Add new CrossEncoder model

e187118 verified 16 days ago

28.6 kB

metadata

language:
  - en
tags:
  - sentence-transformers
  - cross-encoder
  - reranker
  - generated_from_trainer
  - dataset_size:9960000
  - loss:BinaryCrossEntropyLoss
base_model: jhu-clsp/ettin-encoder-1b
datasets:
  - sentence-transformers/msmarco
pipeline_tag: text-ranking
library_name: sentence-transformers
metrics:
  - map
  - mrr@10
  - ndcg@10
model-index:
  - name: CrossEncoder based on jhu-clsp/ettin-encoder-1b
    results:
      - task:
          type: cross-encoder-reranking
          name: Cross Encoder Reranking
        dataset:
          name: NanoMSMARCO R100
          type: NanoMSMARCO_R100
        metrics:
          - type: map
            value: 0.6418
            name: Map
          - type: mrr@10
            value: 0.6372
            name: Mrr@10
          - type: ndcg@10
            value: 0.7051
            name: Ndcg@10
      - task:
          type: cross-encoder-reranking
          name: Cross Encoder Reranking
        dataset:
          name: NanoNFCorpus R100
          type: NanoNFCorpus_R100
        metrics:
          - type: map
            value: 0.3816
            name: Map
          - type: mrr@10
            value: 0.5641
            name: Mrr@10
          - type: ndcg@10
            value: 0.4148
            name: Ndcg@10
      - task:
          type: cross-encoder-reranking
          name: Cross Encoder Reranking
        dataset:
          name: NanoNQ R100
          type: NanoNQ_R100
        metrics:
          - type: map
            value: 0.724
            name: Map
          - type: mrr@10
            value: 0.7315
            name: Mrr@10
          - type: ndcg@10
            value: 0.7717
            name: Ndcg@10
      - task:
          type: cross-encoder-nano-beir
          name: Cross Encoder Nano BEIR
        dataset:
          name: NanoBEIR R100 mean
          type: NanoBEIR_R100_mean
        metrics:
          - type: map
            value: 0.5824
            name: Map
          - type: mrr@10
            value: 0.6443
            name: Mrr@10
          - type: ndcg@10
            value: 0.6305
            name: Ndcg@10

CrossEncoder based on jhu-clsp/ettin-encoder-1b

This is a Cross Encoder model finetuned from jhu-clsp/ettin-encoder-1b on the msmarco 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: jhu-clsp/ettin-encoder-1b
Maximum Sequence Length: 7999 tokens
Number of Output Labels: 1 label
Training Dataset:
- msmarco
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.

from sentence_transformers import CrossEncoder

# Download from the 🤗 Hub
model = CrossEncoder("kdhole/reranker-ettin-encoder-1b-msmarco-bce-10m")
# Get scores for pairs of texts
pairs = [
    ['what is the zip code for hamilton nj', 'Stats and Demographics for the 21227 ZIP Code. ZIP code 21227 is located in Maryland and covers a slightly less than average land area compared to other ZIP codes in the United States. It also has a slightly larger than average population density.n addition to the primary city for a ZIP code, USPS also publishes a list of other acceptable cities that can be used with ZIP code 21227. However, if you are mailing something to ZIP code 21227, you should not use any of the cities listed as unacceptable.'],
    ['which cooking oil is healthy', "Photo Credit bgfoto126/iStock/Getty Images. Olive oil may get all the press, but safflower oil is emerging as a health superstar in its own right. Made from the seeds of safflowers, which are similar to sunflowers, this oil contains healthy fats and may even help prevent cardiovascular disease.All oils are pure fat, however, which means they're high in calories.hoto Credit bgfoto126/iStock/Getty Images. Olive oil may get all the press, but safflower oil is emerging as a health superstar in its own right. Made from the seeds of safflowers, which are similar to sunflowers, this oil contains healthy fats and may even help prevent cardiovascular disease."],
    ['what canadian cities have the highest crime rates?', 'Saginaw, TX Tarrant County crime report and crime data. Compare the rate to other cities, state average, and national average.'],
    ['how to delete bcc', 'How to delete the CC/BCC rule. 1  Open Auto BCC dialog box. 2  Select the rule you need in the list and click the Delete button on the top toolbar or the Delete key on your keyboard. 3  Confirm deletion and click Ok to save changes.'],
    ['what county is waukegan illinois in?', 'Downtown Waukegan is the urban center of Lake County. Many restaurants, bars, shops, the Waukegan Public Library, the College of Lake County, the Lake County Courthouse (including the William D. Block Memorial Law Library), and much more call Downtown Waukegan their home.'],
]
scores = model.predict(pairs)
print(scores.shape)
# (5,)

