fin-sparse-encoder-doc-v1

Financial Domain Sparse Encoder (doc-v3-gte fine-tuned)

This is a Asymmetric Inference-free SPLADE Sparse Encoder model finetuned from opensearch-project/opensearch-neural-sparse-encoding-doc-v3-gte on the financial-filings-sparse-retrieval-training dataset using the sentence-transformers library. It maps sentences & paragraphs to a 30522-dimensional sparse vector space and can be used for semantic search and sparse retrieval.

Model Details

Model Description

• Model Type: Asymmetric Inference-free SPLADE Sparse Encoder
• Base model: opensearch-project/opensearch-neural-sparse-encoding-doc-v3-gte <!-- at revision 1646fef40807937e8e130c66d327a26421c408d5 -->
• Maximum Sequence Length: 512 tokens
• Output Dimensionality: 30522 dimensions
• Similarity Function: Dot Product
• Training Dataset:
  • financial-filings-sparse-retrieval-training
• Language: en
• License: apache-2.0

Model Sources

• Documentation: Sentence Transformers Documentation
• Documentation: Sparse Encoder Documentation
• Repository: Sentence Transformers on GitHub
• Hugging Face: Sparse Encoders on Hugging Face

Full Model Architecture

SparseEncoder(
  (0): Router(
    (sub_modules): ModuleDict(
      (query): Sequential(
        (0): SparseStaticEmbedding({'frozen': False}, dim=30522, tokenizer=TokenizersBackend)
      )
      (document): Sequential(
        (0): MLMTransformer({'max_seq_length': 512, 'do_lower_case': False, 'architecture': 'NewForMaskedLM'})
        (1): SpladePooling({'pooling_strategy': 'max', 'activation_function': 'log1p_relu', 'word_embedding_dimension': 30522})
      )
    )
  )
)

