This is a sentence-transformers model finetuned from nvidia/llama-nemotron-embed-1b-v2 on the financial-filings-sparse-retrieval-training dataset. It maps sentences & paragraphs to a 2048-dimensional dense vector space and can be used for retrieval.

Model Details

Model Description

Model Type: Sentence Transformer
Base model: nvidia/llama-nemotron-embed-1b-v2
Maximum Sequence Length: 2048 tokens
Output Dimensionality: 2048 dimensions
Similarity Function: Cosine Similarity
Supported Modality: Text
Training Dataset:
- financial-filings-sparse-retrieval-training

Model Sources

Documentation: Sentence Transformers Documentation
Repository: Sentence Transformers on GitHub
Hugging Face: Sentence Transformers on Hugging Face

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': 'LlamaBidirectionalModel'})
  (1): Pooling({'embedding_dimension': 2048, 'pooling_mode': 'mean', '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.

python

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("abanfalvi/finance-llama-nemotron-embed-1b-v2")
# Run inference
queries = [
    'deferred tax assets opening balance inventory loss 2022',
]
documents = [
    'The movements of deferred tax assets and deferred tax liabilities were as follows:\n\nFor the year ended December 31, 2022\n\nd. Deferred tax assets and liabilities\n\nDecember 31 2022 2021 Current tax liabilities Income tax payable $ 473,781 $ 255,744\n\nb. Income tax recognized in other comprehensive income\n\nOpening Balance Recognized in Profit or Loss Recognized in Other Comprehensive Income Closing Balance Deferred tax assets Temporary differences Unrealized loss on write-down of inventories $ 250 $ 8 $ - $ 258 Unrealized employee compensation 241 (241) - - Exchanges difference on foreign operations 73,357 - (23,271) 50,086 Unrealized exchange losses 15,010 (15,010) - - Total $ 88,858 $ (15,243) $ (23,271) $ 50,344 Deferred tax liabilities Temporary differences Unrealized exchange gain $ - $ 2,525 $ - $ 2,525\n\nFor the Year Ended December 31 2022 2021 Deferred tax In respect of the current year $ (23,271) $ (9,510) Translation of foreign operations\n\nc. Current tax assets and liabilities',
    'e. Information on unused loss carryforwards\n\nThe movements of deferred tax assets and deferred tax liabilities were as follows:\n\nFor the year ended December 31, 2021\n\nDecember 31 2022 Loss carryforwards Expiry in 2027 $ - Expiry in 2028 $ 46,056 Expiry in 2029 $ 72,486 Expiry in 2030 $ 97,191 Total $ 215,733\n\nOpening Balance Recognized in Profit or Loss Closing Balance Temporary differences Unrealized exchange gain and loss $ 3 $ (332) Unrealized inventory loss $ 1,113 $ 560 Others $ 402 $ (228) Total $ 1,518 $ -\n\nOpening Balance Recognized in Profit or Loss Closing Balance Temporary differences Unrealized exchange gain and loss $ 113 $ (110) Unrealized inventory loss $ 1,849 $ (736) Others $ (444) $ 846 Total $ 1,518 $ -\n\nUnused Amount Expiry Year $ 46,056 2028 $ 72,486 2029 $ 97,191 2030 Total $ 215,733\n\nd. Items for which no deferred tax assets have been recognized\n\nThe Company offset certain deferred tax assets and deferred tax liabilities which met the offset criteria.\n\nc. Deferred tax assets and liabilities\n\nLoss carryforwards as of December 31, 2022 comprised:\n\nFor the year ended December 31, 2022',
    'PILGRIM FOODSERVICE LIMITED\nNOTES TO THE FINANCIAL STATEMENTS (CONTINUED)\nFOR THE YEAR ENDED 30 APRIL 2025\n\n|   |   | 2025 £ | 2024 £ |\n|---|---|---|---|\n| 17 | Stocks |   |   |\n|   | Finished goods and goods for resale | 4,149,563 | 3,602,784 |\n\n|   |   | 2025 £ | 2024 £ |\n|---|---|---|---|\n| 18 | Debtors |   |   |\n|   | Amounts falling due within one year: |   |   |\n|   | Trade debtors | 4,056,601 | 3,939,861 |\n|   | Corporation tax recoverable | 386,649 | - |\n|   | Other debtors | 51,029 | 285,195 |\n|   | Prepayments and accrued income | 894,513 | 810,422 |\n|   |   | 5,388,792 | 5,035,478 |\n\n|   |   | Notes | 2025 £ | 2024 £ |\n|---|---|---|---|---|\n| 19 | Creditors: amounts falling due within one year |   |   |   |\n|   | Bank loans and overdrafts | 21 | 750,004 | 1,572,186 |\n|   | Obligations under finance leases | 22 | 929,382 | 221,527 |\n|   | Trade creditors |   | 6,972,922 | 5,358,604 |\n|   | Amounts due to group undertakings |   | 49,502 | - |\n|   | Corporation tax |   | 324,698 | 342,382 |\n|   | Other taxation and social security |   | 843,069 | 292,562 |\n|   | Other creditors |   | 3,361,115 | 802,545 |\n|   | Accruals |   |   | 2,269,019 |\n|   |   |   | 13,230,692 | 10,858,915 |\n\n|   |   | Notes | 2025 £ | 2024 £ |\n|---|---|---|---|---|\n| 20 | Creditors: amounts falling due after more than one year |   |   |   |\n|   | Bank loans and overdrafts | 21 | 1,174,972 | 1,925,178 |\n|   | Obligations under finance leases | 22 | 3,026,050 | 927,331 |\n|   | Shareholder loans | 21 | 1,925,125 | 681,083 |\n|   |   |   | 6,126,147 | 3,533,592 |\n\n-26-',
]
query_embeddings = model.encode_query(queries)
document_embeddings = model.encode_document(documents)
print(query_embeddings.shape, document_embeddings.shape)
# [1, 2048] [3, 2048]

