This is a Cross Encoder model finetuned from sentence-transformers/all-MiniLM-L6-v2 on the ms_marco dataset 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: sentence-transformers/all-MiniLM-L6-v2
Maximum Sequence Length: 512 tokens
Number of Output Labels: 1 label
Training Dataset:
- ms_marco
Language: en

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("skfrost19/reranker-msmarco-v1.1-Lion-all-MiniLM-L6-v2-bce")
# Get scores for pairs of texts
pairs = [
    ['what is a ladyfinger', 'A light, delicate sponge cake roughly shaped like a large, fat finger. Used as an accompaniment to ice cream, puddings and other desserts. One of the oldest and most delicate of sponge cakes, from the House of Savoy in eleventh century France. American ladyfingers are smaller and moister than their Italian counterparts.'],
    ['what is itp blood disorder', "Immune thrombocytopenia (THROM-bo-si-toe-PE-ne-ah), or ITP, is a bleeding disorder. In ITP, the blood doesn't clot as it should. This is due to a low number of blood cell fragments called platelets (PLATE-lets) or thrombocytes (THROM-bo-sites). Platelets are made in your bone marrow along with other kinds of blood cells."],
    ['what is spark programming', 'Spark 1.5.1 works with Java 7 and higher. If you are using Java 8, Spark supports lambda expressions for concisely writing functions, otherwise you can use the classes in the org.apache.spark.api.java.function package. To write a Spark application in Java, you need to add a dependency on Spark. The first thing a Spark program must do is to create a JavaSparkContext object, which tells Spark how to access a cluster. To create a SparkContext you first need to build a SparkConf object that contains information about your application.'],
    ['cost to replace a driveway', "The average national cost of a driveway installation is $3,647, with most homeowners spending between $2,026 and $5,278. This data is based on actual project costs as reported by HomeAdvisor members. If you're thinking about installing a driveway, it's important to consider a couple of different things. "],
    ['is quaker oatmeal gluten-free', 'Gluten-Free Confidence Score: 5/10. While pure oats themselves are technically gluten-free, cross contamination can occur while growing, storing, and processing of oats. It is because of this that Quaker Oats cannot guarantee that their oats in their oatmeal are truly gluten-free.'],
]
scores = model.predict(pairs)
print(scores.shape)
# (5,)

# Or rank different texts based on similarity to a single text
ranks = model.rank(
    'what is a ladyfinger',
    [
        'A light, delicate sponge cake roughly shaped like a large, fat finger. Used as an accompaniment to ice cream, puddings and other desserts. One of the oldest and most delicate of sponge cakes, from the House of Savoy in eleventh century France. American ladyfingers are smaller and moister than their Italian counterparts.',
        "Immune thrombocytopenia (THROM-bo-si-toe-PE-ne-ah), or ITP, is a bleeding disorder. In ITP, the blood doesn't clot as it should. This is due to a low number of blood cell fragments called platelets (PLATE-lets) or thrombocytes (THROM-bo-sites). Platelets are made in your bone marrow along with other kinds of blood cells.",
        'Spark 1.5.1 works with Java 7 and higher. If you are using Java 8, Spark supports lambda expressions for concisely writing functions, otherwise you can use the classes in the org.apache.spark.api.java.function package. To write a Spark application in Java, you need to add a dependency on Spark. The first thing a Spark program must do is to create a JavaSparkContext object, which tells Spark how to access a cluster. To create a SparkContext you first need to build a SparkConf object that contains information about your application.',
        "The average national cost of a driveway installation is $3,647, with most homeowners spending between $2,026 and $5,278. This data is based on actual project costs as reported by HomeAdvisor members. If you're thinking about installing a driveway, it's important to consider a couple of different things. ",
        'Gluten-Free Confidence Score: 5/10. While pure oats themselves are technically gluten-free, cross contamination can occur while growing, storing, and processing of oats. It is because of this that Quaker Oats cannot guarantee that their oats in their oatmeal are truly gluten-free.',
    ]
)
# [{'corpus_id': ..., 'score': ...}, {'corpus_id': ..., 'score': ...}, ...]