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 SparseEncoder

# Download from the 🤗 Hub
model = SparseEncoder("oneryalcin/fin-sparse-encoder-doc-v1")
# Run inference
queries = [
    "How much did the company charge for depreciation of tangible assets in 2025?",
]
documents = [
    "CoolerAid Holdings Ltd\nNotes to the Consolidated Financial Statements (continued)\nYear ended 31 March 2025\n\n5. Operating profit\nOperating profit or loss is stated after charging/crediting\n\n|| 2025| 2024|\n|------------------------------------------------|---------|-----------|\n| Depreciation of tangible assets| 870,340 | 752,425   |\n| Loss/(gains) on disposal of tangible assets| 1,535   | (173,401) |\n| Impairment of trade debtors| 32,465  | 15,860|\n| Operating lease rentals| 112,292 | 139,908   |\n\n6. Auditor's remuneration\n\n|| 2025  | 2024  |\n|--------------------------------------------------------------|-------|-------|\n| Fees payable for the audit of the consolidated financial statements | 5,000 | 5,000 |\n\n7. Staff costs\nThe average number of persons employed by the group during the year, including the directors, amounted to:\n\n|| 2025 | 2024 |\n|---------------------|------|------|\n| Distribution staff  | 44   | 42   |\n| Administrative staff| 14   | 13   |\n| Management staff| 2| 2|\n| Directors| 4| 4|\n|| 64   | 61   |\n\nThe aggregate payroll costs incurred during the year, relating to the above, were:\n\n|| 2025| 2024|\n|---------------------|-----------|-----------|\n| Wages and salaries  | 2,828,063 | 2,596,870 |\n| Social security costs| 314,830   | 303,444   |\n| Other pension costs | 79,769| 68,413|\n|| 3,222,662 | 2,968,727 |\n\n8. Directors' remuneration\nThe directors' aggregate remuneration in respect of qualifying services was:\n\n|| 2025   | 2024   |\n|---------------|--------|--------|\n| Remuneration  | 18,880 | 18,880 |\n\n- 20 -",
    'Kenneth Forbes (Holdings) Limited\nNotes to the Financial Statements (continued)\nYear ended 30 April 2025\n\n13. Tangible assets.\n\n| Group and company | Freehold property £ | Plant and machinery £ | Fixtures and fittings £ | Motor vehicles £ | Assets held under the course of construction £ | Total £ |\n|---|---|---|---|---|---|---|\n| Cost |  |  |  |  |  |  |\n| At 1 May 2024 | 3,941,316 | 3,684,443 | 1,326,032 | 294,882 | 41,948 | 9,288,621 |\n| Additions | 570,435 | 395,946 | 187,049 | 194,503 | 106,519 | 1,454,452 |\n| Disposals |  | (317,265) | (214,785) | (164,244) |  | (696,294) |\n| At 30 Apr 2025 | 4,511,751 | 3,763,124 | 1,298,296 | 325,141 | 148,467 | 10,046,779 |\n| Depreciation |  |  |  |  |  |  |\n| At 1 May 2024 | 1,158,100 | 3,257,887 | 904,088 | 156,223 |  | 5,476,298 |\n| Charge for the year | 133,789 | 160,512 | 101,477 | 78,993 |  | 474,771 |\n| Disposals |  | (317,265) | (214,785) | (133,585) |  | (665,635) |\n| At 30 Apr 2025 | 1,291,889 | 3,101,134 | 790,780 | 101,631 |  | 5,285,434 |\n| Carrying amount |  |  |  |  |  |  |\n| At 30 Apr 2025 | 3,219,862 | 661,990 | 507,516 | 223,510 | 148,467 | 4,761,345 |\n| At 30 Apr 2024 | 2,783,216 | 426,556 | 421,944 | 138,659 | 41,948 | 3,812,323 |\n\nFreehold land amounting to £475,010 (2024: £475,010) is not depreciated.\nAll assets are held by the company but used by the group undertakings or associates.\n\nTangible assets held at valuation\n\nIn respect of tangible assets held at valuation, aggregate cost, depreciation and comparable carrying\namount that would have been recognised if the assets had been carried under the historical cost model are\nas follows:\n\nGroup and company\n\n|  | Freehold property £ |\n|---|---|\n| At 30 April 2025 |  |\n| Aggregate cost | 3,854,887 |\n| Aggregate depreciation | (1,963,225) |\n| Carrying value | 1,891,662 |\n| At 30 April 2024 |  |\n| Aggregate cost | 3,854,887 |\n| Aggregate depreciation | (1,700,013) |\n| Carrying value | 2,154,874 |\n\n1\n-24-\n1',
    'WHOCANFIXMYCAR.COM LTD\nNOTES TO THE FINANCIAL STATEMENTS\nFOR THE YEAR ENDED 31 MARCH 2025\n\n7. Tangible fixed assets\n\n| | Fixtures and fittings £ | Computer equipment £ | Right of use assets £ | Total £ | £ |\n|---|---|---|---|---|---|\n| **Cost** | | | | | |\n| At 1 April 2024 | 61,966 | 131,878 | - | 193,844 |  |\n| Additions | - | 3,187 | 287,367 | 290,554 |  |\n| At 31 March 2025 | 61,966 | 135,065 | 287,367 | 484,398 |  |\n| **Depreciation** | | | | | |\n| At 1 April 2024 | 61,966 | 99,890 | - | 161,856 |  |\n| Charge for the year | - | 12,982 | 82,657 | 95,639 |  |\n| At 31 March 2025 | 61,966 | 112,872 | 82,657 | 257,495 |  |\n| **Net book value** | | | | | |\n| At 31 March 2025 | - | 22,193 | 204,710 | 226,903 |  |\n| At 31 March 2024 | - | 31,988 | - | 31,988 |  |\n\nThe net book value of owned and leased assets included as "Tangible fixed assets" in the Balance Sheet is as follows:\n\n| | 2025 £ | 2024 £ |\n|---|---|---|\n| Tangible fixed assets owned | 22,193 | 31,988 |\n| Right-of-use tangible fixed assets | 204,710 | - |\n| | 226,903 | 31,988 |\n\nInformation about right-of-use assets is summarised below:\n\nNet book value\n\n| | 2025 £ | 2024 £ |\n|---|---|---|\n| Office and computer equipment | 204,710 | - |\n| | 204,710 | - |\n\nPage 11',
]
query_embeddings = model.encode_query(queries)
document_embeddings = model.encode_document(documents)
print(query_embeddings.shape, document_embeddings.shape)
# [1, 30522] [3, 30522]

# Get the similarity scores for the embeddings
similarities = model.similarity(query_embeddings, document_embeddings)
print(similarities)
# tensor([[18.4437, 21.8515, 17.4189]])

<!-- ### 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

Domain Evaluation: Financial Document Retrieval

Evaluated on 2,028 held-out financial test examples (SEC filings + earnings call transcripts). Each example has 1 query + 1 positive + up to 7 hard negatives. We re-rank candidates per query using sparse dot-product similarity.

