CrossEncoder based on cross-encoder/ms-marco-TinyBERT-L2

This is a Cross Encoder model finetuned from cross-encoder/ms-marco-TinyBERT-L2 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-TinyBERT-L2
Maximum Sequence Length: 384 tokens
Number of Output Labels: 1 label
Supported Modality: Text

Model Sources

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

Full Model Architecture

CrossEncoder(
  (0): Transformer({'transformer_task': 'sequence-classification', 'modality_config': {'text': {'method': 'forward', 'method_output_name': 'logits'}}, 'module_output_name': 'scores', 'architecture': 'BertForSequenceClassification'})
)

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("SumedhJoshi/convmix-tinybert-negative-sampling-followup-skip-3")
# Get scores for pairs of inputs
pairs = [
    ['What English football club was the first to achieve the European treble?', "Incidentally, Real Madrid from Spain have won this many by themselves. The most recent English win came in 2012 when Chelsea F.C. beat FC Bayern Munich on penalties at the Allianz Arena after the game ended 1-1 in extra time. A record five teams have won Europe's premier club competition: Liverpool five times, the second English team to retain the cup (1977, 1978, 1981, 1984, 2005), Manchester United three times (1968, 1999, 2008), Nottingham Forest twice, being the second English team to retain the European Cup (1979, 1980), Aston Villa once (1982) and Chelsea once (2012). English clubs also hold the records for the most consecutive tournament victories by clubs from one country (six wins between 1977 and 1982 by Liverpool, Forest, and Villa) as well as the most consecutive defeats in the final (four teams were runners-up once each between 2006 and 2009). European Cup era. Early years: 1955–1967."],
    ['What is the full name of Roque Júnior?', 'According to Swivel, an eclectic range of songs were recorded, including ballads, "weird ethereal things" and 1990\'s R&B and Afrobeat-inspired songs.'],
    ['What company originally published the 1973 novel "The Taking of Pelham One Two Three?"', "Graham Greene There have been six editions of the novel, the first from Heinemann in 1935, priced at 7\xa0shillings and sixpence, the equivalent of about £ in 2022. Ten years later a second edition was published by Grey Walls Press, with the addition of illustrations by Felix Kelly. A third edition, for which Graham Greene wrote an introduction focusing on the novel's autobiographical elements, was published by Eyre and Spottiswoode in 1947. The first American edition was published in New York by New Directions in 1948, with an introduction by Kenneth Rexroth. Penguin Books published a fifth edition in 1979, which included the 1947 introduction by Greene. A sixth edition, published by R. Clark, appeared in 1989 and was reprinted in 1995, both containing Greene's introduction."],
    ['Spanish name of the TV series Money Heist?', 'There were four episodes in the first series, which started on 19 June 2011: Series 2 (2012). The first episode of the second series was shown in the UK on 16 September 2012. The series had three episodes: Series 3 (2014). The third series features four episodes. The first episode was first shown on 27 December 2013, on Sweden\'s SVT, with episodes 2 and 3 shown in the following weeks. Philip Mould described the appearance on Swedish television weeks ahead of the British premiere as a "weird BBC World cock-up".'],
    ['What is the release date of the film Swades?', 'The film was known under the working title "Jackpot" but was later renamed as "Sahasam". The film was mostly shot in Ladakh, and Hyderabad. The film\'s first look was released on the eve of Ugadi i.e. 11 April 2013. The film was later dubbed into Hindi as "The Real Jackpot" by Goldmines Telefilms in 2013. Plot. In 2003, a Pakistani Archaeologist Dr. M. Alam Siddiqui after seeking permission from the Government of Pakistan starts researching about the hidden treasure of King Kanishka whose treasure map is hidden in an old box whose key is a Garuda Locket which was last seen in the research books of David, a British archaeologist.'],
]
scores = model.predict(pairs)
print(scores)
# [0.2202 0.0022 0.0959 0.0095 0.0251]

# Or rank different texts based on similarity to a single text
ranks = model.rank(
    'What English football club was the first to achieve the European treble?',
    [
        "Incidentally, Real Madrid from Spain have won this many by themselves. The most recent English win came in 2012 when Chelsea F.C. beat FC Bayern Munich on penalties at the Allianz Arena after the game ended 1-1 in extra time. A record five teams have won Europe's premier club competition: Liverpool five times, the second English team to retain the cup (1977, 1978, 1981, 1984, 2005), Manchester United three times (1968, 1999, 2008), Nottingham Forest twice, being the second English team to retain the European Cup (1979, 1980), Aston Villa once (1982) and Chelsea once (2012). English clubs also hold the records for the most consecutive tournament victories by clubs from one country (six wins between 1977 and 1982 by Liverpool, Forest, and Villa) as well as the most consecutive defeats in the final (four teams were runners-up once each between 2006 and 2009). European Cup era. Early years: 1955–1967.",
        'According to Swivel, an eclectic range of songs were recorded, including ballads, "weird ethereal things" and 1990\'s R&B and Afrobeat-inspired songs.',
        "Graham Greene There have been six editions of the novel, the first from Heinemann in 1935, priced at 7\xa0shillings and sixpence, the equivalent of about £ in 2022. Ten years later a second edition was published by Grey Walls Press, with the addition of illustrations by Felix Kelly. A third edition, for which Graham Greene wrote an introduction focusing on the novel's autobiographical elements, was published by Eyre and Spottiswoode in 1947. The first American edition was published in New York by New Directions in 1948, with an introduction by Kenneth Rexroth. Penguin Books published a fifth edition in 1979, which included the 1947 introduction by Greene. A sixth edition, published by R. Clark, appeared in 1989 and was reprinted in 1995, both containing Greene's introduction.",
        'There were four episodes in the first series, which started on 19 June 2011: Series 2 (2012). The first episode of the second series was shown in the UK on 16 September 2012. The series had three episodes: Series 3 (2014). The third series features four episodes. The first episode was first shown on 27 December 2013, on Sweden\'s SVT, with episodes 2 and 3 shown in the following weeks. Philip Mould described the appearance on Swedish television weeks ahead of the British premiere as a "weird BBC World cock-up".',
        'The film was known under the working title "Jackpot" but was later renamed as "Sahasam". The film was mostly shot in Ladakh, and Hyderabad. The film\'s first look was released on the eve of Ugadi i.e. 11 April 2013. The film was later dubbed into Hindi as "The Real Jackpot" by Goldmines Telefilms in 2013. Plot. In 2003, a Pakistani Archaeologist Dr. M. Alam Siddiqui after seeking permission from the Government of Pakistan starts researching about the hidden treasure of King Kanishka whose treasure map is hidden in an old box whose key is a Garuda Locket which was last seen in the research books of David, a British archaeologist.',
    ]
)
# [{'corpus_id': ..., 'score': ...}, {'corpus_id': ..., 'score': ...}, ...]

