tomaarsen/reranker-ModernBERT-base-gooaq-lambda

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-lambda")
# Get scores for pairs of texts
pairs = [
    ['How many calories in an egg', 'There are on average between 55 and 80 calories in an egg depending on its size.'],
    ['How many calories in an egg', 'Egg whites are very low in calories, have no fat, no cholesterol, and are loaded with protein.'],
    ['How many calories in an egg', 'Most of the calories in an egg come from the yellow yolk in the center.'],
]
scores = model.predict(pairs)
print(scores.shape)
# (3,)

# Or rank different texts based on similarity to a single text
ranks = model.rank(
    'How many calories in an egg',
    [
        'There are on average between 55 and 80 calories in an egg depending on its size.',
        'Egg whites are very low in calories, have no fat, no cholesterol, and are loaded with protein.',
        'Most of the calories in an egg come from the yellow yolk in the center.',
    ]
)
# [{'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	0.7164 (+0.1853)
mrr@10	0.7148 (+0.1908)
ndcg@10	0.7601 (+0.1689)

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	0.4853 (-0.0042)	0.3379 (+0.0769)	0.5390 (+0.1194)
mrr@10	0.4772 (-0.0003)	0.5293 (+0.0294)	0.5479 (+0.1212)
ndcg@10	0.5514 (+0.0110)	0.3714 (+0.0464)	0.5941 (+0.0934)

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	0.4541 (+0.0640)
mrr@10	0.5181 (+0.0501)
ndcg@10	0.5056 (+0.0503)

Training Details

Training Dataset

Unnamed Dataset

• Size: 95,939 training samples
• Columns: question, answer, and labels
• Approximate statistics based on the first 1000 samples:

  	question	answer	labels
  type	string	list	list
  details	min: 18 characters mean: 43.5 characters max: 101 characters	size: 6 elements	size: 6 elements

• Samples:

  question	answer	labels
  can u get ip banned from discord?	['Yes you very much can, infact its already done. When you ban a person its an IP ban (also an account ban) There are no ways to bypass it without a new account.', 'Yes, your account is banned if you see the “Your account has been suspended/terminated for violating the Terms of Service” message when logging in to Pokémon GO.', 'This means that Snap is identifying devices and not users. So if a user, after getting banned, tries to access Snapchat from a different account but the same device, then that account also gets banned automatically. “The jailbreaking ban is apparently actually a device ban.', "When you block someone on Discord, they won't be able to send you private messages, and will servers you share will hide their messages. If the person you blocked was on your Friends list, they'll be removed immediately.", "You will for sure get an e-mail telling you that you were banned. That error happens quite often to me. Just login again from the title screen and game on. It's a commo...	[1, 0, 0, 0, 0, ...]
  what is the difference between methylphenidate cd and er?	['Metadate CD is a once-a-day capsule with biphasic release; initially there is a rapid release of methylphenidate, then a continuous-release phase. Metadate ER, on the other hand, is a tablet given two to three times per day.', 'Irregular Heartbeat Risk Associated with Common ADHD Med. Children who take a common drug to treat attention-deficit/hyperactivity disorder may be at an increased risk for developing an irregular heartbeat. The drug, methylphenidate, is the active ingredient in Concerta, Daytrana and Ritalin.', "Vyvanse contains the drug lisdexamfetamine dimesylate, while Ritalin contains the drug methylphenidate. Both Vyvanse and Ritalin are used to treat ADHD symptoms such as poor focus, reduced impulse control, and hyperactivity. However, they're also prescribed to treat other conditions.", 'Tolerance develops to the side effects of Adderall IR and XR in five to seven days. Side effects that persist longer than one week can be quickly managed by lowering the dose or changin...	[1, 0, 0, 0, 0, ...]
  who has the most championships in hockey?	['Having lifted the trophy a total of 24 times, the Montreal Canadiens are the team with more Stanley Cup titles than any other franchise.', "['Ivy League – 46 National Championships.', 'Big Ten – 39 National Championships. ... ', 'SEC – 29 National Championships. ... ', 'ACC – 18 National Championships. ... ', 'Independents – 17 National Championships. ... ', 'Pac-12 – 15 National Championships. ... ', 'Big 12 – 11 National Championships. ... ']", 'Boston Celtics center Bill Russell holds the record for the most NBA championships won with 11 titles during his 13-year playing career.', 'Alabama can claim the most NCAA titles in the poll era, with only three of its 15 coming prior. With the 15th title — a win in the College Football Playoff in 2017, coach Nick Saban tied the legendary Bear Bryant with five championships recognized by the NCAA.', 'American football is the most popular sport to watch in the United States, followed by baseball, basketball, and ice hockey, which makes up th...	[1, 0, 0, 0, 0, ...]

