Add new CrossEncoder model

e239c5f verified 5 months ago

21.2 kB

language:
  - multilingual
license: cc-by-nc-4.0
tags:
  - sentence-transformers
  - cross-encoder
  - reranker
  - generated_from_trainer
  - dataset_size:170000
  - loss:BinaryCrossEntropyLoss
base_model: jinaai/jina-reranker-v2-base-multilingual
pipeline_tag: text-ranking
library_name: sentence-transformers
metrics:
  - map
  - mrr@10
  - ndcg@10
model-index:
  - name: cometadata/jina-reranker-v2-multilingual-affiliations-large
    results:
      - task:
          type: cross-encoder-reranking
          name: Cross Encoder Reranking
        dataset:
          name: affiliation val
          type: affiliation-val
        metrics:
          - type: map
            value: 0.988
            name: Map
          - type: mrr@10
            value: 0.988
            name: Mrr@10
          - type: ndcg@10
            value: 0.9933
            name: Ndcg@10

cometadata/jina-reranker-v2-multilingual-affiliations-large

This is a Cross Encoder model finetuned from jinaai/jina-reranker-v2-base-multilingual 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: jinaai/jina-reranker-v2-base-multilingual
Maximum Sequence Length: 1024 tokens
Number of Output Labels: 1 label
Language: multilingual
License: cc-by-nc-4.0

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("cometadata/jina-reranker-v2-multilingual-affiliations-large")
# Get scores for pairs of texts
pairs = [
    ['Instituto Multidisciplinar para el Estudio del Medio ‘Ramón Margalef’, Universidad de Alicante, Alicante, Spain', 'Departamento de Matemática Aplicada, Universidad de Alicante, San Vicente del Raspeig (Alicante), España'],
    ['Instituto Multidisciplinar para el Estudio del Medio ‘Ramón Margalef’, Universidad de Alicante, Alicante, Spain', 'Research Institute of Physics and Aerospace Science, University of Vigo, Vigo, Spain'],
    ['Departamento de Patologia Básica, Setor de Ciências Biológicas, Universidade Federal do Paraná, 81531-970 Curitiba, PR, Brasil', 'Laboratory of Hematology, Department of Medical Pathology, Federal University of Paraná, Curitiba, Brazil'],
    ['Departamento de Patologia Básica, Setor de Ciências Biológicas, Universidade Federal do Paraná, 81531-970 Curitiba, PR, Brasil', 'Laboratório de Patologia Experimental Pontifícia Universidade Católica do Paraná Curitiba Brazil'],
    ['Institute of Information & Control, Hangzhou Dianzi University, Hangzhou 310018, P.R. China', 'College of Media &amp; Design Hangzhou Dianzi University Hangzhou 310018 China'],
]
scores = model.predict(pairs)
print(scores.shape)
# (5,)

# Or rank different texts based on similarity to a single text
ranks = model.rank(
    'Instituto Multidisciplinar para el Estudio del Medio ‘Ramón Margalef’, Universidad de Alicante, Alicante, Spain',
    [
        'Departamento de Matemática Aplicada, Universidad de Alicante, San Vicente del Raspeig (Alicante), España',
        'Research Institute of Physics and Aerospace Science, University of Vigo, Vigo, Spain',
        'Laboratory of Hematology, Department of Medical Pathology, Federal University of Paraná, Curitiba, Brazil',
        'Laboratório de Patologia Experimental Pontifícia Universidade Católica do Paraná Curitiba Brazil',
        'College of Media &amp; Design Hangzhou Dianzi University Hangzhou 310018 China',
    ]
)
# [{'corpus_id': ..., 'score': ...}, {'corpus_id': ..., 'score': ...}, ...]

Evaluation

Metrics

Cross Encoder Reranking

Dataset: affiliation-val

Evaluated with CrossEncoderRerankingEvaluator with these parameters:

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

Metric	Value
map	0.9880 (-0.0120)
mrr@10	0.9880 (-0.0120)
ndcg@10	0.9933 (-0.0067)

Training Details

Training Dataset

Unnamed Dataset

Size: 170,000 training samples
Columns: query, document, and label

Approximate statistics based on the first 1000 samples:

	query	document	label
type	string	string	int
details	min: 22 characters mean: 89.21 characters max: 209 characters	min: 25 characters mean: 101.13 characters max: 279 characters	0: ~50.00% 1: ~50.00%

Samples:

query	document	label
`Max-Planck-Institut für Astronomie, Königgstuhl 17, D-69117 Heidelberg, Germany`	`Max-Planck-Institute for Astronomy, Königstuhl 17, 69117 Heidelberg, Germany e-mail: beuther@mpia.de`	`1`
`Max-Planck-Institut für Astronomie, Königgstuhl 17, D-69117 Heidelberg, Germany`	`Clinical Trials Center Cardiovascular Research Foundation New York City NY USA`	`0`
`Stowers Institute for Medical Research, 64110, Kansas City, Missouri, USA`	`Stowers Institute for Medical Research, 1,000 East 50th Street, Kansas City, MO 64110, USA`	`1`