The fine-tuned model ranks the correct document first (median_rank=1.0) more often than not, compared to position 2 for the base model. Sparsity also improved: corpus active dims decreased from ~2,500 to ~1,666, meaning faster inverted index lookups at inference time.

Metrics

Sparse Information Retrieval

• Datasets: NanoNFCorpus, NanoSciFact and NanoFiQA2018
• Evaluated with <code>SparseInformationRetrievalEvaluator</code>

Sparse Nano BEIR

• Dataset: NanoBEIR_mean
• Evaluated with <code>SparseNanoBEIREvaluator</code> with these parameters:

  {
      "dataset_names": [
          "nfcorpus",
          "scifact",
          "fiqa2018"
      ],
      "dataset_id": "sentence-transformers/NanoBEIR-en"
  }
  

<!-- ## 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

financial-filings-sparse-retrieval-training

• Dataset: financial-filings-sparse-retrieval-training at 23e44ab
• Size: 18,247 training samples
• Columns: <code>query</code>, <code>positive</code>, <code>negative_0</code>, <code>negative_1</code>, <code>negative_2</code>, <code>negative_3</code>, <code>negative_4</code>, <code>negative_5</code>, and <code>negative_6</code>
• Approximate statistics based on the first 1000 samples:

  	query	positive	negative_0	negative_1	negative_2	negative_3	negative_4	negative_5	negative_6
  type	string	string	string	string	string	string	string	string	string
  details	<ul><li>min: 9 tokens</li><li>mean: 20.51 tokens</li><li>max: 79 tokens</li></ul>	<ul><li>min: 51 tokens</li><li>mean: 331.12 tokens</li><li>max: 512 tokens</li></ul>	<ul><li>min: 54 tokens</li><li>mean: 360.35 tokens</li><li>max: 512 tokens</li></ul>	<ul><li>min: 56 tokens</li><li>mean: 357.16 tokens</li><li>max: 512 tokens</li></ul>	<ul><li>min: 2 tokens</li><li>mean: 303.81 tokens</li><li>max: 512 tokens</li></ul>	<ul><li>min: 2 tokens</li><li>mean: 263.98 tokens</li><li>max: 512 tokens</li></ul>	<ul><li>min: 2 tokens</li><li>mean: 221.87 tokens</li><li>max: 512 tokens</li></ul>	<ul><li>min: 2 tokens</li><li>mean: 193.47 tokens</li><li>max: 512 tokens</li></ul>	<ul><li>min: 2 tokens</li><li>mean: 166.03 tokens</li><li>max: 512 tokens</li></ul>

• Samples:

  query	positive	negative_0	negative_1	negative_2	negative_3	negative_4	negative_5	negative_6
  <code>What was the actuarial gain on defined benefit pension plans for the year 2021?</code>	<code>2021 £'000 2020 £'000 Loss for the financial year (9,731) (50,454) Other comprehensive income/(expense) for the financial period Actuarial gain/(loss) on defined benefit pension plans 31,200 (4,400) Deferred tax impact of actuarial gain/(loss) (10,920) 1,540 Other comprehensive income/(expense) 20,280 (2,860) Total comprehensive income/(expense) for the financial period 10,549 (53,314)<br><br>STATEMENT OF COMPREHENSIVE INCOME FOR THE YEAR ENDED 31 DECEMBER 2021<br><br>11</code>	<code>Defined benefit pension plans 2022 2021 Equities 12% 26% Bonds 56% 31% Liability driven investment 32% 39% Other - 4% 100% 100%<br><br>Defined benefit pension plans 2022 €'000 2021 €'000 Actuarial losses from changes in financial assumptions 36 391 Actuarial gain/(losses) 3,629 (4,830) 3,665 (4,439)<br><br>markdown Dunlop International Europe Limited Notes to the Financial Statements - continued for the year ended 31 December 2022<br><br>Defined benefit pension plans 2022 €'000 2021 €'000 Opening fair value of scheme assets 90,460 85,266 Contributions by employer 1,129 1,189 Expected return 1,697 1,183 Actuarial (losses) (33,630) (591) Benefits paid (3,397) (3,408) Exchange differences on foreign plans (4,631) 6,821 51,628 90,460<br><br>Changes in the fair value of scheme assets are as follows:<br><br>Page 24 <br><br>The major categories of scheme assets as a percentage of total scheme assets are as follows:<br><br>Changes in the present value of the defined benefit obligation are as follows:<br><br>Defined benefit pension pla...</code>	<code>The components of net periodic pension benefit cost recognized in our Consolidated Statements of Operations for the periods presented are as follows:<br><br>Years Ended December 31, 2021 Projected benefit obligation, beginning of year $ 97,740 Service cost $ 1,282 Interest cost $ 1,452 Actuarial (gain) loss $ (8,682) Benefits paid $ (2,010) Translation adjustment $ (4,006) Projected benefit obligation, end of year $ 85,776 Fair value of plan assets, beginning of year $ 17,293 Actual return on plan assets $ 641 Contributions $ 1,775 Benefits paid $ (1,112) Actuarial gain $ 71 Translation adjustment $ (147) Fair value of plan assets, end of year $ 18,521 Funded status of plan $ (67,255)<br><br>Our projected benefit obligation and plan assets for defined benefit pension plans and the related assumptions used to determine the related liabilities are as follows:<br><br>Defined Benefit Plan<br><br>We maintain defined benefit pension plans for certain of our non-U.S. employees in the U.K., Germany, and Philippines. ...</code>	<code></code>	<code></code>	<code></code>	<code></code>	<code></code>
  <code>What is the interest rate for the GBP short term loan with Canadian Natural Resources Limited?</code>	<code>CREDITORS: amounts falling due within one year<br><br>33<br><br>DEBTORS<br><br>The carrying amount of debtors is a reasonable approximation to fair value. Trade debtors and other receivables are not overdue as payment terms have not been exceeded. The expected credit loss on the trade debtor's balance was negligible and therefore no adjustment has been applied.<br><br>Other creditors includes £4.9 million (2022 - £4.3 million) in respect of future share options (note 21). The short term loan with CNR International (U.K.) Developments Limited expired on 31 December 2023. Interest was generated at a rate of Secured Overnight Financing Rate (SOFR) + 1.75%.<br><br>Amounts owed by group undertakings are unsecured and repayable on demand. The GBP short term loan with Canadian Natural Resources Limited generates interest at a rate of Sterling Overnight Index Average (SONIA) + 1.15% per annum. Trading balances are interest free. Management considered the expected credit loss on amounts due from Group undertakings at 31 Dec...</code>	<code>for the year ended 31 December 2022<br><br>The share options creditor relates to amounts payable to the ultimate parent Canadian Natural Resources Limited relating to employee options to purchase stock in the aforementioned company. The provision is based on the specifics of the agreed plan with Canadian Natural Resources Limited. The current portion of this totalling £4.3 million (2021 - £3.7 million) is included in creditor amounts falling due within one year.