# Get the similarity scores for the embeddings
similarities = model.similarity(query_embeddings, document_embeddings)
print(similarities)
# tensor([[0.5039, 0.4746, 0.2891]], dtype=torch.bfloat16)

Evaluation

Metrics

Information Retrieval

Dataset: financial_corpus
Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.663
cosine_accuracy@3	0.787
cosine_accuracy@5	0.837
cosine_accuracy@10	0.897
cosine_precision@1	0.663
cosine_precision@3	0.2623
cosine_precision@5	0.1674
cosine_precision@10	0.0897
cosine_recall@1	0.663
cosine_recall@3	0.787
cosine_recall@5	0.837
cosine_recall@10	0.897
cosine_ndcg@10	0.7762
cosine_mrr@10	0.7381
cosine_map@100	0.7426

Training Details

Training Dataset

financial-filings-sparse-retrieval-training

Dataset: financial-filings-sparse-retrieval-training
Size: 15,509 training samples
Columns: anchor, positive, and negative
Approximate statistics based on the first 1000 samples:
anchor positive negative
type string string string
details
min: 5 tokens
mean: 16.07 tokens
max: 61 tokens
min: 52 tokens
mean: 341.78 tokens
max: 1566 tokens
min: 56 tokens
mean: 373.04 tokens
max: 1713 tokens

Samples:

anchor	positive	negative
`What is the unit of measurement used for Oil EUR on the vertical axis of the Relative Peer Well Performance chart?`	Relative Peer Well Performance[object Object][object Object]Source: RSEG (2019)[object Object][object Object]Note: Bubbles are sized by well count and colored by oil %[object Object][object Object]Oil EUR (Mbbl/1,000')[object Object][object Object]\| \| 0 \| 5 \| 10 \| 15 \| 20 \| 25 \| 30 \| 35 \| 40 \| 45 \| 50 \| \|---\|---\|---\|---\|---\|---\|---\|---\|---\|---\|---\|---\|---\| \| Average Horizontal Interwell Spacing (ft) \| 0 \| 100 \| 200 \| 300 \| 400 \| 500 \| 600 \| 700 \| 80\|[object Object][object Object]Wellhead Liquids (%)[object Object][object Object]<40 \| 40-55 \| >55	`Exhibit B`