Loss: BinaryCrossEntropyLoss with these parameters:

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

Evaluation Dataset

Unnamed Dataset

Size: 30,000 evaluation samples
Columns: query, document, and label

Approximate statistics based on the first 1000 samples:

	query	document	label
type	string	string	int
details	min: 28 characters mean: 113.4 characters max: 298 characters	min: 24 characters mean: 104.23 characters max: 272 characters	0: ~50.00% 1: ~50.00%

Samples:

query	document	label
`Instituto Multidisciplinar para el Estudio del Medio ‘Ramón Margalef’, Universidad de Alicante, Alicante, Spain`	`Departamento de Matemática Aplicada, Universidad de Alicante, San Vicente del Raspeig (Alicante), España`	`1`
`Instituto Multidisciplinar para el Estudio del Medio ‘Ramón Margalef’, Universidad de Alicante, Alicante, Spain`	`Research Institute of Physics and Aerospace Science, University of Vigo, Vigo, Spain`	`0`
`Departamento de Patologia Básica, Setor de Ciências Biológicas, Universidade Federal do Paraná, 81531-970 Curitiba, PR, Brasil`	`Laboratory of Hematology, Department of Medical Pathology, Federal University of Paraná, Curitiba, Brazil`	`1`

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: 128
per_device_eval_batch_size: 128
learning_rate: 3e-05
warmup_ratio: 0.1
bf16: True
load_best_model_at_end: True
push_to_hub: True
hub_model_id: cometadata/jina-reranker-v2-multilingual-affiliations-large

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
eval_strategy: steps
prediction_loss_only: True
per_device_train_batch_size: 128
per_device_eval_batch_size: 128
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: 3e-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: 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: 42
data_seed: None
jit_mode_eval: 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}
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: True
resume_from_checkpoint: None
hub_model_id: cometadata/jina-reranker-v2-multilingual-affiliations-large
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
multi_dataset_batch_sampler: proportional
router_mapping: {}
learning_rate_mapping: {}

Training Logs

Epoch	Step	Training Loss	Validation Loss	affiliation-val_ndcg@10
-1	-1	-	-	0.9200 (-0.0800)
0.0008	1	0.1129	-	-
0.0752	100	0.3049	-	-
0.1505	200	0.1295	-	-
0.1881	250	-	0.6259	0.9715 (-0.0285)
0.2257	300	0.1076	-	-
0.3010	400	0.0978	-	-
0.3762	500	0.1031	0.2822	0.9871 (-0.0129)
0.4515	600	0.0932	-	-
0.5267	700	0.1015	-	-
0.5643	750	-	0.2395	0.9890 (-0.0110)
0.6020	800	0.0999	-	-
0.6772	900	0.1112	-	-
0.7524	1000	0.1196	0.1980	0.9921 (-0.0079)
0.8277	1100	0.1288	-	-
0.9029	1200	0.1295	-	-
0.9406	1250	-	0.1773	0.9929 (-0.0071)
0.9782	1300	0.1338	-	-
1.0534	1400	0.0585	-	-
1.1287	1500	0.0295	0.3412	0.9879 (-0.0121)
1.2039	1600	0.0412	-	-
1.2792	1700	0.0491	-	-
1.3168	1750	-	0.2622	0.9903 (-0.0097)
1.3544	1800	0.0619	-	-
1.4296	1900	0.0612	-	-
1.5049	2000	0.0676	0.2131	0.9919 (-0.0081)
1.5801	2100	0.073	-	-
1.6554	2200	0.0801	-	-
1.6930	2250	-	0.1940	0.9927 (-0.0073)
1.7306	2300	0.0963	-	-
1.8059	2400	0.1114	-	-
1.8811	2500	0.1083	0.1773	0.9933 (-0.0067)
1.9564	2600	0.1203	-	-
2.0316	2700	0.0841	-	-
2.0692	2750	-	0.2898	0.9907 (-0.0093)
2.1068	2800	0.0248	-	-
2.1821	2900	0.032	-	-
2.2573	3000	0.0468	0.2455	0.9915 (-0.0085)
2.3326	3100	0.0497	-	-
2.4078	3200	0.0585	-	-
2.4454	3250	-	0.2142	0.9921 (-0.0079)
2.4831	3300	0.0653	-	-
2.5583	3400	0.0701	-	-
2.6336	3500	0.0758	0.2034	0.9924 (-0.0076)
2.7088	3600	0.0903	-	-
2.7840	3700	0.1037	-	-
2.8217	3750	-	0.1984	0.9927 (-0.0073)
2.8593	3800	0.113	-	-
2.9345	3900	0.1199	-	-
-1	-1	-	-	0.9933 (-0.0067)

The bold row denotes the saved checkpoint.

Framework Versions

Python: 3.12.12
Sentence Transformers: 5.2.0
Transformers: 4.57.3
PyTorch: 2.9.1+cu128
Accelerate: 1.12.0
Datasets: 4.4.2
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",
}