<br><br>34<br><br>The Company settled an intercompany term loan on 24 October 2022. The amount drawn down by the Company at 24 October 2022 was US$440.0 million (2021 - US$440.0 million). Interest was charged at US$ LIBOR + 2.8175% per annum on amounts drawn down from this facility until settlement.<br><br>2022 (£'000) 2021 (£'000) Share options 3,550 4,276 Lease liabilities (note 13) 4,786 6,428 Total 8,336 10,704<br><br>The short term loan with CNR International (U.K.) Developments Limited generates interest at a rate of USS LIBOR + 1.75%.<br><br>18. CREDITORS: ...</code>	<code>18 Cash and cash equivalents<br><br>Group 31 Dec 2021 (£000) Group 31 Dec 2020 (£000) Company 31 Dec 2021 (£000) Company 31 Dec 2020 (£000) Sterling 19,641 37,349 19,498 37,347 United States Dollar 8,574 6,826 — — Euros 361 — — — Canadian Dollar 227 364 — — Polish Zloty 225 283 — — Singapore Dollar 29 — — — Japanese Yen 7 — — — Total 29,064 44,822 19,498 37,347<br><br>Cash and cash equivalents are denominated in the following currencies:<br><br>On 14 October 2021, the Group and Company entered into a loan agreement with Bank Of Ireland Group plc consisting of a £10 million term loan in addition to a revolving credit facility of £10 million. The loan is secured on the assets of the Group. Operating covenants are limited to the Group’s net debt leverage and interest cover. The term loan is repayable over five years with an initial 12-month repayment holiday followed by annual capital repayments of £1,250,000. At the end of the term, a bullet payment of £5 million is due. The loan is denominated in Pound S...</code>	<code></code>	<code></code>	<code></code>	<code></code>	<code></code>
  <code>What is the amount for Charges à imputer relative au personnel?</code>	<code>Ventilation de la rubrique 492/3 du passif si celle-ci représente un montant important<br><br>COMPTES DE RÉGULARISATION<br><br>Description Montant Charges à imputer relative au personnel 128.084.824,70 Charges à imputer: intérêts courus et non échus 37.536.987,21 Charges à imputer diverses 12.774.405,94 Charges à imputer: protocoles et conventions avec autres opérateurs et réseaux 7.240.507,84 Produits à reporter divers 7.841.755,07 Produits à reporter relatifs au trafic 116.123.554,80 Produis à reporter: Hello Belgium Railpass 21.396.182,64 Produis à reporter: NPV 20.088.866,29 Produits à reporter: financements alternatifs 15.285.835,38<br><br>72<br><br>First - C-Cap2022 - 38 / 82</code>	<code>markdown N° 0203.430.576 C-Cap 6.9<br><br>Charges à imputer diverses Exercice Charges à imputer: protocoles et conventions avec autres opérateurs et reseaux 2.101.406,29 Charges à imputer relatives au personnel 1.378.996,29 Charges à imputer: intérêts courus et non échus 139.424.041,35 Produits à reporter divers 39.103.099,47 Produits à reporter relatifs au trafic 6.816.893,14 Produits à reporter: financements altematifs 146.853.054,23 Produis à reporter: NPV 9.576.751,43 15.185.940,48<br><br>71 Rapport annuel SNCB 2023 <br><br>COMPTES DE RÉGULARISATION<br><br>Ventilation de la rubrique 492/3 du passif si celle-ci représente un montant important</code>	<code>DETTES FISCALES, SALARIALES ET SOCIALES (€)<br><br>CODES BNB 2024 VENTILATION DE LA RUBRIQUE 492/3 DU PASSIF SI CELLE-CI REPRÉSENTE UN MONTANT IMPORTANT Intérêts crédits KBC à imputer Autres charges à imputer<br><br>ÉTAT DES DETTES (€)<br><br>COMPTES DE REGULARISATION (€)<br><br>12.4 Etat des dettes et comptes de régularisation du passif<br><br>CODES BNB 2024 VENTILATION DES DETTES A L'ORIGINE A PLUS D'UN AN, EN FONCTION DE LEUR DUREE RESDIEUELLE Dettes à plus d'un an échéant dans l'année Dettes financières 8801 Etablissements de crédit 8841 Total des dettes à plus d'un an échéant dans l'année (42) Dettes ayant plus d'un an mais 5 ans au plus à courir Dettes financières 8802 Etablissement de crédit 8842 Total des dettes ayant plus d'un an mais 5 ans au plus à courir 8912<br><br>12.5 Résultats d'exploitation (en milliers €)<br><br>DETTES GARANTIES (€)<br><br>CODES BNB 2024 Rémunérations et charges sociales Autres dettes salariales et sociales 9077<br><br>```markdown<br><br>CODES BNB 2023 2024 CHARGES D'EXPLOITATION Travailleurs pour lesquels la ...