stella_en_400M_v5-FinanceRAG-v2

SentenceTransformer based on thomaskim1130/stella_en_400M_v5-FinanceRAG

This is a sentence-transformers model finetuned from thomaskim1130/stella_en_400M_v5-FinanceRAG. It maps sentences & paragraphs to a 1024-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: thomaskim1130/stella_en_400M_v5-FinanceRAG <!-- at revision f10dfc9df857453911fda9d3af68b25bb47dd353 -->
• Maximum Sequence Length: 512 tokens
• Output Dimensionality: 1024 tokens
• Similarity Function: Cosine Similarity

<!-- - **Training Dataset:** Unknown --> <!-- - **Language:** Unknown --> <!-- - **License:** Unknown -->

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': False}) with Transformer model: NewModel 
  (1): Pooling({'word_embedding_dimension': 1024, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, '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): Dense({'in_features': 1024, 'out_features': 1024, 'bias': True, 'activation_function': 'torch.nn.modules.linear.Identity'})
)

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("sentence_transformers_model_id")
# Run inference
sentences = [
    "Instruct: Given a web search query, retrieve relevant passages that answer the query.\nQuery: Title: \nText: In the year with lowest amount of Deposits with banks Average volume, what's the increasing rate of Deposits with banks Average volume?",
    'Title: \nText: Additional Interest Rate Details Average Balances and Interest Ratesé\x88¥æ\x93\x9cssets(1)(2)(3)(4)\n|  | Average volume | Interest revenue | % Average rate |\n| In millions of dollars, except rates | 2015 | 2014 | 2013 | 2015 | 2014 | 2013 | 2015 | 2014 | 2013 |\n| Assets |  |  |  |  |  |  |  |  |  |\n| Deposits with banks-5 | $133,790 | $161,359 | $144,904 | $727 | $959 | $1,026 | 0.54% | 0.59% | 0.71% |\n| Federal funds sold and securities borrowed or purchased under agreements to resell-6 |  |  |  |  |  |  |  |  |  |\n| In U.S. offices | $150,359 | $153,688 | $158,237 | $1,211 | $1,034 | $1,133 | 0.81% | 0.67% | 0.72% |\n| In offices outside the U.S.-5 | 84,006 | 101,177 | 109,233 | 1,305 | 1,332 | 1,433 | 1.55 | 1.32 | 1.31 |\n| Total | $234,365 | $254,865 | $267,470 | $2,516 | $2,366 | $2,566 | 1.07% | 0.93% | 0.96% |\n| Trading account assets-7(8) |  |  |  |  |  |  |  |  |  |\n| In U.S. offices | $114,639 | $114,910 | $126,123 | $3,945 | $3,472 | $3,728 | 3.44% | 3.02% | 2.96% |\n| In offices outside the U.S.-5 | 103,348 | 119,801 | 127,291 | 2,141 | 2,538 | 2,683 | 2.07 | 2.12 | 2.11 |\n| Total | $217,987 | $234,711 | $253,414 | $6,086 | $6,010 | $6,411 | 2.79% | 2.56% | 2.53% |\n| Investments |  |  |  |  |  |  |  |  |  |\n| In U.S. offices |  |  |  |  |  |  |  |  |  |\n| Taxable | $214,714 | $188,910 | $174,084 | $3,812 | $3,286 | $2,713 | 1.78% | 1.74% | 1.56% |\n| Exempt from U.S. income tax | 20,034 | 20,386 | 18,075 | 443 | 626 | 811 | 2.21 | 3.07 | 4.49 |\n| In offices outside the U.S.-5 | 102,376 | 113,163 | 114,122 | 3,071 | 3,627 | 3,761 | 3.00 | 3.21 | 3.30 |\n| Total | $337,124 | $322,459 | $306,281 | $7,326 | $7,539 | $7,285 | 2.17% | 2.34% | 2.38% |\n| Loans (net of unearned income)(9) |  |  |  |  |  |  |  |  |  |\n| In U.S. offices | $354,439 | $361,769 | $354,707 | $24,558 | $26,076 | $25,941 | 6.93% | 7.21% | 7.31% |\n| In offices outside the U.S.