tomaarsen/reranker-ModernBERT-base-gooaq-bce-soft-negs

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
• 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

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("tomaarsen/reranker-ModernBERT-base-gooaq-bce-soft-negs")
# Get scores for pairs of texts
pairs = [
    ['what is the difference between ground level ozone and the ozone layer?', 'Here, ground-level or "bad" ozone is an air pollutant that is harmful to breathe and it damages crops, trees and other vegetation. ... The stratosphere or "good" ozone layer extends upward from about 6 to 30 miles and protects life on Earth from the sun\'s harmful ultraviolet (UV) rays.'],
    ['what is the difference between ground level ozone and the ozone layer?', 'In the stratosphere, temperature increases with altitude. The reason is that the direct heat source for the stratosphere is the Sun. A layer of ozone molecules absorbs solar radiation, which heats the stratosphere.'],
    ['what is the difference between ground level ozone and the ozone layer?', "Atmosphere layers. Earth's atmosphere is divided into five main layers: the exosphere, the thermosphere, the mesosphere, the stratosphere and the troposphere. ... Ozone is abundant here and it heats the atmosphere while also absorbing harmful radiation from the sun."],
    ['what is the difference between ground level ozone and the ozone layer?', "['Water vapor (H. 2O)', 'Carbon dioxide (CO. ... ', 'Methane (CH. ... ', 'Nitrous oxide (N. 2O)', 'Ozone (O. ... ', 'Chlorofluorocarbons (CFCs)', 'Hydrofluorocarbons (includes HCFCs and HFCs)']"],
    ['what is the difference between ground level ozone and the ozone layer?', "Gases in the atmosphere, such as carbon dioxide, trap heat just like the glass roof of a greenhouse. These heat-trapping gases are called greenhouse gases. During the day, the Sun shines through the atmosphere. Earth's surface warms up in the sunlight."],
]
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 difference between ground level ozone and the ozone layer?',
    [
        'Here, ground-level or "bad" ozone is an air pollutant that is harmful to breathe and it damages crops, trees and other vegetation. ... The stratosphere or "good" ozone layer extends upward from about 6 to 30 miles and protects life on Earth from the sun\'s harmful ultraviolet (UV) rays.',
        'In the stratosphere, temperature increases with altitude. The reason is that the direct heat source for the stratosphere is the Sun. A layer of ozone molecules absorbs solar radiation, which heats the stratosphere.',
        "Atmosphere layers. Earth's atmosphere is divided into five main layers: the exosphere, the thermosphere, the mesosphere, the stratosphere and the troposphere. ... Ozone is abundant here and it heats the atmosphere while also absorbing harmful radiation from the sun.",
        "['Water vapor (H. 2O)', 'Carbon dioxide (CO. ... ', 'Methane (CH. ... ', 'Nitrous oxide (N. 2O)', 'Ozone (O. ... ', 'Chlorofluorocarbons (CFCs)', 'Hydrofluorocarbons (includes HCFCs and HFCs)']",
        "Gases in the atmosphere, such as carbon dioxide, trap heat just like the glass roof of a greenhouse. These heat-trapping gases are called greenhouse gases. During the day, the Sun shines through the atmosphere. Earth's surface warms up in the sunlight.",
    ]
)
# [{'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	0.7089 (+0.1778)
mrr@10	0.7076 (+0.1836)
ndcg@10	0.7550 (+0.1637)

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	0.5540 (+0.0644)	0.3421 (+0.0811)	0.6312 (+0.2116)
mrr@10	0.5472 (+0.0697)	0.5284 (+0.0286)	0.6380 (+0.2113)
ndcg@10	0.6229 (+0.0825)	0.3792 (+0.0541)	0.6915 (+0.1908)

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	0.5091 (+0.1190)
mrr@10	0.5712 (+0.1032)
ndcg@10	0.5645 (+0.1092)

Training Details

Training Dataset

Unnamed Dataset

• Size: 482,388 training samples
• Columns: question, answer, and label
• Approximate statistics based on the first 1000 samples:

  	question	answer	label
  type	string	string	int
  details	min: 17 characters mean: 43.7 characters max: 91 characters	min: 53 characters mean: 250.44 characters max: 393 characters	0: ~79.30% 1: ~20.70%

