bge

This is a sentence-transformers model finetuned from BAAI/bge-m3 on the finmodernbert-triplets dataset. It maps sentences & paragraphs to a 1024-dimensional dense vector space and can be used for retrieval.

Model Details

Model Description

  • Model Type: Sentence Transformer
  • Base model: BAAI/bge-m3
  • Maximum Sequence Length: 512 tokens
  • Output Dimensionality: 1024 dimensions
  • Similarity Function: Cosine Similarity
  • Supported Modality: Text
  • Training Dataset:
    • finmodernbert-triplets
  • Language: en
  • License: apache-2.0

Model Sources

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': 'XLMRobertaModel'})
  (1): Pooling({'embedding_dimension': 1024, '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("sentence_transformers_model_id")
# Run inference
queries = [
    'Adobe reports that approximately 42% of its total revenue is generated from outside the Americas, exposing the company to international risks including currency fluctuations, trade barriers, and complex global data privacy regulations like GDPR and PIPL. Additionally, with 49% of its employees located outside the U.S., the company faces significant operational costs and legal risks related to varying international labor laws and evolving data security requirements.',
]
documents = [
    't generate a corresponding significant increase in revenue and we are unable to achieve the efficiencies we anticipate. In addition, the loss of key sales employees could impact our customer relationships and future ability to sell to certain accounts covered by such employees. We face various risks associated with our operating as a multinational corporation. As a global business that generates approximately 42% of our total revenue from sales to customers outside of the Americas, we are subject to a number of risks, including: • inflation and actions taken by central banks to counter inflation; • foreign currency fluctuations and controls; • international and regional economic, political and labor conditions, including any instability or security concerns abroad, such as uncertainty caused by economic sanctions, trade disputes, armed conflicts and wars; • tax laws (including U.S. taxes on foreign subsidiaries); • increased financial accounting and reporting burdens and complexities; • changes in, or impositions of, legislative or regulatory requirements; • changes in laws governing the free flow of data across international borders; • failure of laws to protect our intellectual property rights adequately; • inadequate local infrastructure and difficulties in managing and staffing international operations; • delays resulting from difficulty in obtaining export licenses for certain technology, tariffs, quotas and other trade barriers; • the imposition of governmental economic sanctions on countries in which we do business or where we plan to expand our business; • costs and delays associated with developing products in multiple languages; • operating in locations with a higher incidence of corruption and fraudulent business practices; and • other factors beyond our control, such as terrorism, war, natural disasters, climate change and pandemics, including the COVID-19 pandemic and resulting restrictions on business activity, which may vary significantly by region. S',
    "y reference as part of Exhibit 4.1 of the Company’s Form 8-K filed on March 11, 2021 (SEC File No. 001-12291). 4.(k) Form of Treasury Unit, incorporated herein by reference as part of Exhibit 4.1 of the Company’s Form 8-K filed on March 11, 2021 (SEC File No. 001-12291). 4.(l) Form of Cash Settled Unit, incorporated herein by reference as part of Exhibit 4.1 of the Company’s Form 8-K filed on March 11, 2021 (SEC File No. 001-12291). 4.(m) Form of Series A Cumulative Perpetual Convertible Preferred Stock Certificate, incorporated herein by reference to Exhibit 4.5 of the Company’s Form 8-K filed on March 11, 2021 (SEC File No. 001-12291). 4.(n) Form of Series B Cumulative Perpetual Preferred Stock Certificate, incorporated herein by reference to Exhibit 4.6 of the Company’s Form 8-K filed on March 11, 2021 (SEC File No. 001-12291). 205 | 2022 Annual Report 10.1 The AES Corporation Profit Sharing and Stock Ownership Plan are incorporated herein by reference to Exhibit 4(c)(1) of the Registration Statement on Form S-8 (Registration No. 33-49262) filed on July 2, 1992. (P) 10.2 The AES Corporation Incentive Stock Option Plan of 1991, as amended, is incorporated herein by reference to Exhibit 10.30 of the Company's Form 10-K for the year ended December 31, 1995 (SEC File No. 00019281). (P) 10.3 Applied Energy Services, Inc. Incentive Stock Option Plan of 1982 is incorporated herein by reference to Exhibit 10.31 of the Registration Statement on Form S-1 (Registration No. 33-40483). (P) 10.4 Deferred Compensation Plan for Executive Officers, as amended, is incorporated herein by reference to Exhibit 10.32 of Amendment No. 1 to the Registration Statement on Form S-1 (Registration No. 33-40483). (P) 10.5 Deferred Compensation Plan for Directors, as amended and restated, on February 17, 2012 is incorporated herein by reference to Exhibit 10.5 of the Company's Form 10-K for the year ended December 31, 2012. 10.6 The AES Corporation Stock Option Plan for Outside Directors, as a",
    "tized premium or discount and valuation allowances. Valuation allowances are established for impaired loans when it is probable that contractual principal and interest will not be collected. Valuation allowances for impaired loans reduce the carrying value to the fair value of the collateral or the present value of the loan's expected future repayment cash flows discounted at the loan's original effective interest rate. Short-term investments are carried at cost or amortized cost that approximates fair value, and include the reinvestment of collateral received in connection with certain securities included in repurchase, resale and lending activities and collateral posted by counterparties in derivative transactions. For these transactions, the Company records an offsetting liability in other liabilities and accrued expenses for the Company's obligation to repay the collateral. Other investments, which consist primarily of policy loans, are carried at the unpaid principal balances. Investment income consists primarily of interest and dividends, net investment income from partnership interests and income from certain derivative transactions. Interest is recognized on an accrual basis and dividends are recorded at the ex-dividend date. Interest income on mortgage-backed and asset-backed securities is determined using the effective yield method, based on estimated principal repayments. Interest income on certain beneficial interests in securitized financial assets is determined using the prospective yield method, based upon projections of expected future cash flows. Income from investments in partnership interests, accounted for on the cost basis, is recognized upon receipt of amounts distributed by the partnerships as income. Accrual of income is suspended for fixed income securities and mortgage loans that are in default or when the receipt of interest payments is in doubt. Realized capital gains and losses include gains and losses on investment dispositions, write-d",
]
query_embeddings = model.encode_query(queries)
document_embeddings = model.encode_document(documents)
print(query_embeddings.shape, document_embeddings.shape)
# [1, 1024] [3, 1024]