-5 | 273,072 | 296,656 | 292,852 | 15,988 | 18,723 | 19,660 | 5.85 | 6.31 | 6.71 |\n| Total | $627,511 | $658,425 | $647,559 | $40,546 | $44,799 | $45,601 | 6.46% | 6.80% | 7.04% |\n| Other interest-earning assets-10 | $55,060 | $40,375 | $38,233 | $1,839 | $507 | $602 | 3.34% | 1.26% | 1.57% |\n| Total interest-earning assets | $1,605,837 | $1,672,194 | $1,657,861 | $59,040 | $62,180 | $63,491 | 3.68% | 3.72% | 3.83% |\n| Non-interest-earning assets-7 | $218,000 | $224,721 | $222,526 |  |  |  |  |  |  |\n| Total assets from discontinued operations | — | — | 2,909 |  |  |  |  |  |  |\n| Total assets | $1,823,837 | $1,896,915 | $1,883,296 |  |  |  |  |  |  |\nNet interest revenue includes the taxable equivalent adjustments related to the tax-exempt bond portfolio (based on the U. S.  federal statutory tax rate of 35%) of $487 million, $498 million and $521 million for 2015, 2014 and 2013, respectively.\nInterest rates and amounts include the effects of risk management activities associated with the respective asset categories.\nMonthly or quarterly averages have been used by certain subsidiaries where daily averages are unavailable.\nDetailed average volume, Interest revenue and Interest expense exclude Discontinued operations.\nSee Note 2 to the Consolidated Financial Statements.\nAverage rates reflect prevailing local interest rates, including inflationary effects and monetary corrections in certain countries.\nAverage volumes of securities borrowed or purchased under agreements to resell are reported net pursuant to ASC 210-20-45.\nHowever, Interest revenue excludes the impact of ASC 210-20-45.\nThe fair value carrying amounts of derivative contracts are reported net, pursuant to ASC 815-10-45, in Non-interest-earning assets and Other non-interest bearing liabilities.\nInterest expense on Trading account liabilities of ICG is reported as a reduction of Interest revenue.\nInterest revenue and Interest expense on cash collateral positions are reported in interest on Trading account assets and Trading account liabilities, respectively.\nIncludes cash-basis loans.\nIncludes brokerage receivables.\nDuring 2015, continued management actions, primarily the sale or transfer to held-for-sale of approximately $1.5 billion of delinquent residential first mortgages, including $0.9 billion in the fourth quarter largely associated with the transfer of CitiFinancial loans to held-for-sale referenced above, were the primary driver of the overall improvement in delinquencies within Citi Holdings\x80\x99 residential first mortgage portfolio.\nCredit performance from quarter to quarter could continue to be impacted by the amount of delinquent loan sales or transfers to held-for-sale, as well as overall trends in HPI and interest rates.\nNorth America Residential First Mortgages\x80\x94State Delinquency Trends The following tables set forth the six U. S.  states and/or regions with the highest concentration of Citi\x80\x99s residential first mortgages.\n| In billions of dollars | December 31, 2015 | December 31, 2014 |\n| State-1 | ENR-2 | ENRDistribution | 90+DPD% | %LTV >100%-3 | RefreshedFICO | ENR-2 | ENRDistribution | 90+DPD% | %LTV >100%-3 | RefreshedFICO |\n| CA | $19.2 | 37% | 0.2% | 1% | 754 | $18.9 | 31% | 0.6% | 2% | 745 |\n| NY/NJ/CT-4 | 12.7 | 25 | 0.8 | 1 | 751 | 12.2 | 20 | 1.9 | 2 | 740 |\n| VA/MD | 2.2 | 4 | 1.2 | 2 | 719 | 3.0 | 5 | 3.0 | 8 | 695 |\n| IL-4 | 2.2 | 4 | 1.0 | 3 | 735 | 2.5 | 4 | 2.5 | 9 | 713 |\n| FL-4 | 2.2 | 4 | 1.1 | 4 | 723 | 2.8 | 5 | 3.0 | 14 | 700 |\n| TX | 1.9 | 4 | 1.0 | — | 711 | 2.5 | 4 | 2.7 | — | 680 |\n| Other | 11.0 | 21 | 1.3 | 2 | 710 | 18.2 | 30 | 3.3 | 7 | 677 |\n| Total-5 | $51.5 | 100% | 0.7% | 1% | 738 | $60.1 | 100% | 2.1% | 4% | 715 |\nNote: Totals may not sum due to rounding.