CrossEncoder based on cross-encoder/ms-marco-MiniLM-L6-v2

This is a Cross Encoder model finetuned from cross-encoder/ms-marco-MiniLM-L6-v2 using the sentence-transformers library. It computes scores for pairs of texts, which can be used for text reranking and semantic search.

Model Details

Model Description

Model Type: Cross Encoder
Base model: cross-encoder/ms-marco-MiniLM-L6-v2
Maximum Sequence Length: 512 tokens
Number of Output Labels: 1 label

Model Sources

Documentation: Sentence Transformers Documentation
Documentation: Cross Encoder Documentation
Repository: Sentence Transformers on GitHub
Hugging Face: Cross Encoders on Hugging Face

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 CrossEncoder

# Download from the 🤗 Hub
model = CrossEncoder("cross_encoder_model_id")
# Get scores for pairs of texts
pairs = [
    ['(I) Designation The taxpayer shall designate the imputed income limitation of each unit taken into account under such clause?', '(I) Designation\nThe taxpayer shall designate the imputed income limitation of each unit taken into\naccount under such clause.\n(II) Average test\nThe average of the imputed income limitations designated under subclause (I) shall not\nexceed 60 percent of area median gross income.\n(III) 10-percent increments\nThe designated imputed income limitation of any unit under subclause (I) shall be 20\npercent, 30 percent, 40 percent, 50 percent, 60 percent, 70 percent, or 80 percent of area\nmedian gross income.\nAny election under this paragraph, once made, shall be irrevocable. For purposes of this\nparagraph, any property shall not be treated as failing to be residential rental property merely\nbecause part of the building in which such property is located is used for purposes other than\nresidential rental purposes.\n(2) Rent-restricted units\n(A) In general\nFor purposes of paragraph (1), a residential unit is rent-restricted if the gross rent with respect'],
    ['Summarize the key points from usc26@118-78.pdf.', 'turn or other schedules. List the type and \namount of tax.\nOther taxes to be listed include the \nfollowing.\nForm 8978 adjustment. Complete the \nNegative Form 8978 Adjustment Work-\nsheet—Schedule 2 (Line 17z) if you are \nfiling Form 8978 and completed the \nworksheet in the Schedule 3, line 6l, in-\nstructions and the amount on line 3 of \nthat worksheet is negative.\n100'],
    ['Summarize the key points from usc26@118-78.pdf.', 'program who are included in a unit of employees covered by an agreement which the Secretary of Labor finds\nto be a collective bargaining agreement between employee representatives and one or more employers, if there\nis evidence that educational assistance benefits were the subject of good faith bargaining between such\nemployee representatives and such employer or employers."\nPub. L. 99–514, §1114(b)(4), substituted "highly compensated employees (within the meaning of section\n414(q))" for "officers, owners, or highly compensated,".\nSubsec. (b)(6). Pub. L. 99–514, §1151(c)(4)(B), struck out par. (6) which read as follows: "\n.—Reasonable notification of the availability and terms of the program\nNOTIFICATION OF EMPLOYEES\nmust be provided to eligible employees."\nSubsec. (d). Pub. L. 99–514, §1162(a)(1), substituted "December 31, 1987" for "December 31, 1985".\n1984—Subsec. (a). Pub. L. 98–611, §1(b), amended subsec. generally, substituting "Exclusion from gross'],
    ['Summarize the key points from usc26@118-78.pdf.', 'the taxpayer or by law. Taxpayers have the right to expect \nappropriate action will be taken against employees, return \npreparers, and others who wrongfully use or disclose taxpayer \nreturn information.\n9. The Right to Retain Representation\nTaxpayers have the right to retain an authorized representative \nof their choice to represent them in their dealings with the \nIRS. Taxpayers have the right to seek assistance from a Low \nIncome Taxpayer Clinic if they cannot afford representation.\n10. The Right to a Fair and Just Tax System\nTaxpayers have the right to expect the tax system to consider \nfacts and circumstances that might affect their underlying \nliabilities, ability to pay, or ability to provide information timely. \nTaxpayers have the right to receive assistance from the \nTaxpayer Advocate Service if they are experiencing ﬁnancial \ndifﬁculty or if the IRS has not resolved their tax issues properly \nand timely through its normal channels. \n113'],
    ['Summarize the key points from usc26@118-78.pdf.', 'ceived, include the amount withheld in \nthe total on line 25b. This should be \nshown in box 4 of Form 1099, box 6 of \nForm SSA-1099, or box 10 of Form \nRRB-1099.\nLine 25c—Other Forms\nInclude on line 25c any federal income \ntax withheld on your Form(s) W-2G. \nThe amount withheld should be shown \nin box 4. Attach Form(s) W-2G to the \nfront of your return if federal income tax \nwas withheld.\nIf you had Additional Medicare Tax \nwithheld, include the amount shown on \nForm 8959, line 24, in the total on \nline 25c. Attach Form 8959.\nInclude on line 25c any federal in-\ncome tax withheld that is shown on a \nSchedule K-1.\nAlso include on line 25c any tax \nwithheld that is shown on Form 1042-S, \nForm 8805, or Form 8288-A. You \nshould attach the form to your return to \nclaim a credit for the withholding.\nLine 26\n2023 Estimated Tax \nPayments\nEnter any estimated federal income tax \npayments you made for 2023. Include \nany overpayment that you applied to \nyour 2023 estimated tax from your 2022'],
]
scores = model.predict(pairs)
print(scores.shape)
# (5,)