Total Stockholder Return for the Company and each of the Peer Companies shall be calculated in accordance with the following formula, with the result expressed as a percentage:

The Company’s Relative Total Stockholder Return measured against the Peer Group shall be determined by first ranking the Company and each of the Peer Companies by their respective Total Stockholder Returns (highest to lowest) over the Performance Period. The Company’s Relative Total Stockholder Return shall be the Company’s percentile ranking determined from such numerical ranking, which percentile ranking shall be calculated as 100 multiplied by a fraction, the numerator of which is (x) the number of Peer Companies that are ranked lower than the Company by their respective Total Stockholder Returns and the denominator of which is (y) the number of Peer Companies in the Peer Group at the time of the determination minus one (1).

The number of Conditional PSUs with which you are credited, if any, at t... | | Q3 FY22 EBITDA decline reasons inflation |

Net sales is growing at healthy CAGR of 8.7% over 2 years, despite pandemic induced setbacks on our summer season brand sales[object Object][object Object]Metric YoY Gr 2 Year CAGR Net Sales 2.3 % 8.7% EBIDTA (34.8) % (7.0) % PBT * 1205.7 % 104.0% PAT * 1239.1 % 134.4%[object Object][object Object]Quarterly Performance - Q3 FY 22[object Object][object Object]EBIDTA de-grew mainly due to impact on gross margins on account of product mix and inflationary price increases in key raw materials & packing materials[object Object][object Object]PBT after exceptional items is growing mainly due to reduction in finance cost due to repayment of intercompany loan and nil exceptional item in the current quarter compared to Rs.342 million in previous year comparable period[object Object][object Object]The previous year comparable period includes exceptional items of Rs.342 million.

| PVC Pipes and Fittings volume registered a y-o-y growth of 4.2% to 1,58,266 MT.


[object Object]
[object Object]
[object Object]
[object Object]
[object Object]
[object Object]
[object Object]
[object Object]
[object Object]
[object Object]
[object Object]
[object Object]
[object Object]
[object Object]
[object Object]
[object Object]
[object Object]
[object Object]

Strong YTD perfo... | | What was the total value of trade creditors in 2022? |

2022 (£) 2021 (£) Trade debtors 1,849,235 1,877,661 Other debtors 105,874 75,543 Prepayments and accrued income 36,698 61,917 Total 1,991,807 2,015,121 Deferred tax asset (note 10) 341,000 408,000 Total Debtors 2,332,807 2,423,121


[object Object]
[object Object]
[object Object]
[object Object]
[object Object]
[object Object]
[object Object]
[object Object]
[object Object]
[object Object]
[object Object]

20... | 15 Other non-current liabilities


[object Object]
[object Object]
[object Object]
[object Object]
[object Object]
[object Object]
[object Object]

Trade and other debtors are shown after deducting a provision for impairment against tenant debto... |

Loss: MultipleNegativesRankingLoss with these parameters:

json
1{ 2 "scale": 20.0, 3 "similarity_fct": "cos_sim", 4 "gather_across_devices": false, 5 "directions": [ 6 "query_to_doc" 7 ], 8 "partition_mode": "joint", 9 "hardness_mode": null, 10 "hardness_strength": 0.0 11}