# Get the similarity scores for the embeddings
similarities = model.similarity(query_embeddings, document_embeddings)
print(similarities)
# tensor([[0.6508, 0.2301, 0.3242]])

Evaluation

Metrics

Triplet

Metric Value
cosine_accuracy 0.8667

Training Details

Training Dataset

finmodernbert-triplets

  • Dataset: finmodernbert-triplets
  • Size: 70 training samples
  • Columns: query, positive, and negative
  • Approximate statistics based on the first 70 samples:
    query positive negative
    type string string string
    modality text text text
    details
    • min: 72 tokens
    • mean: 99.54 tokens
    • max: 145 tokens
    • min: 392 tokens
    • mean: 447.49 tokens
    • max: 512 tokens
    • min: 376 tokens
    • mean: 454.26 tokens
    • max: 512 tokens
  • Samples:
    query positive negative
    In 2017, Ball's Beverage Packaging, Europe segment accounted for 21% of consolidated net sales, producing 27 billion containers with a 42% market share in Europe. The Food and Aerosol Packaging segment contributed 10% of 2017 net sales, while the Aerospace segment accounted for 9%, with 98% of its sales derived from federal government contracts. sult of the inclusion of provisions in most metal beverage container sales contracts to pass through aluminum ingot price changes, as well as through the use of derivative instruments. Beverage Packaging, Europe, Segment The beverage packaging, Europe, segment, which accounted for 21 percent of Ball’s consolidated net sales in 2017. Our European operations consist of 20 facilities throughout Europe. The European beverage container market is approximately 64 billion containers, including Russia, and we are the largest producer with an estimated 42 percent of European shipments. The European market is highly regional in terms of sales growth rates and packaging mix. Four companies manufacture substantially all of the metal beverage containers in Europe. Our European beverage facilities produced 27 billion beverage containers in 2017, the vast majority of which were produced from aluminum. Historically, sales volumes of metal beverage containers in Europe tend to be highest during the per... have a significant amount of goodwill recorded on our consolidated balance sheet as of December 31, 2020. We are required at least annually to test the recoverability of goodwill. The recoverability test of goodwill is based on the current fair value of our identified reporting units. Fair value measurement requires assumptions and estimates of many critical factors, including revenue and market growth, operating cash flows and discount rates. If general market conditions deteriorate in portions of our business, we could experience a significant decline in the fair value of our reporting units. This decline could lead to an impairment of all or a significant portion of the goodwill balance, which could materially affect our U.S. GAAP net earnings and net assets. ​ If the investments in Ball’s pension plans, or in the multi-employer pension plans in which Ball participates, do not perform as expected, we may have to contribute additional amounts to the plans, which would otherwise be av...
    ADP provides a comprehensive suite of HCM products including RUN Powered by ADP, ADP Workforce Now, and ADP Vantage HCM, serving various business sizes with payroll, benefits, talent management, and workforce management solutions. The company processes payroll for over 23 million workers in the U.S., moved more than $2.3 trillion in client funds in its last fiscal year, and manages compliance, insurance, and retirement services for over 100,000 employers. lutions. Our premier suite of HCM products offers complete solutions that assist employers of all types and sizes in all stages of the employment cycle, from recruitment to retirement. Our suite of HCM solutions are powered by our strategic, cloud-based, award-winning platforms: • RUN Powered by ADP combines a software platform for small business payroll, HR management and tax compliance administration, with 24/7 service and support from our team of small business experts. RUN Powered by ADP also integrates with other ADP solutions, such as workforce management, workers’ compensation insurance premium payment