\n(1) Certain of the states are included as part of a region based on Citi\x80\x99s view of similar HPI within the region.\n(2) Ending net receivables.\nExcludes loans in Canada and Puerto Rico, loans guaranteed by U. S.  government agencies, loans recorded at fair value and loans subject to long term standby commitments (LTSCs).\nExcludes balances for which FICO or LTV data are unavailable.\n(3) LTV ratios (loan balance divided by appraised value) are calculated at origination and updated by applying market price data.\n(4) New York, New Jersey, Connecticut, Florida and Illinois are judicial states.\n(5) Improvement in state trends during 2015 was primarily due to the sale or transfer to held-for-sale of residential first mortgages, including the transfer of CitiFinancial residential first mortgages to held-for-sale in the fourth quarter of 2015.\nForeclosures A substantial majority of Citi\x80\x99s foreclosure inventory consists of residential first mortgages.\nAt December 31, 2015, Citi\x80\x99s foreclosure inventory included approximately $0.1 billion, or 0.2%, of the total residential first mortgage portfolio, compared to $0.6 billion, or 0.9%, at December 31, 2014, based on the dollar amount of ending net receivables of loans in foreclosure inventory, excluding loans that are guaranteed by U. S.  government agencies and loans subject to LTSCs.\nNorth America Consumer Mortgage Quarterly Credit Trends \x80\x94Net Credit Losses and Delinquencies\x80\x94Home Equity Loans Citi\x80\x99s home equity loan portfolio consists of both fixed-rate home equity loans and loans extended under home equity lines of credit.\nFixed-rate home equity loans are fully amortizing.\nHome equity lines of credit allow for amounts to be drawn for a period of time with the payment of interest only and then, at the end of the draw period, the then-outstanding amount is converted to an amortizing loan (the interest-only payment feature during the revolving period is standard for this product across the industry).\nAfter conversion, the home equity loans typically have a 20-year amortization period.\nAs of December 31, 2015, Citi\x80\x99s home equity loan portfolio of $22.8 billion consisted of $6.3 billion of fixed-rate home equity loans and $16.5 billion of loans extended under home equity lines of credit (Revolving HELOCs).',
    'Title: \nText: Issuer Purchases of Equity Securities Repurchases of common stock are made to support the Company\x80\x99s stock-based employee compensation plans and for other corporate purposes.\nOn February 13, 2006, the Board of Directors authorized the purchase of $2.0 billion of the Company\x80\x99s common stock between February 13, 2006 and February 28, 2007.\nIn August 2006, 3M\x80\x99s Board of Directors authorized the repurchase of an additional $1.0 billion in share repurchases, raising the total authorization to $3.0 billion for the period from February 13, 2006 to February 28, 2007.\nIn February 2007, 3M\x80\x99s Board of Directors authorized a twoyear share repurchase of up to $7.0 billion for the period from February 12, 2007 to February 28, 2009.',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 1024]