• Loss: LambdaLoss with these parameters:

  {
      "weighting_scheme": "sentence_transformers.cross_encoder.losses.LambdaLoss.NDCGLoss2PPScheme",
      "k": null,
      "sigma": 1.0,
      "eps": 1e-10,
      "reduction_log": "binary",
      "activation_fct": "torch.nn.modules.linear.Identity",
      "mini_batch_size": 16
  }
  

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	-	0.1318 (-0.4594)	0.0314 (-0.5091)	0.3145 (-0.0105)	0.0444 (-0.4562)	0.1301 (-0.3253)
0.0007	1	2.1483	-	-	-	-	-
0.0667	100	2.0302	-	-	-	-	-
0.1333	200	1.0684	-	-	-	-	-
0.1667	250	-	0.7116 (+0.1204)	0.4469 (-0.0935)	0.3483 (+0.0233)	0.6251 (+0.1244)	0.4734 (+0.0181)
0.2	300	0.6541	-	-	-	-	-
0.2667	400	0.5459	-	-	-	-	-
0.3333	500	0.5159	0.7425 (+0.1513)	0.5219 (-0.0186)	0.3722 (+0.0471)	0.6300 (+0.1294)	0.5080 (+0.0526)
0.4	600	0.4852	-	-	-	-	-
0.4667	700	0.4655	-	-	-	-	-
0.5	750	-	0.7545 (+0.1633)	0.5572 (+0.0167)	0.3726 (+0.0476)	0.6188 (+0.1182)	0.5162 (+0.0608)
0.5333	800	0.448	-	-	-	-	-
0.6	900	0.4283	-	-	-	-	-
0.6667	1000	0.4296	0.7582 (+0.1670)	0.5540 (+0.0136)	0.3723 (+0.0473)	0.6142 (+0.1136)	0.5135 (+0.0581)
0.7333	1100	0.4237	-	-	-	-	-
0.8	1200	0.4165	-	-	-	-	-
0.8333	1250	-	0.7600 (+0.1687)	0.5574 (+0.0169)	0.3676 (+0.0426)	0.5671 (+0.0665)	0.4974 (+0.0420)
0.8667	1300	0.4258	-	-	-	-	-
0.9333	1400	0.4192	-	-	-	-	-
1.0	1500	0.425	0.7601 (+0.1689)	0.5514 (+0.0110)	0.3714 (+0.0464)	0.5941 (+0.0934)	0.5056 (+0.0503)
-1	-1	-	0.7601 (+0.1689)	0.5514 (+0.0110)	0.3714 (+0.0464)	0.5941 (+0.0934)	0.5056 (+0.0503)

• 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: 1.2.0
• 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",
}

LambdaLoss

@inproceedings{wang2018lambdaloss,
  title={The lambdaloss framework for ranking metric optimization},
  author={Wang, Xuanhui and Li, Cheng and Golbandi, Nadav and Bendersky, Michael and Najork, Marc},
  booktitle={Proceedings of the 27th ACM international conference on information and knowledge management},
  pages={1313--1322},
  year={2018}
}