plans, and retirement plan administration systems. • ADP Workforce Now is a flexible HCM solution used across mid-sized and large businesses in North America to manage their employees. More businesses use ADP Workforce Now in North America than any other HCM solution designed for both mid-sized and large businesses. • ADP Vantage HCM is a solution for large enterprises in the Unit... d accrued expenses 2,444 Other current liabilities 190 Other liabilities 451 Total liabilities related to real estate and other assets held for sale 104,627 Net real estate and other assets held for sale $ 25,872 11. Real Estate We provide build-to-suit services for our clients and also develop or purchase certain projects which we intend to sell to institutional investors upon project completion or redevelopment. Therefore, we have ownership of real estate until such projects are sold or otherwise disposed. Certain real estate assets secure the outstanding balances of underlying mortgage or construction loans. Our real estate is reported in our Development Services and Global Investment Management segments and consisted of the following (dollars in thousands): Land Buildings and Improvements Other Total At December 31, 2013 Real estate under development (current) $ 667 $ 18,466 $ — $ 19,133 Real estate under development (non-current) 822 — — 822 Real estate held for investment 24,717 ...
    AIG's 2020 Form 10-K details critical accounting estimates for various insurance lines, highlighting that a 5% increase in loss cost trends for U.S. Excess Casualty would increase loss reserves by $1,000 million, while a 6-month slower loss development factor would increase reserves by $1,050 million. The report also notes that for U.S. Run-Off Long Tail Insurance, the tail factor beyond 30 years could vary by 10 percent above or below the 2020 reserve review estimates. s are also an important assumption, but are less critical than for U.S. Excess Casualty. Because these classes are written on a claims made basis, the loss reporting and development tail is much shorter than for U.S. Excess Casualty. However, the high severity nature of the losses does create the potential for significant deviations in loss development patterns from one year to the next. Similar to U.S. Excess Casualty, after evaluating the historical loss development factors from prior accident years since the early 1990s, in our judgment, it is reasonably likely that actual loss development factors could change by an amount equivalent to a shift by six months from those actually utilized in the year-end 2020 reserve review. Europe Casualty and Financial Lines Similar to U.S. business, European Casualty and Financial Lines can be significantly impacted by loss cost trends and changes in loss development factors. The variation in such factors can differ significantly by product and reg... of risks that aggregate and/or correlate across divisions; • identify, analyze, and mitigate significant external risks that could impair the financial condition of ACE and/or hinder its business objectives; • coordinate accumulation guidelines and actual exposure relative to guidelines, risk codes, and other risk processes; • provide analysis and maintain accumulation and economic capital and information systems that enable business leaders to make appropriate and consistent risk/return decisions; • identify and assess emerging risk issues; and • develop and communicate to the Company’s business lines consistent risk management processes. The Company’s Enterprise Risk Management Board (ERMB) reports to and assists the Chief Executive Officer in the oversight and review of ACE’s ERM framework. The ERMB is chaired by the Company’s Chief Risk Officer and Chief Actuary. The ERMB oversees and monitors the processes and guidelines used to manage insurance risk, financial risk, strategic ris...
  • Loss: CachedMultipleNegativesRankingLoss with these parameters:
    {
        "scale": 20.0,
        "similarity_fct": "cos_sim",
        "mini_batch_size": 4,
        "gather_across_devices": false,
        "directions": [
            "query_to_doc"
        ],
        "partition_mode": "joint",
        "hardness_mode": null,
        "hardness_strength": 0.0
    }
    

