kwondw/reranker-ModernBERT-base-gooaq-bce

ModernBERT-base trained on GooAQ

This is a Cross Encoder model finetuned from answerdotai/ModernBERT-base 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: answerdotai/ModernBERT-base
• Maximum Sequence Length: 8192 tokens
• Number of Output Labels: 1 label
• Supported Modality: Text
• Language: en
• License: apache-2.0

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': 'ModernBertForSequenceClassification'})
)

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("kwondw/reranker-ModernBERT-base-gooaq-bce")
# Get scores for pairs of inputs
pairs = [
    ['what to spend my nectar points on?', "There are lots of different places and ways you can spend your points. You can spend straight from your card by swiping it at Sainsbury's, Vue Cinemas and Argos or you can spend them online at nectar.com on great things like holidays, hotels and days out."],
    ['what to spend my nectar points on?', "If you have already registered for Online Services, just log in to see your points balance. For detailed points transactions, go to Membership Rewards page and click on 'Your points Account' and then 'My Points' at the top of the page."],
    ['what to spend my nectar points on?', 'The SDIP authorized insurance rate increase are as follows: 1 point=25%, 2 points=45%, 3 points=65%, 4 points=90%, 5 points=120%, 6 points=150%, 7 points=180%, 8 points=220%, 9 points=260%, 10 points=300%, 11 points=350%, and 12 insurance points equals a 400% increase in your insurance premiums, ouch!'],
    ['what to spend my nectar points on?', 'To redeem your Amex Blue Sky points, log in to your account via the Amex travel portal. At the checkout, there will be an option to pay with points. Points are worth 1 cent each toward flights and 0.7 cents each toward other travel costs.'],
    ['what to spend my nectar points on?', 'Yes, TD rewards points are worth $0.005 each, and must be redeemed in increments of 10,000, at a value of $50 per 10,000 points.'],
]
scores = model.predict(pairs)
print(scores)
# [9.7463e-01 2.7169e-02 1.4214e-03 5.0407e-03 4.4422e-04]

# Or rank different texts based on similarity to a single text
ranks = model.rank(
    'what to spend my nectar points on?',
    [
        "There are lots of different places and ways you can spend your points. You can spend straight from your card by swiping it at Sainsbury's, Vue Cinemas and Argos or you can spend them online at nectar.com on great things like holidays, hotels and days out.",
        "If you have already registered for Online Services, just log in to see your points balance. For detailed points transactions, go to Membership Rewards page and click on 'Your points Account' and then 'My Points' at the top of the page.",
        'The SDIP authorized insurance rate increase are as follows: 1 point=25%, 2 points=45%, 3 points=65%, 4 points=90%, 5 points=120%, 6 points=150%, 7 points=180%, 8 points=220%, 9 points=260%, 10 points=300%, 11 points=350%, and 12 insurance points equals a 400% increase in your insurance premiums, ouch!',
        'To redeem your Amex Blue Sky points, log in to your account via the Amex travel portal. At the checkout, there will be an option to pay with points. Points are worth 1 cent each toward flights and 0.7 cents each toward other travel costs.',
        'Yes, TD rewards points are worth $0.005 each, and must be redeemed in increments of 10,000, at a value of $50 per 10,000 points.',
    ]
)
# [{'corpus_id': ..., 'score': ...}, {'corpus_id': ..., 'score': ...}, ...]

Evaluation

Metrics

Cross Encoder Reranking

• Dataset: gooaq-dev
• Evaluated with CrossEncoderRerankingEvaluator with these parameters:

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

Metric	Value
map	0.7842 (+0.1739)
mrr@10	0.7827 (+0.1773)
ndcg@10	0.8182 (+0.1499)

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	0.4839 (-0.0056)	0.3422 (+0.0812)	0.6052 (+0.1856)
mrr@10	0.4753 (-0.0022)	0.5542 (+0.0544)	0.6189 (+0.1922)
ndcg@10	0.5681 (+0.0277)	0.3908 (+0.0658)	0.6657 (+0.1651)

Cross Encoder Nano BEIR

• Dataset: NanoBEIR_R100_mean
• Evaluated with CrossEncoderNanoBEIREvaluator with these parameters:

  {
      "dataset_names": [
          "msmarco",
          "nfcorpus",
          "nq"
      ],
      "dataset_id": "sentence-transformers/NanoBEIR-en",
      "rerank_k": 100,
      "at_k": 10,
      "always_rerank_positives": true
  }
  

Metric	Value
map	0.4771 (+0.0870)
mrr@10	0.5495 (+0.0814)
ndcg@10	0.5415 (+0.0862)

Training Details

Training Dataset

Unnamed Dataset

• Size: 292,161 training samples
• Columns: question, answer, and label
• Approximate statistics based on the first 100 samples:

  	question	answer	label
  type	string	string	int
  modality	text	text
  details	min: 8 tokens mean: 11.17 tokens max: 14 tokens	min: 23 tokens mean: 57.12 tokens max: 115 tokens	0: ~82.69% 1: ~17.31%

• Samples:

  question	answer	label
  what to spend my nectar points on?	There are lots of different places and ways you can spend your points. You can spend straight from your card by swiping it at Sainsbury's, Vue Cinemas and Argos or you can spend them online at nectar.com on great things like holidays, hotels and days out.	1
  what to spend my nectar points on?	If you have already registered for Online Services, just log in to see your points balance. For detailed points transactions, go to Membership Rewards page and click on 'Your points Account' and then 'My Points' at the top of the page.	0
  what to spend my nectar points on?	The SDIP authorized insurance rate increase are as follows: 1 point=25%, 2 points=45%, 3 points=65%, 4 points=90%, 5 points=120%, 6 points=150%, 7 points=180%, 8 points=220%, 9 points=260%, 10 points=300%, 11 points=350%, and 12 insurance points equals a 400% increase in your insurance premiums, ouch!	0