Training Details

Training Dataset

Unnamed Dataset

Size: 8,494 training samples
Columns: sentence_0, sentence_1, and label

Approximate statistics based on the first 100 samples:

	sentence_0	sentence_1	label
type	string	string	float
modality	text	text
details	min: 7 tokens mean: 13.67 tokens max: 22 tokens	min: 11 tokens mean: 88.07 tokens max: 272 tokens	min: 0.0 mean: 0.06 max: 1.0

Samples:

sentence_0	sentence_1	label
`What English football club was the first to achieve the European treble?`	Incidentally, Real Madrid from Spain have won this many by themselves. The most recent English win came in 2012 when Chelsea F.C. beat FC Bayern Munich on penalties at the Allianz Arena after the game ended 1-1 in extra time. A record five teams have won Europe's premier club competition: Liverpool five times, the second English team to retain the cup (1977, 1978, 1981, 1984, 2005), Manchester United three times (1968, 1999, 2008), Nottingham Forest twice, being the second English team to retain the European Cup (1979, 1980), Aston Villa once (1982) and Chelsea once (2012). English clubs also hold the records for the most consecutive tournament victories by clubs from one country (six wins between 1977 and 1982 by Liverpool, Forest, and Villa) as well as the most consecutive defeats in the final (four teams were runners-up once each between 2006 and 2009). European Cup era. Early years: 1955–1967.	`1.0`
`What is the full name of Roque Júnior?`	`According to Swivel, an eclectic range of songs were recorded, including ballads, "weird ethereal things" and 1990's R&B and Afrobeat-inspired songs.`	`0.0`
`What company originally published the 1973 novel "The Taking of Pelham One Two Three?"`	Graham Greene There have been six editions of the novel, the first from Heinemann in 1935, priced at 7 shillings and sixpence, the equivalent of about £ in 2022. Ten years later a second edition was published by Grey Walls Press, with the addition of illustrations by Felix Kelly. A third edition, for which Graham Greene wrote an introduction focusing on the novel's autobiographical elements, was published by Eyre and Spottiswoode in 1947. The first American edition was published in New York by New Directions in 1948, with an introduction by Kenneth Rexroth. Penguin Books published a fifth edition in 1979, which included the 1947 introduction by Greene. A sixth edition, published by R. Clark, appeared in 1989 and was reprinted in 1995, both containing Greene's introduction.	`0.0`

Loss: BinaryCrossEntropyLoss with these parameters:

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

Training Hyperparameters

Non-Default Hyperparameters

per_device_train_batch_size: 16
num_train_epochs: 5
per_device_eval_batch_size: 16

All Hyperparameters

Click to expand

per_device_train_batch_size: 16
num_train_epochs: 5
max_steps: -1
learning_rate: 5e-05
lr_scheduler_type: linear
lr_scheduler_kwargs: None
warmup_steps: 0
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
label_smoothing_factor: 0.0
bf16: False
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: 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: 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: None
batch_sampler: batch_sampler
multi_dataset_batch_sampler: proportional
router_mapping: {}
learning_rate_mapping: {}
warmup_ratio: None

Training Logs

Epoch	Step	Training Loss
0.9416	500	0.2823
1.8832	1000	0.2272
2.8249	1500	0.2107
3.7665	2000	0.2046
4.7081	2500	0.2016

Training Time

Training: 23.2 minutes

Framework Versions

Python: 3.13.16
Sentence Transformers: 6.2.0
Transformers: 5.20.2
PyTorch: 2.11.0+cu130
Accelerate: 1.15.0
Datasets: 5.1.0
Tokenizers: 0.23.3

Additional Resources

Training and Finetuning Reranker Models with Sentence Transformers: the end-to-end guide for training or finetuning Cross Encoder (reranker) models.
Multimodal Embedding & Reranker Models with Sentence Transformers: use text, image, audio, and video reranker models through the same API.
Training and Finetuning Multimodal Embedding & Reranker Models with Sentence Transformers: training multimodal Cross Encoders.

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",
}