Evaluation Dataset

finmodernbert-triplets

  • Dataset: finmodernbert-triplets
  • Size: 15 evaluation samples
  • Columns: query, positive, and negative
  • Approximate statistics based on the first 15 samples:
    query positive negative
    type string string string
    modality text text text
    details
    • min: 75 tokens
    • mean: 100.13 tokens
    • max: 117 tokens
    • min: 396 tokens
    • mean: 447.6 tokens
    • max: 512 tokens
    • min: 377 tokens
    • mean: 451.87 tokens
    • max: 512 tokens
  • Samples:
    query positive negative
    The Company reported a goodwill balance of $3.50 billion as of December 31, 2022, and 2021, and utilizes various valuation techniques, including discounted cash flow and market-to-book multiples, to estimate the fair value of its goodwill reporting units. The passage also details the acquisition of National General Holdings Corp. on January 4, 2021, and the accounting treatments for credit loss allowances, intangible assets, and insurance claim reserves. des, but is not limited to, as applicable, the remaining payment terms of the security, prepayment speeds, the financial condition and future earnings potential of the issue or issuer, expected defaults, expected recoveries, the value of underlying collateral, origination vintage year, geographic concentration of underlying collateral, available reserves or escrows, current subordination levels, third-party guarantees and other credit enhancements. Other information, such as industry analyst reports and forecasts, sector credit ratings, financial condition of the bond insurer for insured fixed income securities, and other market data relevant to the realizability of contractual cash flows, may also be considered. The estimated fair value of collateral will be used to estimate recovery value if we determine that the security is dependent on the liquidation of collateral for ultimate settlement. If we do not expect to receive cash flows sufficient to recover the entire amortized cost bas... s, and profitability. While we seek to take advantage of the efficiencies and opportunities that large retail customers can offer, they often seek lower pricing and purchase volume flexibility, offer competing own-label products, and represent a large number of other competing products. If the buying power of these large retail customers continues to increase, it could negatively affect our financial results. We might not succeed in our strategies for acquisitions and dispositions. From time to time, we acquire or invest in additional brands or businesses. We expect to continue to seek acquisition and investment opportunities that we believe will increase long-term shareholder value, but we may not be able to find and purchase brands or businesses at acceptable prices and terms. Acquisitions involve risks and uncertainties, including potential difficulties integrating acquired brands and personnel; the possible loss of key customers or employees most knowledgeable about the acquired bu...
    Abbott reported 2019 gross profit margins of 52.5% on $2.4 billion in R&D expenses, driven by growth in Structural Heart (MitraClip), Electrophysiology, and Heart Failure (HeartMate 3) segments. The company also detailed the financial impact of the $23.6 billion St. Jude Medical acquisition and reported flat SG&A expenses in 2019. of clip sizes and facilitation of procedure assessment in real time to offer doctors further options when treating mitral valve disease. In 2018, growth in Structural Heart was driven by several product areas including MitraClip and the AMPLATZER ® PFO occluder, a device designed to close a hole-like opening in the heart. In September 2018, Abbott announced positive clinical results from its COAPT study, which demonstrated that MitraClip improved survival and clinical outcomes for select patients with functional MR. In September 2019, Abbott announced additional data from its COAPT trial that shows that MitraClip is projected to increase life expectancy and quality of life compared to guideline-directed medical therapy alone in heart failure patients with secondary MR. In 2019, the growth in Electrophysiology revenue reflects higher sales of cardiac diagnostic and ablation catheters in both the U.S. and internationally. In January 2019, Abbott announced U.S. FDA approval of its TactiCa... icipants the convenient choice of picking up their 90-day supply of maintenance medications at any CVS/pharmacy store or obtaining them through mail order in either case at the cost of mail for both the payer and the plan participant); Bridge Supply (which enables eligible plan participants to avoid gaps in care while waiting for their medications to arrive in the mail by obtaining a bridge supply of their prescriptions at any CVS/pharmacy store at no additional charge); and a new ExtraCare ® Health Card program (which offers discounts to eligible plan participants on certain Flexible Spending Account-eligible and over-the-counter health care products sold in any of our CVS/pharmacy stores). We are also creating new compliance and persistency programs designed to ensure that patients take their medications in the correct manner as well as enhanced disease management programs that are targeted at managing chronic disease states. In addition, we are working with our clients to (i) decrea...
    Bank of America is subject to extensive oversight by U.S. and international regulators, including the SEC, CFTC, and the Federal Reserve, which impose capital, liquidity, and transaction limits on its banking and non-bank subsidiaries. The Corporation's ability to pay dividends and repurchase stock is contingent on maintaining regulatory capital levels, with its stress capital buffer increasing to 3.4% and its G-SIB surcharge expected to rise to 3.0% on January 1, 2024. ion by the SEC, Financial Industry Regulatory Authority and New York Stock Exchange, among others; our futures commission merchant subsidiaries supporting commodities and derivatives businesses in the U.S. are subject to regulation by and supervision of the U.S. Commodity Futures Trading Commission (CFTC), National Futures Association, the Chicago Mercantile Exchange, and in the case of the Banks, certain banking regulators; our insurance activities are subject to licensing and regulation by state insurance regulatory agencies; and our consumer financial products and services are regulated by the Consumer Financial Protection Bureau (CFPB). Our non-U.S. businesses are also subject to extensive regulation by various non-U.S. regulators, including governments, securities exchanges, prudential regulators, central banks and other regulatory bodies, in the jurisdictions in which those businesses operate. For example, our financial services entities in the United Kingdom (U.K.), Ireland and ... e corrections, increasing consumer and corporate leverage, increases in corporate bond spreads, rising or elevated unemployment levels, rising or elevated inflation, fluctuations in foreign exchange or interest rates, as well as widespread health emergencies or pandemics, extreme weather events and the impacts of climate change and domestic and global efforts to transition to a low-carbon economy. Significant economic or market stresses and disruptions typically have a negative impact on the business environment and financial markets, which could impact the underlying credit quality of our borrowers, counterparties and assets. Property value declines or asset price corrections could increase the risk of borrowers or counterparties defaulting or becoming delinquent in their obligations to us, and could decrease the value of the collateral we hold, which could increase credit losses. Credit risk could also be magnified by lending to leveraged borrowers or declining asset prices, includin...
  • Loss: CachedMultipleNegativesRankingLoss with these parameters:
    {
        "scale": 20.0,
        "similarity_fct": "cos_sim",
        "mini_batch_size": 4,
        "gather_across_devices": false,
        "directions": [
            "query_to_doc"
        ],
        "partition_mode": "joint",
        "hardness_mode": null,
        "hardness_strength": 0.0
    }
    

