This is a Cross Encoder model finetuned from nreimers/BERT-Tiny_L-2_H-128_A-2 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: nreimers/BERT-Tiny_L-2_H-128_A-2
Maximum Sequence Length: 64 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.

python

from sentence_transformers import CrossEncoder

# Download from the 🤗 Hub
model = CrossEncoder("lucasflins/CE-BERT-Tiny_L-2_H-128_A-2-RankNet-20260207-093355")
# Get scores for pairs of texts
pairs = [
    ['escada 5', 'escada mor alumínio 5 degraus uso doméstico'],
    ['escada 5', 'escada alumínio 5 degraus mor'],
    ['escada 5', 'escada doméstica 05 degraus alumínio - real'],
    ['escada 5', 'escada alumínio 5 degraus - mor'],
    ['escada 5', 'escada de alumínio 5 degraus com fita de segurança mor'],
]
scores = model.predict(pairs)
print(scores.shape)
# (5,)

# Or rank different texts based on similarity to a single text
ranks = model.rank(
    'escada 5',
    [
        'escada mor alumínio 5 degraus uso doméstico',
        'escada alumínio 5 degraus mor',
        'escada doméstica 05 degraus alumínio - real',
        'escada alumínio 5 degraus - mor',
        'escada de alumínio 5 degraus com fita de segurança mor',
    ]
)
# [{'corpus_id': ..., 'score': ...}, {'corpus_id': ..., 'score': ...}, ...]

Training Details

Training Dataset

Unnamed Dataset

Size: 164,849 training samples
Columns: query, text, and label
Approximate statistics based on the first 1000 samples:
query text label
type string list list
details
min: 3 characters
mean: 24.5 characters
max: 97 characters
min: 3 elements
mean: 92.48 elements
max: 450 elements
min: 3 elements
mean: 92.48 elements
max: 450 elements

Samples:

query	text	label
`blusas femininas de verao`	`['blusa de malha feminina com decote v - manga curta - moda verão 2024', 'kit 2 regata feminina verão soltinhas confortáveis plus', 'blusa feminina manga princesa crepe verão básico em tendência 2025.', 'regata feminina malha fria senhoras plus verão', 'blusa cropped feminino tricot modal decote cruzado com alça verão', ...]`	`[1.0, 0.9726344517752807, 0.8374873988889217, 0.8008303826878321, 0.7915794516443119, ...]`
`boneca grande bebe reborn`	`['bebê reborn boneca princesa grande com vários acessórios', 'boneca reborn bebê realista grande com vestido florido', 'boneca bebe reborn grande 60cm corpo de pano com girafinha de pelúcia', 'bebe reborn loira realista grande 100% silicone 48cm banho', 'bebe reborn realista princesa grande 100% silicone 20 itens', ...]`	`[1.0, 0.9116494092065375, 0.8928073970303235, 0.8879649368952169, 0.8731986703319667, ...]`
`porta tempero de inox`	`['kit porta temperos e condimentos vidro tampa inox 6un 200ml', 'kit potes de vidro hermético 150ml porta tempero tampa inox 3 un', 'porta tempero condimento inox magnético imã cozinha 3 potes', 'kit 6 porta tempero condimento vidro suporte giratório inox cozinha pote organizador', 'porta tempero condimento giratório redondo 20 peças inox', ...]`	`[0.9655845721090657, 0.8396654217445704, 0.7738607871762022, 0.6343641152653602, 0.566489008709851, ...]`

Loss: RankNetLoss with these parameters:
json
1{ 2 "k": null, 3 "sigma": 1.0, 4 "eps": 1e-10, 5 "reduction_log": "binary", 6 "activation_fn": "torch.nn.modules.linear.Identity", 7 "mini_batch_size": null 8}

Evaluation Dataset

Unnamed Dataset

Size: 54,809 evaluation samples
Columns: query, text, and label
Approximate statistics based on the first 1000 samples:
query text label
type string list list
details
min: 3 characters
mean: 25.32 characters
max: 90 characters
min: 3 elements
mean: 91.67 elements
max: 511 elements
min: 3 elements
mean: 91.67 elements
max: 511 elements

