akr2002/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
• 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("akr2002/reranker-ModernBERT-base-gooaq-bce")
# Get scores for pairs of texts
pairs = [
    ['how do you find mass?', "Divide the object's weight by the acceleration of gravity to find the mass. You'll need to convert the weight units to Newtons. For example, 1 kg = 9.807 N. If you're measuring the mass of an object on Earth, divide the weight in Newtons by the acceleration of gravity on Earth (9.8 meters/second2) to get mass."],
    ['how do you find mass?', "In general use, 'High Mass' means a full ceremonial Mass, most likely with music, and also with incense if they're particularly traditional. ... Incense is used quite a lot. Low Mass in the traditional rite is celebrated by one priest, and usually only one or two altar servers."],
    ['how do you find mass?', 'A neutron has a slightly larger mass than the proton. These are often given in terms of an atomic mass unit, where one atomic mass unit (u) is defined as 1/12th the mass of a carbon-12 atom. You can use that to prove that a mass of 1 u is equivalent to an energy of 931.5 MeV.'],
    ['how do you find mass?', 'Mass is the amount of matter in a body, normally measured in grams or kilograms etc. Weight is a force that pulls on a mass and is measured in Newtons. ... Density basically means how much mass is occupied in a specific volume or space. Different materials of the same size may have different masses because of its density.'],
    ['how do you find mass?', 'Receiver – Mass communication is the transmission of the message to a large number of recipients. This mass of receivers, are often called as mass audience. The Mass audience is large, heterogenous and anonymous in nature. The receivers are scattered across a given village, state or country.'],
]
scores = model.predict(pairs)
print(scores.shape)
# (5,)

# Or rank different texts based on similarity to a single text
ranks = model.rank(
    'how do you find mass?',
    [
        "Divide the object's weight by the acceleration of gravity to find the mass. You'll need to convert the weight units to Newtons. For example, 1 kg = 9.807 N. If you're measuring the mass of an object on Earth, divide the weight in Newtons by the acceleration of gravity on Earth (9.8 meters/second2) to get mass.",
        "In general use, 'High Mass' means a full ceremonial Mass, most likely with music, and also with incense if they're particularly traditional. ... Incense is used quite a lot. Low Mass in the traditional rite is celebrated by one priest, and usually only one or two altar servers.",
        'A neutron has a slightly larger mass than the proton. These are often given in terms of an atomic mass unit, where one atomic mass unit (u) is defined as 1/12th the mass of a carbon-12 atom. You can use that to prove that a mass of 1 u is equivalent to an energy of 931.5 MeV.',
        'Mass is the amount of matter in a body, normally measured in grams or kilograms etc. Weight is a force that pulls on a mass and is measured in Newtons. ... Density basically means how much mass is occupied in a specific volume or space. Different materials of the same size may have different masses because of its density.',
        'Receiver – Mass communication is the transmission of the message to a large number of recipients. This mass of receivers, are often called as mass audience. The Mass audience is large, heterogenous and anonymous in nature. The receivers are scattered across a given village, state or country.',
    ]
)
# [{'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	0.7258 (+0.1946)
mrr@10	0.7245 (+0.2005)
ndcg@10	0.7686 (+0.1774)

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	0.4807 (-0.0089)	0.3866 (+0.1256)	0.5595 (+0.1399)
mrr@10	0.4689 (-0.0086)	0.6058 (+0.1060)	0.5752 (+0.1485)
ndcg@10	0.5499 (+0.0095)	0.4233 (+0.0982)	0.6191 (+0.1184)

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	0.4756 (+0.0855)
mrr@10	0.5500 (+0.0820)
ndcg@10	0.5308 (+0.0754)

Training Details

Training Dataset

Unnamed Dataset

• Size: 578,402 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: 44.75 characters max: 84 characters	min: 54 characters mean: 252.51 characters max: 388 characters	0: ~83.00% 1: ~17.00%

• Samples:

  question	answer	label
  how do you find mass?	Divide the object's weight by the acceleration of gravity to find the mass. You'll need to convert the weight units to Newtons. For example, 1 kg = 9.807 N. If you're measuring the mass of an object on Earth, divide the weight in Newtons by the acceleration of gravity on Earth (9.8 meters/second2) to get mass.	1
  how do you find mass?	In general use, 'High Mass' means a full ceremonial Mass, most likely with music, and also with incense if they're particularly traditional. ... Incense is used quite a lot. Low Mass in the traditional rite is celebrated by one priest, and usually only one or two altar servers.	0
  how do you find mass?	A neutron has a slightly larger mass than the proton. These are often given in terms of an atomic mass unit, where one atomic mass unit (u) is defined as 1/12th the mass of a carbon-12 atom. You can use that to prove that a mass of 1 u is equivalent to an energy of 931.5 MeV.	0