Evaluation Dataset

financial-filings-sparse-retrieval-training

Dataset: financial-filings-sparse-retrieval-training
Size: 2,738 evaluation samples
Columns: anchor, positive, and negative
Approximate statistics based on the first 1000 samples:
anchor positive negative
type string string string
details
min: 4 tokens
mean: 16.07 tokens
max: 52 tokens
min: 56 tokens
mean: 325.18 tokens
max: 1272 tokens
min: 64 tokens
mean: 355.84 tokens
max: 1099 tokens

Samples:

anchor	positive	negative
`What was the regional distribution of holdings within the United States and Canada during the spring of 2020?`	OWNERSHIP STAKE[object Object][object Object]March 2020 March 2019 Fully operational 97% 97% Construction 3% 3%[object Object][object Object]70 HICL ANNUAL REPORT 2020[object Object][object Object]March 2020 March 2019 UK 76% 77% EU 17% 15% North America 7% 8%[object Object][object Object]March 2020 March 2019 PPP projects 72% 71% Demand-based assets 20% 21% Regulated assets 8% 8%[object Object][object Object]MARKET SEGMENT[object Object][object Object]GEOGRAPHIC LOCATION[object Object][object Object]INVESTMENT STATUS[object Object][object Object]3.6 Portfolio Analysis[object Object][object Object]March 2020 March 2019 100% ownership 26% 25% 50%-100% ownership 34% 32% Less than 50% ownership 40% 43%[object Object][object Object]SECTOR[object Object][object Object]as at 31 March 2020[object Object][object Object]March 2020 March 2019 Accommodation 11% 11% Education 14% 15% Electricity, Gas & Water 8% 8% Health 30% 28% Fire, Law & Order 7% 7% Transport 30% 31%	`PROPERTY, PLANT & EQUIPMENT`

(a) At cost, except Factory land which is at cost, less amounts written off.

Particulars Land & Building Factory Building Office Equipment's Furniture & Fixtures Plant & Machinery Vehicles Computers Electrical fittings Total Tangible Assets Intangible Assets Software Total Intangible Assets GROSS BLOCK As on 1st April 2019 17.22 443.21 46.92 37.11 2,339.36 8.32 25.50 1.80 2,919.44 4.67 4.67 Additions - 3.74 0.32 45.17 - 1.53 0.40 51.16 3.76 3.76 Disposals - - - 2.64 - - - 2.64 - - At 31 March 2020 17.22 443.21 50.67 37.43 2,381.89 8.32 27.03 2.20 2,967.96 8.43 8.43 As on 1st April 2020 17.22 443.21 50.67 37.43 2,381.89 8.32 27.03 2.20 2,967.96 8.43 8.43 Additions 24.52 - 49.09 42.36 2.08 6.27 124.33 - - Disposals - - - - - - - - - - - At 31 March 2021 41.74 443.21 100.67 80.59 2,430.98 50.68 29.11 8.47 3,092.29 8.43 8.43 DEPRECIATION As on 1st April 2019 - 320.72 41.98 27.39 2,153.01 5.39 19.32 0.12 2,567.94 4.67 4.67 Charge for the year 17.21 2.98 2.52 68... | | What is the interest rate charged on the loan to Karbon Carlton Staindrop LLP? | Unrelieved tax losses of £68,994 (2023: £68,994) remain available to offset against future taxable profits.


[object Object]
[object Object]
[object Object]
[object Object]
[object Object]
[object Object]
[object Object]
[object Object]
[object Object]

2024 (£) 2023 (£) Amounts owed by Karbon Carlton Staindrop LLP - loan capital 1,776... |

RWK GOODMAN LLP
NOTES TO THE FINANCIAL STATEMENTS (CONTINUED)
FOR THE YEAR ENDED 31 MARCH 2025


[object Object]



































[object Object]
[object Object]