• Samples:

  question	answer	label
  what is the difference between ground level ozone and the ozone layer?	Here, ground-level or "bad" ozone is an air pollutant that is harmful to breathe and it damages crops, trees and other vegetation. ... The stratosphere or "good" ozone layer extends upward from about 6 to 30 miles and protects life on Earth from the sun's harmful ultraviolet (UV) rays.	1
  what is the difference between ground level ozone and the ozone layer?	In the stratosphere, temperature increases with altitude. The reason is that the direct heat source for the stratosphere is the Sun. A layer of ozone molecules absorbs solar radiation, which heats the stratosphere.	0
  what is the difference between ground level ozone and the ozone layer?	Atmosphere layers. Earth's atmosphere is divided into five main layers: the exosphere, the thermosphere, the mesosphere, the stratosphere and the troposphere. ... Ozone is abundant here and it heats the atmosphere while also absorbing harmful radiation from the sun.	0

• Loss: BinaryCrossEntropyLoss with these parameters:

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

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}
• 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	gooaq-dev_ndcg@10	NanoMSMARCO_R100_ndcg@10	NanoNFCorpus_R100_ndcg@10	NanoNQ_R100_ndcg@10	NanoBEIR_R100_mean_ndcg@10
-1	-1	-	0.1488 (-0.4424)	0.0573 (-0.4832)	0.2647 (-0.0604)	0.0388 (-0.4619)	0.1202 (-0.3351)
0.0001	1	1.3143	-	-	-	-	-
0.0265	200	1.2539	-	-	-	-	-
0.0531	400	0.9497	-	-	-	-	-
0.0796	600	0.5613	-	-	-	-	-
0.1061	800	0.4687	-	-	-	-	-
0.1327	1000	0.4042	0.7103 (+0.1191)	0.5262 (-0.0142)	0.3298 (+0.0048)	0.5589 (+0.0583)	0.4717 (+0.0163)
0.1592	1200	0.3562	-	-	-	-	-
0.1857	1400	0.3543	-	-	-	-	-
0.2123	1600	0.3467	-	-	-	-	-
0.2388	1800	0.3153	-	-	-	-	-
0.2653	2000	0.3033	0.7317 (+0.1405)	0.5662 (+0.0258)	0.3859 (+0.0609)	0.6828 (+0.1822)	0.5450 (+0.0896)
0.2919	2200	0.2986	-	-	-	-	-
0.3184	2400	0.3016	-	-	-	-	-
0.3449	2600	0.2984	-	-	-	-	-
0.3715	2800	0.2646	-	-	-	-	-
0.3980	3000	0.3048	0.7359 (+0.1447)	0.5713 (+0.0309)	0.3987 (+0.0736)	0.6960 (+0.1953)	0.5553 (+0.1000)
0.4245	3200	0.2714	-	-	-	-	-
0.4510	3400	0.2773	-	-	-	-	-
0.4776	3600	0.2621	-	-	-	-	-
0.5041	3800	0.2529	-	-	-	-	-
0.5306	4000	0.2533	0.7459 (+0.1546)	0.5893 (+0.0489)	0.3887 (+0.0637)	0.6749 (+0.1743)	0.5510 (+0.0956)
0.5572	4200	0.2822	-	-	-	-	-
0.5837	4400	0.2299	-	-	-	-	-
0.6102	4600	0.2554	-	-	-	-	-
0.6368	4800	0.2373	-	-	-	-	-
0.6633	5000	0.2248	0.7497 (+0.1584)	0.6110 (+0.0706)	0.3782 (+0.0531)	0.6885 (+0.1878)	0.5592 (+0.1038)
0.6898	5200	0.2315	-	-	-	-	-
0.7164	5400	0.2313	-	-	-	-	-
0.7429	5600	0.2294	-	-	-	-	-
0.7694	5800	0.2384	-	-	-	-	-
0.7960	6000	0.2195	0.7530 (+0.1617)	0.6249 (+0.0845)	0.3873 (+0.0623)	0.6773 (+0.1766)	0.5632 (+0.1078)
0.8225	6200	0.2047	-	-	-	-	-
0.8490	6400	0.2192	-	-	-	-	-
0.8756	6600	0.1926	-	-	-	-	-
0.9021	6800	0.2185	-	-	-	-	-
0.9286	7000	0.2365	0.7550 (+0.1637)	0.6229 (+0.0825)	0.3792 (+0.0541)	0.6915 (+0.1908)	0.5645 (+0.1092)
0.9552	7200	0.2173	-	-	-	-	-
0.9817	7400	0.2249	-	-	-	-	-
-1	-1	-	0.7550 (+0.1637)	0.6229 (+0.0825)	0.3792 (+0.0541)	0.6915 (+0.1908)	0.5645 (+0.1092)

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.11.10
• Sentence Transformers: 3.5.0.dev0
• Transformers: 4.49.0
• PyTorch: 2.5.1+cu124
• Accelerate: 1.5.2
• Datasets: 2.21.0
• Tokenizers: 0.21.0

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",
}