tomaarsen/reranker-ModernBERT-base-gooaq-bce-0margin-3min-100max-5top

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-0margin-3min-100max-5top")
# Get scores for pairs of texts
pairs = [
    ['what is baking powder bicarbonate soda?', 'Baking soda and bicarbonate of soda are actually different names for the same thing. ... Both bicarbonate of soda and baking powder are leavening (raising) agents. When included in a batter, the leavening agent creates air bubbles that expand when cooked, and cause it to rise.'],
    ['what is baking powder bicarbonate soda?', "What is baking soda? Baking soda is a leavening agent used in baked goods like cakes, muffins, and cookies. Formally known as sodium bicarbonate, it's a white crystalline powder that is naturally alkaline, or basic (1). Baking soda becomes activated when it's combined with both an acidic ingredient and a liquid."],
    ['what is baking powder bicarbonate soda?', 'The chemical name for baking powder is sodium hydrogencarbonate. You may see it called bicarbonate of soda in the supermarket. This is the old name for the same stuff. It has the chemical formula NaHCO3.'],
    ['what is baking powder bicarbonate soda?', "Substituting baking soda for baking powder What's more, baking soda has much stronger leavening power than baking powder. As a rule of thumb, about 1 teaspoon of baking powder is equivalent to 1/4 teaspoon of baking soda."],
    ['what is baking powder bicarbonate soda?', "Baking soda is a leavening agent used in baked goods like cakes, muffins, and cookies. Formally known as sodium bicarbonate, it's a white crystalline powder that is naturally alkaline, or basic (1). Baking soda becomes activated when it's combined with both an acidic ingredient and a liquid."],
]
scores = model.predict(pairs)
print(scores.shape)
# (5,)

# Or rank different texts based on similarity to a single text
ranks = model.rank(
    'what is baking powder bicarbonate soda?',
    [
        'Baking soda and bicarbonate of soda are actually different names for the same thing. ... Both bicarbonate of soda and baking powder are leavening (raising) agents. When included in a batter, the leavening agent creates air bubbles that expand when cooked, and cause it to rise.',
        "What is baking soda? Baking soda is a leavening agent used in baked goods like cakes, muffins, and cookies. Formally known as sodium bicarbonate, it's a white crystalline powder that is naturally alkaline, or basic (1). Baking soda becomes activated when it's combined with both an acidic ingredient and a liquid.",
        'The chemical name for baking powder is sodium hydrogencarbonate. You may see it called bicarbonate of soda in the supermarket. This is the old name for the same stuff. It has the chemical formula NaHCO3.',
        "Substituting baking soda for baking powder What's more, baking soda has much stronger leavening power than baking powder. As a rule of thumb, about 1 teaspoon of baking powder is equivalent to 1/4 teaspoon of baking soda.",
        "Baking soda is a leavening agent used in baked goods like cakes, muffins, and cookies. Formally known as sodium bicarbonate, it's a white crystalline powder that is naturally alkaline, or basic (1). Baking soda becomes activated when it's combined with both an acidic ingredient and a liquid.",
    ]
)
# [{'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	0.7234 (+0.1923)
mrr@10	0.7223 (+0.1984)
ndcg@10	0.7676 (+0.1764)

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	0.4711 (-0.0185)	0.3601 (+0.0992)	0.6047 (+0.1851)
mrr@10	0.4565 (-0.0210)	0.5969 (+0.0971)	0.6064 (+0.1797)
ndcg@10	0.5342 (-0.0062)	0.4250 (+0.0999)	0.6652 (+0.1646)

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	0.4786 (+0.0886)
mrr@10	0.5533 (+0.0853)
ndcg@10	0.5415 (+0.0861)

Training Details

Training Dataset

Unnamed Dataset

• Size: 577,957 training samples
• Columns: question, answer, and label
• Approximate statistics based on the first 1000 samples:

  	question	answer	label
  type	string	string	int
  details	min: 21 characters mean: 42.64 characters max: 76 characters	min: 54 characters mean: 250.97 characters max: 376 characters	0: ~83.00% 1: ~17.00%