• Loss: BinaryCrossEntropyLoss with these parameters:

  {
      "activation_fn": "torch.nn.modules.linear.Identity",
      "pos_weight": 5
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• per_device_train_batch_size: 64
• num_train_epochs: 1
• learning_rate: 2e-05
• warmup_steps: 0.1
• fp16: True
• per_device_eval_batch_size: 64
• load_best_model_at_end: True
• seed: 12
• dataloader_num_workers: 2

All Hyperparameters

Click to expand

• per_device_train_batch_size: 64
• num_train_epochs: 1
• max_steps: -1
• learning_rate: 2e-05
• lr_scheduler_type: linear
• lr_scheduler_kwargs: None
• warmup_steps: 0.1
• 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: 1.0
• label_smoothing_factor: 0.0
• bf16: False
• fp16: True
• 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: 64
• 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: True
• ignore_data_skip: False
• restore_callback_states_from_checkpoint: False
• full_determinism: False
• seed: 12
• 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: 2
• dataloader_pin_memory: True
• dataloader_persistent_workers: False
• dataloader_prefetch_factor: None
• 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
• warmup_ratio: None
• local_rank: -1
• prompts: None
• batch_sampler: batch_sampler
• multi_dataset_batch_sampler: proportional
• router_mapping: {}
• learning_rate_mapping: {}

Training Logs

Epoch	Step	Training Loss	gooaq-dev_ndcg@10	NanoMSMARCO_R100_ndcg@10	NanoNFCorpus_R100_ndcg@10	NanoNQ_R100_ndcg@10	NanoBEIR_R100_mean_ndcg@10
-1	-1	-	0.1149 (-0.5534)	0.0212 (-0.5192)	0.2526 (-0.0724)	0.0288 (-0.4719)	0.1009 (-0.3545)
0.0002	1	1.3195	-	-	-	-	-
0.0438	200	1.1767	-	-	-	-	-
0.0876	400	1.0007	-	-	-	-	-
0.1095	500	-	0.7463 (+0.0779)	0.4619 (-0.0785)	0.3117 (-0.0133)	0.5694 (+0.0688)	0.4477 (-0.0077)
0.1314	600	0.6751	-	-	-	-	-
0.1752	800	0.6153	-	-	-	-	-
0.2190	1000	0.5813	0.7963 (+0.1279)	0.5554 (+0.0150)	0.3518 (+0.0267)	0.6018 (+0.1012)	0.5030 (+0.0476)
0.2628	1200	0.5345	-	-	-	-	-
0.3066	1400	0.5544	-	-	-	-	-
0.3285	1500	-	0.8035 (+0.1351)	0.5196 (-0.0208)	0.3772 (+0.0522)	0.6118 (+0.1111)	0.5029 (+0.0475)
0.3504	1600	0.5324	-	-	-	-	-
0.3942	1800	0.5199	-	-	-	-	-
0.4380	2000	0.5015	0.8045 (+0.1362)	0.5731 (+0.0326)	0.3638 (+0.0387)	0.6564 (+0.1558)	0.5311 (+0.0757)
0.4818	2200	0.4899	-	-	-	-	-
0.5256	2400	0.4703	-	-	-	-	-
0.5475	2500	-	0.8030 (+0.1346)	0.5821 (+0.0417)	0.3820 (+0.0569)	0.6695 (+0.1688)	0.5445 (+0.0892)
0.5694	2600	0.4863	-	-	-	-	-
0.6132	2800	0.4823	-	-	-	-	-
0.6570	3000	0.4656	0.8066 (+0.1382)	0.5458 (+0.0054)	0.3882 (+0.0632)	0.6386 (+0.1379)	0.5242 (+0.0688)
0.7008	3200	0.4689	-	-	-	-	-
0.7446	3400	0.4595	-	-	-	-	-
0.7665	3500	-	0.8146 (+0.1462)	0.5861 (+0.0456)	0.3904 (+0.0654)	0.6578 (+0.1572)	0.5448 (+0.0894)
0.7884	3600	0.4484	-	-	-	-	-
0.8322	3800	0.4597	-	-	-	-	-
0.876	4000	0.4482	0.8182 (+0.1499)	0.5681 (+0.0277)	0.3908 (+0.0658)	0.6657 (+0.1651)	0.5415 (+0.0862)
0.9198	4200	0.4488	-	-	-	-	-
0.9636	4400	0.4548	-	-	-	-	-
0.9855	4500	-	0.8168 (+0.1484)	0.5612 (+0.0208)	0.4006 (+0.0755)	0.6864 (+0.1858)	0.5494 (+0.0940)
1.0	4566	-	0.8166 (+0.1482)	0.5640 (+0.0236)	0.4006 (+0.0755)	0.6838 (+0.1832)	0.5495 (+0.0941)
-1	-1	-	0.8182 (+0.1499)	0.5681 (+0.0277)	0.3908 (+0.0658)	0.6657 (+0.1651)	0.5415 (+0.0862)

• The bold row denotes the saved checkpoint.

Training Time

• Training: 1.5 hours

Framework Versions

• Python: 3.12.13
• Sentence Transformers: 5.6.0
• Transformers: 5.13.1
• PyTorch: 2.11.0+cu128
• Accelerate: 1.14.0
• Datasets: 4.0.0
• Tokenizers: 0.22.2

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",
}