <br> <br>135 <br> <br>259 <br> <br>259 <br>Depreciation and amortization<br> <br>579 <br> <br>569 <br>EBITDA, EBIT, Net income and EPS<br> <br>1,829 <br> <br>1,250 <br> <br>805 <br> <br>52.9 <br> <br>2,080 <br> <br>1,510 <br> <br>1,048 <br> <br>70.5 <br>2019 Bemis Integration Plan<br> <br>37 <br> <br>37 <br> <br>37 <br> <br>2.5 <br> <br> <br> <br> <br> <br> <br> <br> <br>Net loss on disposals(2)<br> <br>10 <br> <br>10 <br> <br>10 <br> <br>0.7 <br> <br> <br> <br> <br> <br> <br> <br> <br>Impact of hyperinflation<br> <br>16 <br> <br>16 <br> <br>16 <br> <br>1.0 <br> <br>24 <br> <br>24 <br> <br>24 <br> <br>1.9 <br>Property and other losses, net(3)<br> <br>13 <br> <br>13 <br> <br>13 <br> <br>0.8 <br> <br>2 <br> <br>2 <br> <br>2 <br> <br>0.1 <br>Russia-Ukraine conflict impacts(4)<br> <br>200 <br> <br>200 <br> <br>200 <br> <br>13.2 <br> <br>(90) <br>(90) <br>(90) <br>(6.0) <br>Pension settlements<br> <br>8...</code>	<code>What Was AMCOR's Adjusted Non GAAP EBITDA for FY 2023</code>
  <code>SQUARE,INC.<br>CONSOLIDATEDBALANCESHEETS<br>(In thousands, except share and per share data)<br><br>December31,<br><br>2016<br><br>2015<br>Assets<br><br> <br>Currentassets:<br><br> <br>Cashandcashequivalents<br>$<br>452,030 $<br>461,329<br>Short-terminvestments<br>59,901 <br><br>Restrictedcash<br>22,131 <br>13,537<br>Settlementsreceivable<br>321,102 <br>142,727<br>Customerfundsheld<br>43,574 <br>9,446<br>Loansheldforsale<br>42,144 <br>604<br>Merchantcashadvancereceivable,net<br>4,212 <br>36,473<br>Othercurrentassets<br>56,331 <br>41,447<br>Totalcurrentassets<br>1,001,425 <br>705,563<br>Propertyandequipment,net<br>88,328 <br>87,222<br>Goodwill<br>57,173 <br>56,699<br>Acquiredintangibleassets,net<br>19,292 <br>26,776<br>Long-terminvestments<br>27,366 <br><br>Restrictedcash<br>14,584 <br>14,686<br>Otherassets<br>3,194 <br>3,826<br>Totalassets<br>$<br>1,211,362 $<br>894,772<br>LiabilitiesandStockholdersEquity<br><br> <br>Currentliabilities:<br><br> <br>Accountspayable<br>$<br>12,602 $<br>18,869<br>Customerspayable<br>388,058 <br>215,365<br>Customerfundsobligation<br>43,574 <br>9,446<br>Accruedtransactionlosses<br>20,064 <br>17,176<br>Accruedexpenses<br>39,543 <br>44,401<br>Othercurrentliabilities<br>73,623 <br>28,945<br>Totalcurrentliabilities<br>577,464 <br>33...</code>	<code>Considering the data in the balance sheet, what is Block's (formerly known as Square) FY2016 working capital ratio? Define working capital ratio as total current assets divided by total current liabilities. Round your answer to two decimal places.</code>
  <code>Consolidated Balance Sheets <br>Verizon Communications Inc. and Subsidiaries <br>(dollars in millions, except per share amounts) <br>At December 31,<br>2022<br>2021 <br>Assets <br>Current assets <br>Cash and cash equivalents<br>$ <br>2,605 <br>$ <br>2,921 <br>Accounts receivable<br> <br>25,332 <br> <br>24,742 <br>Less Allowance for credit losses<br> <br>826 <br> <br>896 <br>Accounts receivable, net <br> <br>24,506 <br> <br>23,846 <br>Inventories<br> <br>2,388 <br> <br>3,055 <br>Prepaid expenses and other<br> <br>8,358 <br> <br>6,906 <br>Total current assets<br> <br>37,857 <br> <br>36,728 <br>Property, plant and equipment<br> <br>307,689 <br> <br>289,897 <br>Less Accumulated depreciation<br> <br>200,255 <br> <br>190,201 <br>Property, plant and equipment, net<br> <br>107,434 <br> <br>99,696 <br>Investments in unconsolidated businesses<br> <br>1,071 <br> <br>1,061 <br>Wireless licenses<br> <br>149,796 <br> <br>147,619 <br>Goodwill<br> <br>28,671 <br> <br>28,603 <br>Other intangible assets, net<br> <br>11,461 <br> <br>11,677 <br>Operating lease right-of-use assets<br> <br>26,130 <br> <br>27,883 <br>Other assets<br> <br>17,260 <br> <br>13,329 <br>Total assets<br>$ <br>379,680 <br>$ <br>366,596 <br>Liabilities and Equity <br>Current liabilities <br>Debt maturing within o...</code>	<code>Does Verizon have a reasonably healthy liquidity profile based on its quick ratio for FY 2022? If the quick ratio is not relevant to measure liquidity, please state that and explain why.</code>

• Loss: <code>MatryoshkaLoss</code> 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: 16
• 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

<details><summary>Click to expand</summary>

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: epoch
• prediction_loss_only: True
• per_device_train_batch_size: 16
• 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: 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

</details>

Training Logs

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.11.11
• Sentence Transformers: 3.4.1
• Transformers: 4.48.3
• PyTorch: 2.5.1+cu124
• Accelerate: 1.3.0
• Datasets: 3.3.2
• 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}
}

<!-- ## Glossary *Clearly define terms in order to be accessible across audiences.* --> <!-- ## Model Card Authors *Lists the people who create the model card, providing recognition and accountability for the detailed work that goes into its construction.* --> <!-- ## Model Card Contact *Provides a way for people who have updates to the Model Card, suggestions, or questions, to contact the Model Card authors.* -->