# Or rank different texts based on similarity to a single text
ranks = model.rank(
    '(I) Designation The taxpayer shall designate the imputed income limitation of each unit taken into account under such clause?',
    [
        '(I) Designation\nThe taxpayer shall designate the imputed income limitation of each unit taken into\naccount under such clause.\n(II) Average test\nThe average of the imputed income limitations designated under subclause (I) shall not\nexceed 60 percent of area median gross income.\n(III) 10-percent increments\nThe designated imputed income limitation of any unit under subclause (I) shall be 20\npercent, 30 percent, 40 percent, 50 percent, 60 percent, 70 percent, or 80 percent of area\nmedian gross income.\nAny election under this paragraph, once made, shall be irrevocable. For purposes of this\nparagraph, any property shall not be treated as failing to be residential rental property merely\nbecause part of the building in which such property is located is used for purposes other than\nresidential rental purposes.\n(2) Rent-restricted units\n(A) In general\nFor purposes of paragraph (1), a residential unit is rent-restricted if the gross rent with respect',
        'turn or other schedules. List the type and \namount of tax.\nOther taxes to be listed include the \nfollowing.\nForm 8978 adjustment. Complete the \nNegative Form 8978 Adjustment Work-\nsheet—Schedule 2 (Line 17z) if you are \nfiling Form 8978 and completed the \nworksheet in the Schedule 3, line 6l, in-\nstructions and the amount on line 3 of \nthat worksheet is negative.\n100',
        'program who are included in a unit of employees covered by an agreement which the Secretary of Labor finds\nto be a collective bargaining agreement between employee representatives and one or more employers, if there\nis evidence that educational assistance benefits were the subject of good faith bargaining between such\nemployee representatives and such employer or employers."\nPub. L. 99–514, §1114(b)(4), substituted "highly compensated employees (within the meaning of section\n414(q))" for "officers, owners, or highly compensated,".\nSubsec. (b)(6). Pub. L. 99–514, §1151(c)(4)(B), struck out par. (6) which read as follows: "\n.—Reasonable notification of the availability and terms of the program\nNOTIFICATION OF EMPLOYEES\nmust be provided to eligible employees."\nSubsec. (d). Pub. L. 99–514, §1162(a)(1), substituted "December 31, 1987" for "December 31, 1985".\n1984—Subsec. (a). Pub. L. 98–611, §1(b), amended subsec. generally, substituting "Exclusion from gross',
        'the taxpayer or by law. Taxpayers have the right to expect \nappropriate action will be taken against employees, return \npreparers, and others who wrongfully use or disclose taxpayer \nreturn information.\n9. The Right to Retain Representation\nTaxpayers have the right to retain an authorized representative \nof their choice to represent them in their dealings with the \nIRS. Taxpayers have the right to seek assistance from a Low \nIncome Taxpayer Clinic if they cannot afford representation.\n10. The Right to a Fair and Just Tax System\nTaxpayers have the right to expect the tax system to consider \nfacts and circumstances that might affect their underlying \nliabilities, ability to pay, or ability to provide information timely. \nTaxpayers have the right to receive assistance from the \nTaxpayer Advocate Service if they are experiencing ﬁnancial \ndifﬁculty or if the IRS has not resolved their tax issues properly \nand timely through its normal channels. \n113',
        'ceived, include the amount withheld in \nthe total on line 25b. This should be \nshown in box 4 of Form 1099, box 6 of \nForm SSA-1099, or box 10 of Form \nRRB-1099.\nLine 25c—Other Forms\nInclude on line 25c any federal income \ntax withheld on your Form(s) W-2G. \nThe amount withheld should be shown \nin box 4. Attach Form(s) W-2G to the \nfront of your return if federal income tax \nwas withheld.\nIf you had Additional Medicare Tax \nwithheld, include the amount shown on \nForm 8959, line 24, in the total on \nline 25c. Attach Form 8959.\nInclude on line 25c any federal in-\ncome tax withheld that is shown on a \nSchedule K-1.\nAlso include on line 25c any tax \nwithheld that is shown on Form 1042-S, \nForm 8805, or Form 8288-A. You \nshould attach the form to your return to \nclaim a credit for the withholding.\nLine 26\n2023 Estimated Tax \nPayments\nEnter any estimated federal income tax \npayments you made for 2023. Include \nany overpayment that you applied to \nyour 2023 estimated tax from your 2022',
    ]
)
# [{'corpus_id': ..., 'score': ...}, {'corpus_id': ..., 'score': ...}, ...]