Evaluation

Metrics

Cross Encoder Reranking

Datasets: NanoMSMARCO_R100, NanoNFCorpus_R100 and NanoNQ_R100
Evaluated with CrossEncoderRerankingEvaluator with these parameters:
json
1{ 2 "at_k": 10, 3 "always_rerank_positives": true 4}

Metric	NanoMSMARCO_R100	NanoNFCorpus_R100	NanoNQ_R100
map	0.0637 (-0.4259)	0.2746 (+0.0136)	0.0348 (-0.3848)
mrr@10	0.0361 (-0.4414)	0.2927 (-0.2071)	0.0111 (-0.4156)
ndcg@10	0.0507 (-0.4898)	0.2176 (-0.1074)	0.0300 (-0.4706)

Cross Encoder Nano BEIR

Dataset: NanoBEIR_R100_mean
Evaluated with CrossEncoderNanoBEIREvaluator with these parameters:
json
1{ 2 "dataset_names": [ 3 "msmarco", 4 "nfcorpus", 5 "nq" 6 ], 7 "rerank_k": 100, 8 "at_k": 10, 9 "always_rerank_positives": true 10}

Metric	Value
map	0.1244 (-0.2657)
mrr@10	0.1133 (-0.3547)
ndcg@10	0.0994 (-0.3559)

Training Details

Training Dataset

ms_marco

Dataset: ms_marco at a47ee7a
Size: 666,193 training samples
Columns: query, passage, and label
Approximate statistics based on the first 1000 samples:
query passage label
type string string int
details
min: 8 characters
mean: 33.72 characters
max: 101 characters
min: 121 characters
mean: 417.75 characters
max: 843 characters
0: ~86.60%
1: ~13.40%

Samples:

query	passage	label
`what is signifying`	`Signifyin' directs attention to the connotative, context-bound significance of words, which is accessible only to those who share the cultural values of a given speech community. While Signifyin(g) is the term coined by Henry Louis Gates, Jr. to represent a black vernacular, the idea stems from the thoughts of Ferdinand De Saussure and the process of signifying--the association between words and the ideas they indicate..`	`0`
`definition of scaled`	`scale noun (SIZE). B2 [S or U] the size or level of something, especially when this is large: We don't yet know the scale of the problem. Nuclear weapons cause destruction on a massive scale (= cause a lot of destruction). scale noun. › [S or U] the size or level of something, especially when this is large: the scale of sth We failed to recognize the scale of the problem. on a large/small scale. on a global/national/international scale.`	`0`
`what is planibel glass`	`Product Details. Planibel A is the new high performance LOW-E hard coating specially developed to reach the high ranking within the WER scale (Window Energy Rating).`	`0`

Loss: BinaryCrossEntropyLoss with these parameters:
json
1{ 2 "activation_fn": "torch.nn.modules.linear.Identity", 3 "pos_weight": null 4}

Evaluation Dataset

ms_marco

Dataset: ms_marco at a47ee7a
Size: 10,000 evaluation samples
Columns: query, passage, and label
Approximate statistics based on the first 1000 samples:
query passage label
type string string int
details
min: 12 characters
mean: 34.24 characters
max: 99 characters
min: 79 characters
mean: 421.13 characters
max: 1061 characters
0: ~88.60%
1: ~11.40%