Training Hyperparameters

Non-Default Hyperparameters

  • per_device_train_batch_size: 16
  • num_train_epochs: 10
  • learning_rate: 2e-05
  • warmup_steps: 0.1
  • bf16: True
  • per_device_eval_batch_size: 16
  • load_best_model_at_end: True
  • seed: 777
  • data_seed: 777
  • prompts: {'query': '', 'positive': '', 'negative': ''}
  • batch_sampler: no_duplicates

All Hyperparameters

Click to expand
  • per_device_train_batch_size: 16
  • num_train_epochs: 10
  • max_steps: -1
  • learning_rate: 2e-05
  • lr_scheduler_type: linear
  • 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: 1
  • 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: 777
  • data_seed: 777
  • 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: {'query': '', 'positive': '', 'negative': ''}
  • batch_sampler: no_duplicates
  • multi_dataset_batch_sampler: proportional
  • router_mapping: {}
  • learning_rate_mapping: {}
  • warmup_ratio: None

Training Logs

Epoch Step Training Loss Validation Loss triplet-eval_cosine_accuracy
-1 -1 - - 0.8667
1.0 5 0.3134 0.8631 0.8667
2.0 10 0.2242 0.7348 0.8667
-1 -1 - - 0.8667
  • The bold row denotes the saved checkpoint.

Training Time

  • Training: 40.4 seconds

Framework Versions

  • Python: 3.13.16
  • Sentence Transformers: 5.6.0
  • Transformers: 5.22.1
  • PyTorch: 2.14.1+cu130
  • Accelerate: 1.14.0
  • Datasets: 4.8.5
  • Tokenizers: 0.23.3

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",
}

CachedMultipleNegativesRankingLoss

@misc{gao2021scaling,
    title={Scaling Deep Contrastive Learning Batch Size under Memory Limited Setup},
    author={Luyu Gao and Yunyi Zhang and Jiawei Han and Jamie Callan},
    year={2021},
    eprint={2101.06983},
    archivePrefix={arXiv},
    primaryClass={cs.LG}
}
Downloads last month
-
Safetensors
Model size
0.6B params
Tensor type
BF16
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for finvector/Test_bge_raw_finmodern_tri_merged

Base model

BAAI/bge-m3
Finetuned
(601)
this model

Papers for finvector/Test_bge_raw_finmodern_tri_merged

Evaluation results