Samples:

query	text	label
`escada 5`	`['escada mor alumínio 5 degraus uso doméstico', 'escada alumínio 5 degraus mor', 'escada doméstica 05 degraus alumínio - real', 'escada alumínio 5 degraus - mor', 'escada de alumínio 5 degraus com fita de segurança mor', ...]`	`[1.0, 0.7096172820312041, 0.6436050498616379, 0.6233409190446045, 0.5943286560299376, ...]`
`mascara lola`	`['lola cosmetics danos vorazes máscara de reparação intensiva', 'máscara disciplinante xapadinha 450g lola', 'máscara nutritiva lola bossa crespos e cachos 450g', 'máscara lola morte súbita 450gr', 'kit argan lola cosmetics - shampoo + máscara', ...]`	`[1.0, 0.9644981507636873, 0.9554469702621406, 0.9550674725047994, 0.9500765764997154, ...]`
`brinox roma`	`['chaleira em inox com indução e apito brinox roma 2,7 litros vanilla', 'chaleira roma inox 2,7l com apito de indução vanilla 4875/100 brinox', 'chaleira brinox inox com apito preto marble 2,7l roma']`	`[1.0, 0.29334574198355634, 0.0]`

Loss: RankNetLoss with these parameters:
json
1{ 2 "k": null, 3 "sigma": 1.0, 4 "eps": 1e-10, 5 "reduction_log": "binary", 6 "activation_fn": "torch.nn.modules.linear.Identity", 7 "mini_batch_size": null 8}

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: epoch
per_device_train_batch_size: 256
per_device_eval_batch_size: 32
num_train_epochs: 10
warmup_ratio: 0.1
log_level: info
tf32: True
load_best_model_at_end: True
push_to_hub: True
hub_strategy: end
hub_private_repo: True
hub_always_push: True
eval_on_start: True

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
eval_strategy: epoch
prediction_loss_only: True
per_device_train_batch_size: 256
per_device_eval_batch_size: 32
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
adam_beta1: 0.9
adam_beta2: 0.999
adam_epsilon: 1e-08
max_grad_norm: 1.0
num_train_epochs: 10
max_steps: -1
lr_scheduler_type: linear
lr_scheduler_kwargs: None
warmup_ratio: 0.1
warmup_steps: 0
log_level: info
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: True
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: True
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
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: True
resume_from_checkpoint: None
hub_model_id: None
hub_strategy: end
hub_private_repo: True
hub_always_push: True
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: True
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	Validation Loss
0	0	-	1.0004
1.0	644	0.9963	0.9871
2.0	1288	0.9783	0.9768
3.0	1932	0.9676	0.9725
4.0	2576	0.9613	0.9690
5.0	3220	0.9572	0.9676
6.0	3864	0.9538	0.9652
7.0	4508	0.9514	0.9646
8.0	5152	0.9492	0.9637
9.0	5796	0.9489	0.9631
10.0	6440	0.9473	0.9631

The bold row denotes the saved checkpoint.

Framework Versions

Python: 3.12.11
Sentence Transformers: 5.2.0
Transformers: 4.57.6
PyTorch: 2.7.1+cu128
Accelerate: 1.9.0
Datasets: 4.1.1
Tokenizers: 0.22.1

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",
}

RankNetLoss

bibtex

@inproceedings{burges2005learning,
  title={Learning to Rank using Gradient Descent},
  author={Burges, Chris and Shaked, Tal and Renshaw, Erin and Lazier, Ari and Deeds, Matt and Hamilton, Nicole and Hullender, Greg},
  booktitle={Proceedings of the 22nd international conference on Machine learning},
  pages={89--96},
  year={2005}
}

	query	text	label
type	string	list	list
details	min: 3 characters mean: 24.5 characters max: 97 characters	min: 3 elements mean: 92.48 elements max: 450 elements	min: 3 elements mean: 92.48 elements max: 450 elements