Training Details

Training Dataset

Unnamed Dataset

Size: 2,000 training samples
Columns: sentence_0, sentence_1, and label

Approximate statistics based on the first 1000 samples:

	sentence_0	sentence_1	label
type	string	string	float
details	min: 34 characters mean: 60.95 characters max: 160 characters	min: 109 characters mean: 892.77 characters max: 1000 characters	min: 0.0 mean: 0.5 max: 1.0

Samples:

sentence_0	sentence_1	label
`(I) Designation The taxpayer shall designate the imputed income limitation of each unit taken into account under such clause?`	(I) Designation The taxpayer shall designate the imputed income limitation of each unit taken into account under such clause. (II) Average test The average of the imputed income limitations designated under subclause (I) shall not exceed 60 percent of area median gross income. (III) 10-percent increments The designated imputed income limitation of any unit under subclause (I) shall be 20 percent, 30 percent, 40 percent, 50 percent, 60 percent, 70 percent, or 80 percent of area median gross income. Any election under this paragraph, once made, shall be irrevocable. For purposes of this paragraph, any property shall not be treated as failing to be residential rental property merely because part of the building in which such property is located is used for purposes other than residential rental purposes. (2) Rent-restricted units (A) In general For purposes of paragraph (1), a residential unit is rent-restricted if the gross rent with respect	`1.0`
`Summarize the key points from usc26@118-78.pdf.`	`turn or other schedules. List the type and amount of tax. Other taxes to be listed include the following. Form 8978 adjustment. Complete the Negative Form 8978 Adjustment Work- sheet—Schedule 2 (Line 17z) if you are filing Form 8978 and completed the worksheet in the Schedule 3, line 6l, in- structions and the amount on line 3 of that worksheet is negative. 100`	`0.0`
`Summarize the key points from usc26@118-78.pdf.`	program who are included in a unit of employees covered by an agreement which the Secretary of Labor finds to be a collective bargaining agreement between employee representatives and one or more employers, if there is evidence that educational assistance benefits were the subject of good faith bargaining between such employee representatives and such employer or employers." Pub. L. 99–514, §1114(b)(4), substituted "highly compensated employees (within the meaning of section 414(q))" for "officers, owners, or highly compensated,". Subsec. (b)(6). Pub. L. 99–514, §1151(c)(4)(B), struck out par. (6) which read as follows: " .—Reasonable notification of the availability and terms of the program NOTIFICATION OF EMPLOYEES must be provided to eligible employees." Subsec. (d). Pub. L. 99–514, §1162(a)(1), substituted "December 31, 1987" for "December 31, 1985". 1984—Subsec. (a). Pub. L. 98–611, §1(b), amended subsec. generally, substituting "Exclusion from gross	`1.0`

Loss: BinaryCrossEntropyLoss with these parameters:

{
    "activation_fn": "torch.nn.modules.linear.Identity",
    "pos_weight": null
}

Training Hyperparameters

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
eval_strategy: no
prediction_loss_only: True
per_device_train_batch_size: 8
per_device_eval_batch_size: 8
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: 3
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
bf16: False
fp16: False
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}
parallelism_config: None
deepspeed: None
label_smoothing_factor: 0.0
optim: adamw_torch_fused
optim_args: None
adafactor: False
group_by_length: False
length_column_name: length
project: huggingface
trackio_space_id: trackio
ddp_find_unused_parameters: None
ddp_bucket_cap_mb: None
ddp_broadcast_buffers: False
dataloader_pin_memory: True
dataloader_persistent_workers: False
skip_memory_metrics: True
use_legacy_prediction_loop: False
push_to_hub: False
resume_from_checkpoint: None
hub_model_id: None
hub_strategy: every_save
hub_private_repo: None
hub_always_push: False
hub_revision: None
gradient_checkpointing: False
gradient_checkpointing_kwargs: None
include_inputs_for_metrics: False
include_for_metrics: []
eval_do_concat_batches: True
fp16_backend: auto
push_to_hub_model_id: None
push_to_hub_organization: None
mp_parameters:
auto_find_batch_size: False
full_determinism: False
torchdynamo: None
ray_scope: last
ddp_timeout: 1800
torch_compile: False
torch_compile_backend: None
torch_compile_mode: None
include_tokens_per_second: False
include_num_input_tokens_seen: no
neftune_noise_alpha: None
optim_target_modules: None
batch_eval_metrics: False
eval_on_start: False
use_liger_kernel: False
liger_kernel_config: None
eval_use_gather_object: False
average_tokens_across_devices: True
prompts: None
batch_sampler: batch_sampler
multi_dataset_batch_sampler: proportional
router_mapping: {}
learning_rate_mapping: {}

Training Logs

Epoch	Step	Training Loss
2.0	500	0.2146

Framework Versions

Python: 3.12.12
Sentence Transformers: 5.1.2
Transformers: 4.57.3
PyTorch: 2.9.0+cu126
Accelerate: 1.12.0
Datasets: 4.0.0
Tokenizers: 0.22.1

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",
}