</code>	<code>Résultats financiers Charges financières récurrentes Ventilation des autres charges financières<br><br>11 Dettes fiscales, salariales et sociales<br><br>2022 2021 Impôts (rubriques 450/3 et 179 du passif) Dettes fiscales échues - - Dettes fiscales non échues - - Dettes fiscales estimées - - Rémunérations et charges sociales (rubriques 454/9 et 179 du passif) Dettes échues envers l'Office National de Sécurité Sociale - - Autres dettes salariales et sociales 28.500 25.000<br><br>FINANCIÈRE DE TUBIZE – RAPPORT FINANCIER ANNUEL 2022<br><br>Comptes de régularisation<br><br>FINANCIÈRE DE TUBIZE - RAPPORT ANNUEL 2022 40 41<br><br>2022 2021 Ventilation de la rubrique 492/3 du passif si celle-ci représente un montant important Charges à imputer: intérêts 345.843 40.556 Charges à imputer: commission de réservation 107.126 76.667<br><br>2022 2021 Charges d'exploitation Travailleurs pour lesquels la société a introduit une déclaration DIMONA ou qui sont inscrits au registre général du personnel - - Nombre total à la date de clôture - - Ef...</code>	<code>ANNEXES DES COMPTES DE LA SOCIÉTÉ AU 31 MAI 2025<br><br>SITUATION FISCALE DIFFEREE ET LATENTE<br><br>| Accroissements de la dette future d'impôt | Montant |<br>|---|---|<br>| Impôt dû sur provisions réglementées: | |<br>| Provisions pour hausse de prix | |<br>| Provisions pour fluctuation des cours | |<br>| Provisions pour investissements | |<br>| Amortissements dérogatoires | 3 548 |<br>| Subventions d'investissement | |<br>| TOTAL ACCROISSEMENTS | 3 548 |<br><br>| Allègements de la dette future d'impôt | Montant |<br>|---|---|<br>| Impôt payé d'avance sur: | |<br>| Charges non déductibles temporairement (à déduire l'année suivante) | 1 200 |<br>| Congés payés | |<br>| Participation des salariés | 53 |<br>| Autres | |<br>| A déduire ultérieurement | |<br>| Provisions pour propre assureur | |<br>| Autres | |<br>| TOTAL ALLÈGEMENTS | 1 254 |<br><br>SITUATION FISCALE DIFFÉRÉE NETTE<br>2 294<br><br>IMPÔT DÙ SUR: Plus-values différées<br>43 436<br><br>CREDIT A IMPUTER SUR: Déficits reportables<br><br>CREDIT A IMPUTER SUR: Moins-values à long terme<br><br>SITUATION FISCALE LATENTE NETT...</code>	<code>d'un mois au plus<br><br>Titres à revenu fixe émis par des établissements de crédit<br><br>AUTRES PLACEMENTS DE TRÉSORERIE<br><br>Autres placements de trésorerie non repris ci-avant<br><br>Codes Exercice Exercice précédent 51 8681 8682 8683 52 18.088.896,34 55.716.775,84 8684 53 226.519.496,99 136.305.308,18 8686 85.000.000,00 801.112,99 8687 1.599.690,84 8688 139.919.806,15 135.504.195,19 8689<br><br>Actions, parts et placements autres que placements à revenu fixe<br><br>Titres à revenu fixe<br><br>Actions et parts - Montant non appelé<br><br>Avec une durée résiduelle ou de préavis<br><br>Métaux précieux et œuvres d'art<br><br>Ventilation de la rubrique 490/1 de l'actif si celle-ci représente un montant important<br><br>PLACEMENTS DE TRÉSORERIE ET COMPTES DE RÉGULARISATION DE L'ACTIF<br><br>64 Rapport annuel SNCB 2023<br><br>Actions et parts - Valeur comptable augmentée du montant non appelé<br><br>Comptes à terme détenus auprès des établissements de crédit<br><br>de plus d'un mois à un an au plus<br><br>COMPTES DE RÉGULARISATION<br><br>Exercice Charges à reporter: redevance infrastru...</code>	<code>4.6. Accroissements et allégements de la dette future d'impôt<br><br>Les éléments entraînant un décalage d'imposition conduisent à un accroissement de la dette future d'impôt de 21 278K€ calculé au taux de 25.82%.<br><br>La situation fiscale latente s'analyse comme suit :<br><br>| Base de calcul | Montants en K€ |<br>|---|---|<br>| BASE D'IMPOT SUR : | |<br>| Provisions réglementées : | |<br>| - Ecart de conversion Actif | 0 |<br>| - Ecart de conversion Passif | -4 |<br>| - Provision pour investissements | |<br>| - Amortissements dérogatoires | 94 455 |<br>| Subventions d'investissement | 3 283 |<br>| Produits non imposables temporairement : | |<br>| (à réintégrer l'année de leur acquisition) | |<br>| - plafonnement TP | |<br>| TOTAL ACCROISSEMENTS | 97 734 |<br>| BASE D'IMPOT PAYE D'AVANCE SUR : | |<br>| Charges non déductibles temporairement : | |<br>| (à déduire l'année suivante) | |<br>| - Provision pour risques et charges | -928 |<br>| - Provision pour participation | -4 083 |<br>| - Contribution solidarité | -869 |<br>| - Provisions pou...</code>	<code></code>