# Or rank different texts based on similarity to a single text
ranks = model.rank(
    'what is the zip code for hamilton nj',
    [
        'Stats and Demographics for the 21227 ZIP Code. ZIP code 21227 is located in Maryland and covers a slightly less than average land area compared to other ZIP codes in the United States. It also has a slightly larger than average population density.n addition to the primary city for a ZIP code, USPS also publishes a list of other acceptable cities that can be used with ZIP code 21227. However, if you are mailing something to ZIP code 21227, you should not use any of the cities listed as unacceptable.',
        "Photo Credit bgfoto126/iStock/Getty Images. Olive oil may get all the press, but safflower oil is emerging as a health superstar in its own right. Made from the seeds of safflowers, which are similar to sunflowers, this oil contains healthy fats and may even help prevent cardiovascular disease.All oils are pure fat, however, which means they're high in calories.hoto Credit bgfoto126/iStock/Getty Images. Olive oil may get all the press, but safflower oil is emerging as a health superstar in its own right. Made from the seeds of safflowers, which are similar to sunflowers, this oil contains healthy fats and may even help prevent cardiovascular disease.",
        'Saginaw, TX Tarrant County crime report and crime data. Compare the rate to other cities, state average, and national average.',
        'How to delete the CC/BCC rule. 1  Open Auto BCC dialog box. 2  Select the rule you need in the list and click the Delete button on the top toolbar or the Delete key on your keyboard. 3  Confirm deletion and click Ok to save changes.',
        'Downtown Waukegan is the urban center of Lake County. Many restaurants, bars, shops, the Waukegan Public Library, the College of Lake County, the Lake County Courthouse (including the William D. Block Memorial Law Library), and much more call Downtown Waukegan their home.',
    ]
)
# [{'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:

{
    "at_k": 10,
    "always_rerank_positives": true
}

Metric	NanoMSMARCO_R100	NanoNFCorpus_R100	NanoNQ_R100
map	0.6418 (+0.1522)	0.3816 (+0.1206)	0.7240 (+0.3044)
mrr@10	0.6372 (+0.1597)	0.5641 (+0.0642)	0.7315 (+0.3048)
ndcg@10	0.7051 (+0.1647)	0.4148 (+0.0898)	0.7717 (+0.2710)

Cross Encoder Nano BEIR

Dataset: NanoBEIR_R100_mean

Evaluated with CrossEncoderNanoBEIREvaluator with these parameters:

{
    "dataset_names": [
        "msmarco",
        "nfcorpus",
        "nq"
    ],
    "rerank_k": 100,
    "at_k": 10,
    "always_rerank_positives": true
}

Metric	Value
map	0.5824 (+0.1924)
mrr@10	0.6443 (+0.1763)
ndcg@10	0.6305 (+0.1752)

Training Details

Training Dataset

msmarco

Dataset: msmarco at 9e329ed
Size: 9,960,000 training samples
Columns: query, passage, and score

Approximate statistics based on the first 1000 samples:

	query	passage	score
type	string	string	float
details	min: 10 characters mean: 33.92 characters max: 156 characters	min: 80 characters mean: 344.88 characters max: 1061 characters	min: 0.0 mean: 0.48 max: 1.0

Samples:

query	passage	score
`what is liquid resources`	`Renewable Resources. Renewable resources are resources that are replenished by the environment over relatively short periods of time. This type of resource is much more desirable to use because often a resource renews so fast that it will have regenerated by the time you've used it up.n contrast to renewable resources, non-renewable resources are resources that are not easily replenished by the environment. Let's think about this in terms of that ice cube maker again.`	`0.0`
`halo laser treatment`	`Laser surgery (photocoagulation) With laser surgery, your ophthalmologist uses a laser to make small burns around the retinal tear. The scarring that results seals the retina to the underlying tissue, helping to prevent a retinal detachment. Freezing treatment (cryopexy)`	`0.0`
`three motorcycle car`	`Can-Am Spyder roadster: three wheeled motorcycle. February 20, 2007 BRP has unveiled its first on-road vehicle, the 2008 Can-Am Spyder roadster. This three-wheel vehicle, with two wheels in the front and one in the rear, offers a completely new and stunning look. Powered by a proven 990cc V Twin engine designed and manufactured by BRP-Rotax, Spyder roadster, with its unique Y-architecture, can be described as part motorcycle and part convertible sports car.`	`1.0`

Loss: BinaryCrossEntropyLoss with these parameters:

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

Evaluation Dataset

msmarco

Dataset: msmarco at 9e329ed
Size: 40,000 evaluation samples
Columns: query, passage, and score

Approximate statistics based on the first 1000 samples:

	query	passage	score
type	string	string	float
details	min: 12 characters mean: 34.65 characters max: 132 characters	min: 65 characters mean: 345.56 characters max: 874 characters	min: 0.0 mean: 0.49 max: 1.0