Samples:

query	passage	label
`what is a ladyfinger`	`A light, delicate sponge cake roughly shaped like a large, fat finger. Used as an accompaniment to ice cream, puddings and other desserts. One of the oldest and most delicate of sponge cakes, from the House of Savoy in eleventh century France. American ladyfingers are smaller and moister than their Italian counterparts.`	`0`
`what is itp blood disorder`	`Immune thrombocytopenia (THROM-bo-si-toe-PE-ne-ah), or ITP, is a bleeding disorder. In ITP, the blood doesn't clot as it should. This is due to a low number of blood cell fragments called platelets (PLATE-lets) or thrombocytes (THROM-bo-sites). Platelets are made in your bone marrow along with other kinds of blood cells.`	`0`
`what is spark programming`	Spark 1.5.1 works with Java 7 and higher. If you are using Java 8, Spark supports lambda expressions for concisely writing functions, otherwise you can use the classes in the org.apache.spark.api.java.function package. To write a Spark application in Java, you need to add a dependency on Spark. The first thing a Spark program must do is to create a JavaSparkContext object, which tells Spark how to access a cluster. To create a SparkContext you first need to build a SparkConf object that contains information about your application.	`0`

Loss: BinaryCrossEntropyLoss with these parameters:
json
1{ 2 "activation_fn": "torch.nn.modules.linear.Identity", 3 "pos_weight": null 4}

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: epoch
per_device_train_batch_size: 512
per_device_eval_batch_size: 512
learning_rate: 2e-08
warmup_ratio: 0.1
seed: 12
bf16: True
load_best_model_at_end: 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: 512
per_device_eval_batch_size: 512
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: 2e-08
weight_decay: 0.0
adam_beta1: 0.9
adam_beta2: 0.999
adam_epsilon: 1e-08
max_grad_norm: 1.0
num_train_epochs: 3
max_steps: -1
lr_scheduler_type: linear
lr_scheduler_kwargs: {}
warmup_ratio: 0.1
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: 12
data_seed: None
jit_mode_eval: False
use_ipex: False
bf16: True
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: 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}
tp_size: 0
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}
deepspeed: None
label_smoothing_factor: 0.0
optim: adamw_torch
optim_args: None
adafactor: False
group_by_length: False
length_column_name: length
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
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
dispatch_batches: None
split_batches: None
include_tokens_per_second: False
include_num_input_tokens_seen: False
neftune_noise_alpha: None
optim_target_modules: None
batch_eval_metrics: False
eval_on_start: False
use_liger_kernel: False
eval_use_gather_object: False
average_tokens_across_devices: False
prompts: None
batch_sampler: batch_sampler
multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	Training Loss	Validation Loss	NanoMSMARCO_R100_ndcg@10	NanoNFCorpus_R100_ndcg@10	NanoNQ_R100_ndcg@10	NanoBEIR_R100_mean_ndcg@10
-1	-1	-	-	0.0281 (-0.5123)	0.2158 (-0.1092)	0.0350 (-0.4656)	0.0930 (-0.3624)
0.0008	1	0.6816	-	-	-	-	-
0.7680	1000	0.6802	-	-	-	-	-
1.0	1302	-	0.6756	0.0324 (-0.5081)	0.2171 (-0.1079)	0.0330 (-0.4677)	0.0941 (-0.3612)
1.5361	2000	0.6752	-	-	-	-	-
2.0	2604	-	0.6710	0.0507 (-0.4898)	0.2166 (-0.1084)	0.0184 (-0.4823)	0.0952 (-0.3602)
2.3041	3000	0.6718	-	-	-	-	-
3.0	3906	-	0.6693	0.0507 (-0.4898)	0.2176 (-0.1074)	0.0300 (-0.4706)	0.0994 (-0.3559)
-1	-1	-	-	0.0507 (-0.4898)	0.2176 (-0.1074)	0.0300 (-0.4706)	0.0994 (-0.3559)

The bold row denotes the saved checkpoint.

Framework Versions

Python: 3.11.5
Sentence Transformers: 4.0.1
Transformers: 4.50.3
PyTorch: 2.6.0+cu124
Accelerate: 1.6.0
Datasets: 3.5.0
Tokenizers: 0.21.1

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",
}

	query	passage	label
type	string	string	int
details	min: 8 characters mean: 33.72 characters max: 101 characters	min: 121 characters mean: 417.75 characters max: 843 characters	0: ~86.60% 1: ~13.40%