# 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

• Dataset: Evaluate
• 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

Unnamed Dataset

• Size: 2,256 training samples
• Columns: <code>sentence_0</code> and <code>sentence_1</code>
• Approximate statistics based on the first 1000 samples:

  	sentence_0	sentence_1
  type	string	string
  details	<ul><li>min: 28 tokens</li><li>mean: 45.02 tokens</li><li>max: 114 tokens</li></ul>	<ul><li>min: 23 tokens</li><li>mean: 406.36 tokens</li><li>max: 512 tokens</li></ul>

• Samples:

  sentence_0	sentence_1
  <code>Instruct: Given a web search query, retrieve relevant passages that answer the query.<br>Query: Title: <br>Text: What do all Notional sum up, excluding those negative ones in 2008 for As of December 31, 2008 for Financial assets with interest rate risk? (in million)</code>	<code>Title: <br>Text: Cash Flows Our estimated future benefit payments for funded and unfunded plans are as follows (in millions):<br>1 The expected benefit payments for our other postretirement benefit plans are net of estimated federal subsidies expected to be received under the Medicare Prescription Drug, Improvement and Modernization Act of 2003.<br>Federal subsidies are estimated to be $3 million for the period 2019-2023 and $2 million for the period 2024-2028.<br>The Company anticipates making pension contributions in 2019 of $32 million, all of which will be allocated to our international plans.<br>The majority of these contributions are required by funding regulations or law.</code>
  <code>Instruct: Given a web search query, retrieve relevant passages that answer the query.<br>Query: Title: <br>Text: what's the total amount of No surrender charge of 2010 Individual Fixed Annuities, Change in cash of 2008, and Total reserves of 2010 Individual Variable Annuities ?</code>	<code>Title: <br>Text: 2010 and 2009 Comparison Surrender rates have improved compared to the prior year for group retirement products, individual fixed annuities and individual variable annuities as surrenders have returned to more normal levels.<br>Surrender rates for individual fixed annuities have decreased significantly in 2010 due to the low interest rate environment and the relative competitiveness of interest credited rates on the existing block of fixed annuities versus interest rates on alternative investment options available in the marketplace.<br>Surrender rates for group retirement products are expected to increase in 2011 as certain large group surrenders are anticipated.2009 and 2008 Comparison Surrenders and other withdrawals increased in 2009 for group retirement products primarily due to higher large group surrenders.<br>However, surrender rates and withdrawals have improved for individual fixed annuities and individual variable annuities.<br>The following table presents reserves by surrender charge category and surrender rates:<br>
  <code>Instruct: Given a web search query, retrieve relevant passages that answer the query.<br>Query: Title: <br>Text: What was the total amount of elements for RevPAR excluding those elements greater than 150 in 2016 ?</code>	<code>Title: <br>Text: 2016 Compared to 2015 Comparable?Company-Operated North American Properties<br>

• Loss: <code>MultipleNegativesRankingLoss</code> with these parameters:

  {
      "scale": 20.0,
      "similarity_fct": "cos_sim"
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• per_device_train_batch_size: 16
• per_device_eval_batch_size: 16
• num_train_epochs: 2
• fp16: True
• batch_sampler: no_duplicates
• multi_dataset_batch_sampler: round_robin

All Hyperparameters

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

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: steps
• 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: 1
• eval_accumulation_steps: None
• torch_empty_cache_steps: None
• learning_rate: 5e-05
• weight_decay: 0.0
• adam_beta1: 0.9
• adam_beta2: 0.999
• adam_epsilon: 1e-08
• max_grad_norm: 1
• num_train_epochs: 2
• max_steps: -1
• lr_scheduler_type: linear
• lr_scheduler_kwargs: {}
• warmup_ratio: 0.0
• 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: None
• 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: False
• 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
• 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
• eval_on_start: False
• use_liger_kernel: False
• eval_use_gather_object: False
• batch_sampler: no_duplicates
• multi_dataset_batch_sampler: round_robin

</details>

Training Logs

Framework Versions

• Python: 3.10.12
• Sentence Transformers: 3.1.1
• Transformers: 4.45.2
• PyTorch: 2.5.1+cu121
• Accelerate: 1.1.1
• Datasets: 3.1.0
• Tokenizers: 0.20.3

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",
}

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.* -->