**19 Provision... | | Newmont Corporation 2022 restricted cash reconciliation |

```markdown
NEWMONT CORPORATION
CONSOLIDATED STATEMENTS OF CASH FLOWS

























[object Object]









































[object Object]





























[object Object]

(1) Acquisitions, net for the year ended December 31, 2021 is primarily related to the asset acquisition of the remaining 85.1% of GT Gold. Refer to
Note 1 for additional information.
The accompanying ...

Component Percentage Personal Bonus 6% Base Salary 14% Company Bonus 15% Restricted Stock Units 20% Performance Stock Units 45%[object Object][object Object]Component Objectives/Alignment Stock Price Performance • Value varies with NEM performance • Retention component Relative Stock Performance Long-term incentive to outperform gold competitors: • Absolute share price performance • Relative TSR performance Operating Performance Growth Pipeline/ Sustainability • Safety, Free Cash Flow, CSC, ROCE • Reserves and Resources • Sustainability and External Relations Leadership Measures • Individual objectives (with defined targets) • Leadership Pipeline results Base Salary • Adjusted for performance, scope • Market rate[object Object][object Object]Objectives/Alignment[object Object][object Object]Incentive vehicles balance key performance elements[object Object][object Object]Pay mix (CEO % shown)[object Object][object Object]Executive Compensation Structure[object Object][object Object]BANK OF AMERICA 2020 GLOBAL METALS, MINING & STEEL CONFERENCE NEWMONT CORPORATION[object Object][object Object]Plans support value chain to operating and market performance

Loss: MultipleNegativesRankingLoss with these parameters:

json
1{ 2 "scale": 20.0, 3 "similarity_fct": "cos_sim", 4 "gather_across_devices": false, 5 "directions": [ 6 "query_to_doc" 7 ], 8 "partition_mode": "joint", 9 "hardness_mode": null, 10 "hardness_strength": 0.0 11}

Training Hyperparameters

Non-Default Hyperparameters

per_device_train_batch_size: 2
max_steps: 100
learning_rate: 3e-05
lr_scheduler_type: constant_with_warmup
warmup_steps: 0.03
gradient_accumulation_steps: 8
fp16: True
gradient_checkpointing: True
load_best_model_at_end: True

All Hyperparameters

Click to expand

per_device_train_batch_size: 2
num_train_epochs: 3.0
max_steps: 100
learning_rate: 3e-05
lr_scheduler_type: constant_with_warmup
lr_scheduler_kwargs: None
warmup_steps: 0.03
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: 8
average_tokens_across_devices: True
max_grad_norm: 1.0
label_smoothing_factor: 0.0
bf16: False
fp16: True
bf16_full_eval: False
fp16_full_eval: False
tf32: None
gradient_checkpointing: True
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
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: True
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: 0
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: batch_sampler
multi_dataset_batch_sampler: proportional
router_mapping: {}
learning_rate_mapping: {}

Training Logs

Click to expand