• Samples:

  question	answer	label
  what is baking powder bicarbonate soda?	Baking soda and bicarbonate of soda are actually different names for the same thing. ... Both bicarbonate of soda and baking powder are leavening (raising) agents. When included in a batter, the leavening agent creates air bubbles that expand when cooked, and cause it to rise.	1
  what is baking powder bicarbonate soda?	What is baking soda? Baking soda is a leavening agent used in baked goods like cakes, muffins, and cookies. Formally known as sodium bicarbonate, it's a white crystalline powder that is naturally alkaline, or basic (1). Baking soda becomes activated when it's combined with both an acidic ingredient and a liquid.	0
  what is baking powder bicarbonate soda?	The chemical name for baking powder is sodium hydrogencarbonate. You may see it called bicarbonate of soda in the supermarket. This is the old name for the same stuff. It has the chemical formula NaHCO3.	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	-	0.1293 (-0.4619)	0.0284 (-0.5121)	0.2145 (-0.1105)	0.0134 (-0.4872)	0.0854 (-0.3699)
0.0001	1	1.2576	-	-	-	-	-
0.0221	200	1.2027	-	-	-	-	-
0.0443	400	1.1352	-	-	-	-	-
0.0664	600	0.7686	-	-	-	-	-
0.0886	800	0.6163	-	-	-	-	-
0.1107	1000	0.5764	0.7162 (+0.1250)	0.4924 (-0.0480)	0.3647 (+0.0396)	0.6409 (+0.1403)	0.4993 (+0.0440)
0.1329	1200	0.5488	-	-	-	-	-
0.1550	1400	0.525	-	-	-	-	-
0.1772	1600	0.4987	-	-	-	-	-
0.1993	1800	0.4943	-	-	-	-	-
0.2215	2000	0.4777	0.7508 (+0.1596)	0.5672 (+0.0268)	0.3969 (+0.0718)	0.6236 (+0.1230)	0.5292 (+0.0739)
0.2436	2200	0.4487	-	-	-	-	-
0.2658	2400	0.4582	-	-	-	-	-
0.2879	2600	0.4473	-	-	-	-	-
0.3100	2800	0.4266	-	-	-	-	-
0.3322	3000	0.4374	0.7478 (+0.1565)	0.5851 (+0.0446)	0.3863 (+0.0613)	0.6684 (+0.1678)	0.5466 (+0.0912)
0.3543	3200	0.421	-	-	-	-	-
0.3765	3400	0.4317	-	-	-	-	-
0.3986	3600	0.4206	-	-	-	-	-
0.4208	3800	0.417	-	-	-	-	-
0.4429	4000	0.4113	0.7577 (+0.1665)	0.5611 (+0.0207)	0.3973 (+0.0722)	0.6564 (+0.1557)	0.5382 (+0.0829)
0.4651	4200	0.4008	-	-	-	-	-
0.4872	4400	0.3884	-	-	-	-	-
0.5094	4600	0.4136	-	-	-	-	-
0.5315	4800	0.389	-	-	-	-	-
0.5536	5000	0.3877	0.7609 (+0.1697)	0.5509 (+0.0104)	0.3878 (+0.0627)	0.6807 (+0.1800)	0.5398 (+0.0844)
0.5758	5200	0.3901	-	-	-	-	-
0.5979	5400	0.389	-	-	-	-	-
0.6201	5600	0.3999	-	-	-	-	-
0.6422	5800	0.3703	-	-	-	-	-
0.6644	6000	0.3854	0.7620 (+0.1708)	0.5444 (+0.0039)	0.4040 (+0.0790)	0.6917 (+0.1911)	0.5467 (+0.0913)
0.6865	6200	0.3685	-	-	-	-	-
0.7087	6400	0.3751	-	-	-	-	-
0.7308	6600	0.3709	-	-	-	-	-
0.7530	6800	0.3788	-	-	-	-	-
0.7751	7000	0.3734	0.7672 (+0.1760)	0.5404 (+0.0000)	0.4075 (+0.0824)	0.6638 (+0.1632)	0.5372 (+0.0819)
0.7973	7200	0.3629	-	-	-	-	-
0.8194	7400	0.3547	-	-	-	-	-
0.8415	7600	0.3639	-	-	-	-	-
0.8637	7800	0.3597	-	-	-	-	-
0.8858	8000	0.3522	0.7676 (+0.1764)	0.5342 (-0.0062)	0.4250 (+0.0999)	0.6652 (+0.1646)	0.5415 (+0.0861)
0.9080	8200	0.327	-	-	-	-	-
0.9301	8400	0.344	-	-	-	-	-
0.9523	8600	0.3578	-	-	-	-	-
0.9744	8800	0.3547	-	-	-	-	-
0.9966	9000	0.3491	0.7675 (+0.1763)	0.5423 (+0.0019)	0.4188 (+0.0937)	0.6621 (+0.1614)	0.5411 (+0.0857)
-1	-1	-	0.7676 (+0.1764)	0.5342 (-0.0062)	0.4250 (+0.0999)	0.6652 (+0.1646)	0.5415 (+0.0861)

• 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",
}