Samples:

query	passage	score
`what is the zip code for hamilton nj`	Stats and Demographics for the 21227 ZIP Code. ZIP code 21227 is located in Maryland and covers a slightly less than average land area compared to other ZIP codes in the United States. It also has a slightly larger than average population density.n addition to the primary city for a ZIP code, USPS also publishes a list of other acceptable cities that can be used with ZIP code 21227. However, if you are mailing something to ZIP code 21227, you should not use any of the cities listed as unacceptable.	`0.0`
`which cooking oil is healthy`	Photo Credit bgfoto126/iStock/Getty Images. Olive oil may get all the press, but safflower oil is emerging as a health superstar in its own right. Made from the seeds of safflowers, which are similar to sunflowers, this oil contains healthy fats and may even help prevent cardiovascular disease.All oils are pure fat, however, which means they're high in calories.hoto Credit bgfoto126/iStock/Getty Images. Olive oil may get all the press, but safflower oil is emerging as a health superstar in its own right. Made from the seeds of safflowers, which are similar to sunflowers, this oil contains healthy fats and may even help prevent cardiovascular disease.	`0.0`
`what canadian cities have the highest crime rates?`	`Saginaw, TX Tarrant County crime report and crime data. Compare the rate to other cities, state average, and national average.`	`0.0`

Loss: BinaryCrossEntropyLoss with these parameters:

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

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: steps
per_device_train_batch_size: 64
per_device_eval_batch_size: 64
learning_rate: 2e-05
num_train_epochs: 1
warmup_ratio: 0.1
seed: 12
bf16: True
dataloader_num_workers: 4
load_best_model_at_end: True

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
eval_strategy: steps
prediction_loss_only: True
per_device_train_batch_size: 64
per_device_eval_batch_size: 64
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-05
weight_decay: 0.0
adam_beta1: 0.9
adam_beta2: 0.999
adam_epsilon: 1e-08
max_grad_norm: 1.0
num_train_epochs: 1
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: 4
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_fused
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
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: False
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: False
prompts: None
batch_sampler: batch_sampler
multi_dataset_batch_sampler: proportional
router_mapping: {}
learning_rate_mapping: {}

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	-	-	0.0000 (-0.5404)	0.2648 (-0.0602)	0.0388 (-0.4618)	0.1012 (-0.3541)
0.0000	1	0.8074	-	-	-	-	-
0.0643	10000	0.1734	0.1322	0.6852 (+0.1448)	0.3939 (+0.0689)	0.7987 (+0.2980)	0.6259 (+0.1706)
0.1285	20000	0.1273	0.1189	0.7051 (+0.1647)	0.4148 (+0.0898)	0.7717 (+0.2710)	0.6305 (+0.1752)
0.1928	30000	0.1104	0.1027	0.7116 (+0.1712)	0.4199 (+0.0948)	0.7473 (+0.2467)	0.6263 (+0.1709)
0.2570	40000	0.0958	0.0929	0.6721 (+0.1317)	0.4184 (+0.0933)	0.7930 (+0.2923)	0.6278 (+0.1724)
0.3213	50000	0.0847	0.0773	0.6695 (+0.1291)	0.3852 (+0.0602)	0.7673 (+0.2666)	0.6074 (+0.1520)
0.3855	60000	0.0746	0.0696	0.7101 (+0.1697)	0.4118 (+0.0868)	0.7599 (+0.2592)	0.6273 (+0.1719)
0.4498	70000	0.0672	0.0618	0.6663 (+0.1259)	0.3761 (+0.0511)	0.7701 (+0.2694)	0.6042 (+0.1488)
0.5141	80000	0.0603	0.0552	0.6773 (+0.1369)	0.3854 (+0.0603)	0.7780 (+0.2774)	0.6136 (+0.1582)
0.5783	90000	0.0544	0.0527	0.7135 (+0.1731)	0.3787 (+0.0537)	0.7756 (+0.2749)	0.6226 (+0.1672)
0.6426	100000	0.0487	0.0459	0.6979 (+0.1575)	0.3926 (+0.0676)	0.7583 (+0.2576)	0.6163 (+0.1609)
0.7068	110000	0.045	0.0405	0.7091 (+0.1687)	0.3653 (+0.0403)	0.7517 (+0.2511)	0.6087 (+0.1533)
0.7711	120000	0.0403	0.0386	0.6964 (+0.1560)	0.3840 (+0.0590)	0.7767 (+0.2760)	0.6191 (+0.1637)
0.8353	130000	0.036	0.0322	0.7017 (+0.1613)	0.3732 (+0.0482)	0.7575 (+0.2568)	0.6108 (+0.1554)
0.8996	140000	0.0325	0.0283	0.7006 (+0.1602)	0.3814 (+0.0563)	0.7595 (+0.2588)	0.6138 (+0.1585)
0.9639	150000	0.0298	0.0255	0.6920 (+0.1516)	0.3719 (+0.0469)	0.7797 (+0.2791)	0.6146 (+0.1592)
-1	-1	-	-	0.7051 (+0.1647)	0.4148 (+0.0898)	0.7717 (+0.2710)	0.6305 (+0.1752)

The bold row denotes the saved checkpoint.

Framework Versions

Python: 3.9.18
Sentence Transformers: 5.1.1
Transformers: 4.56.2
PyTorch: 2.8.0+cu128
Accelerate: 1.10.1
Datasets: 4.1.1
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",
}