Epoch	Step	Training Loss	Validation Loss	financial_corpus_cosine_ndcg@10
0.0021	1	0.8457	-	-
0.0041	2	0.7959	-	-
0.0062	3	0.4863	-	-
0.0083	4	0.6768	-	-
0.0103	5	0.5869	-	-
0.0124	6	0.5967	-	-
0.0144	7	0.7900	-	-
0.0165	8	0.7373	-	-
0.0186	9	0.4951	-	-
0.0206	10	0.5439	-	-
0.0227	11	0.4619	-	-
0.0248	12	0.6357	-	-
0.0268	13	0.5322	-	-
0.0289	14	0.4619	-	-
0.0309	15	0.6592	-	-
0.0330	16	0.7598	-	-
0.0351	17	0.4766	-	-
0.0371	18	0.7471	-	-
0.0392	19	0.7686	-	-
0.0413	20	0.4521	-	-
0.0433	21	0.5898	-	-
0.0454	22	0.5352	-	-
0.0474	23	0.7715	-	-
0.0495	24	0.5576	-	-
0.0516	25	0.6650	0.7444	-
0.0536	26	0.2832	-	-
0.0557	27	0.5557	-	-
0.0578	28	0.6738	-	-
0.0598	29	0.5635	-	-
0.0619	30	0.5264	-	-
0.0640	31	0.5576	-	-
0.0660	32	0.7480	-	-
0.0681	33	0.3516	-	-
0.0701	34	0.4668	-	-
0.0722	35	0.3682	-	-
0.0743	36	0.4756	-	-
0.0763	37	0.4404	-	-
0.0784	38	0.4287	-	-
0.0805	39	0.4297	-	-
0.0825	40	0.5137	-	-
0.0846	41	0.5010	-	-
0.0866	42	0.3525	-	-
0.0887	43	0.4746	-	-
0.0908	44	0.4766	-	-
0.0928	45	0.4453	-	-
0.0949	46	0.5273	-	-
0.0970	47	0.5205	-	-
0.0990	48	0.4502	-	-
0.1011	49	0.6289	-	-
0.1031	50	0.6191	0.6595	-
0.1052	51	0.4775	-	-
0.1073	52	0.4365	-	-
0.1093	53	0.4365	-	-
0.1114	54	0.5840	-	-
0.1135	55	0.3516	-	-
0.1155	56	0.3691	-	-
0.1176	57	0.4629	-	-
0.1196	58	0.6191	-	-
0.1217	59	0.5908	-	-
0.1238	60	0.3848	-	-
0.1258	61	0.3730	-	-
0.1279	62	0.5732	-	-
0.1300	63	0.5830	-	-
0.1320	64	0.4727	-	-
0.1341	65	0.3613	-	-
0.1362	66	0.4678	-	-
0.1382	67	0.5986	-	-
0.1403	68	0.2617	-	-
0.1423	69	0.3809	-	-
0.1444	70	0.4395	-	-
0.1465	71	0.3418	-	-
0.1485	72	0.3809	-	-
0.1506	73	0.3721	-	-
0.1527	74	0.3877	-	-
0.1547	75	0.3438	0.6091	-
0.1568	76	0.4131	-	-
0.1588	77	0.6221	-	-
0.1609	78	0.3652	-	-
0.1630	79	0.3164	-	-
0.1650	80	0.3135	-	-
0.1671	81	0.2764	-	-
0.1692	82	0.7373	-	-
0.1712	83	0.3223	-	-
0.1733	84	0.4902	-	-
0.1753	85	0.4219	-	-
0.1774	86	0.3428	-	-
0.1795	87	0.3291	-	-
0.1815	88	0.3135	-	-
0.1836	89	0.3945	-	-
0.1857	90	0.2939	-	-
0.1877	91	0.3135	-	-
0.1898	92	0.3682	-	-
0.1919	93	0.5322	-	-
0.1939	94	0.3594	-	-
0.1960	95	0.2666	-	-
0.1980	96	0.3730	-	-
0.2001	97	0.4292	-	-
0.2022	98	0.4131	-	-
0.2042	99	0.4580	-	-
0.2063	100	0.3711	0.5381	-
-1	-1	-	-	0.7762

The bold row denotes the saved checkpoint.

Training Time

Training: 4.6 hours

Framework Versions

Python: 3.12.13
Sentence Transformers: 5.4.1
Transformers: 5.5.0
PyTorch: 2.10.0+cu128
Accelerate: 1.13.0
Datasets: 4.3.0
Tokenizers: 0.22.2

Citation

BibTeX

Sentence Transformers

bibtex

@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

bibtex

@misc{oord2019representationlearningcontrastivepredictive,
      title={Representation Learning with Contrastive Predictive Coding},
      author={Aaron van den Oord and Yazhe Li and Oriol Vinyals},
      year={2019},
      eprint={1807.03748},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/1807.03748},
}

	anchor	positive	negative
type	string	string	string
details	min: 5 tokens mean: 16.07 tokens max: 61 tokens	min: 52 tokens mean: 341.78 tokens max: 1566 tokens	min: 56 tokens mean: 373.04 tokens max: 1713 tokens