• Loss: BinaryCrossEntropyLoss with these parameters:

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

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• per_device_train_batch_size: 16
• per_device_eval_batch_size: 16
• 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: 16
• per_device_eval_batch_size: 16
• 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}
• 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	gooaq-dev_ndcg@10	NanoMSMARCO_R100_ndcg@10	NanoNFCorpus_R100_ndcg@10	NanoNQ_R100_ndcg@10	NanoBEIR_R100_mean_ndcg@10
-1	-1	-	0.1474 (-0.4438)	0.0356 (-0.5048)	0.2344 (-0.0907)	0.0268 (-0.4739)	0.0989 (-0.3564)
0.0000	1	1.1353	-	-	-	-	-
0.0277	1000	1.1797	-	-	-	-	-
0.0553	2000	0.8539	-	-	-	-	-
0.0830	3000	0.7438	-	-	-	-	-
0.1106	4000	0.7296	0.7119 (+0.1206)	0.5700 (+0.0296)	0.3410 (+0.0160)	0.6012 (+0.1005)	0.5041 (+0.0487)
0.1383	5000	0.6705	-	-	-	-	-
0.1660	6000	0.6624	-	-	-	-	-
0.1936	7000	0.6685	-	-	-	-	-
0.2213	8000	0.6305	0.7328 (+0.1415)	0.5504 (+0.0099)	0.4056 (+0.0805)	0.6947 (+0.1941)	0.5502 (+0.0948)
0.2490	9000	0.6353	-	-	-	-	-
0.2766	10000	0.6118	-	-	-	-	-
0.3043	11000	0.6097	-	-	-	-	-
0.3319	12000	0.6003	0.7423 (+0.1510)	0.5817 (+0.0413)	0.3817 (+0.0566)	0.6152 (+0.1145)	0.5262 (+0.0708)
0.3596	13000	0.5826	-	-	-	-	-
0.3873	14000	0.5935	-	-	-	-	-
0.4149	15000	0.5826	-	-	-	-	-
0.4426	16000	0.5723	0.7557 (+0.1645)	0.5453 (+0.0049)	0.4029 (+0.0779)	0.6260 (+0.1253)	0.5247 (+0.0693)
0.4702	17000	0.582	-	-	-	-	-
0.4979	18000	0.5631	-	-	-	-	-
0.5256	19000	0.5705	-	-	-	-	-
0.5532	20000	0.544	0.7604 (+0.1692)	0.5636 (+0.0232)	0.4112 (+0.0862)	0.6260 (+0.1253)	0.5336 (+0.0782)
0.5809	21000	0.5289	-	-	-	-	-
0.6086	22000	0.5431	-	-	-	-	-
0.6362	23000	0.5449	-	-	-	-	-
0.6639	24000	0.5338	0.7608 (+0.1696)	0.5384 (-0.0020)	0.4327 (+0.1077)	0.5906 (+0.0899)	0.5206 (+0.0652)
0.6915	25000	0.5401	-	-	-	-	-
0.7192	26000	0.5535	-	-	-	-	-
0.7469	27000	0.5353	-	-	-	-	-
0.7745	28000	0.5157	0.7635 (+0.1723)	0.5217 (-0.0188)	0.4171 (+0.0921)	0.5543 (+0.0537)	0.4977 (+0.0423)
0.8022	29000	0.5153	-	-	-	-	-
0.8299	30000	0.5122	-	-	-	-	-
0.8575	31000	0.5108	-	-	-	-	-
0.8852	32000	0.5303	0.7685 (+0.1773)	0.5538 (+0.0134)	0.4147 (+0.0897)	0.6155 (+0.1149)	0.5280 (+0.0727)
0.9128	33000	0.5363	-	-	-	-	-
0.9405	34000	0.4996	-	-	-	-	-
0.9682	35000	0.5193	-	-	-	-	-
0.9958	36000	0.4995	0.7686 (+0.1774)	0.5499 (+0.0095)	0.4233 (+0.0982)	0.6191 (+0.1184)	0.5308 (+0.0754)
-1	-1	-	0.7686 (+0.1774)	0.5499 (+0.0095)	0.4233 (+0.0982)	0.6191 (+0.1184)	0.5308 (+0.0754)

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.12.7
• Sentence Transformers: 4.0.1
• Transformers: 4.50.3
• PyTorch: 2.6.0+cu124
• Accelerate: 1.5.2
• Datasets: 3.5.0
• Tokenizers: 0.21.1

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",
}