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

  {
      "loss": "SparseMultipleNegativesRankingLoss(scale=1.0, similarity_fct='dot_score', gather_across_devices=False)",
      "document_regularizer_weight": 3e-05,
      "query_regularizer_weight": 0.0
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• num_train_epochs: 2
• learning_rate: 2e-05
• warmup_steps: 114
• weight_decay: 0.01
• gradient_accumulation_steps: 4
• bf16: True
• tf32: True
• eval_strategy: steps
• dataloader_num_workers: 4
• batch_sampler: no_duplicates
• router_mapping: {'query': 'query', 'positive': 'document', 'negative_0': 'document', 'negative_1': 'document', 'negative_2': 'document', 'negative_3': 'document', 'negative_4': 'document', 'negative_5': 'document', 'negative_6': 'document'}
• learning_rate_mapping: {'sub_modules\.query\..*': 0.001}

All Hyperparameters

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

• per_device_train_batch_size: 8
• num_train_epochs: 2
• max_steps: -1
• learning_rate: 2e-05
• lr_scheduler_type: linear
• lr_scheduler_kwargs: None
• warmup_steps: 114
• optim: adamw_torch_fused
• optim_args: None
• weight_decay: 0.01
• adam_beta1: 0.9
• adam_beta2: 0.999
• adam_epsilon: 1e-08
• optim_target_modules: None
• gradient_accumulation_steps: 4
• 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: True
• 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: trackio
• eval_strategy: steps
• per_device_eval_batch_size: 8
• 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: False
• 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: 4
• dataloader_pin_memory: True
• dataloader_persistent_workers: False
• dataloader_prefetch_factor: None
• 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_backend: None
• ddp_timeout: 1800
• fsdp: []
• fsdp_config: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}
• deepspeed: None
• debug: []
• skip_memory_metrics: True
• do_predict: False
• resume_from_checkpoint: None
• warmup_ratio: None
• local_rank: -1
• prompts: None
• batch_sampler: no_duplicates
• multi_dataset_batch_sampler: proportional
• router_mapping: {'query': 'query', 'positive': 'document', 'negative_0': 'document', 'negative_1': 'document', 'negative_2': 'document', 'negative_3': 'document', 'negative_4': 'document', 'negative_5': 'document', 'negative_6': 'document'}
• learning_rate_mapping: {'sub_modules\.query\..*': 0.001}

</details>

Training Logs

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

Framework Versions

• Python: 3.11.10
• Sentence Transformers: 5.2.3
• Transformers: 5.2.0
• PyTorch: 2.10.0+cu128
• Accelerate: 1.12.0
• Datasets: 4.5.0
• Tokenizers: 0.22.2

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",
}

SpladeLoss

@misc{formal2022distillationhardnegativesampling,
      title={From Distillation to Hard Negative Sampling: Making Sparse Neural IR Models More Effective},
      author={Thibault Formal and Carlos Lassance and Benjamin Piwowarski and Stéphane Clinchant},
      year={2022},
      eprint={2205.04733},
      archivePrefix={arXiv},
      primaryClass={cs.IR},
      url={https://arxiv.org/abs/2205.04733},
}

SparseMultipleNegativesRankingLoss

@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}
}

FlopsLoss

@article{paria2020minimizing,
    title={Minimizing flops to learn efficient sparse representations},
    author={Paria, Biswajit and Yeh, Chih-Kuan and Yen, Ian EH and Xu, Ning and Ravikumar, Pradeep and P{'o}czos, Barnab{'a}s},
    journal={arXiv preprint arXiv:2004.05665},
    year={2020}
}

<!-- ## 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.* -->