tomaarsen/reranker-MiniLM-L12-H384-msmarco-mse

CrossEncoder based on microsoft/MiniLM-L12-H384-uncased

This is a Cross Encoder model finetuned from microsoft/MiniLM-L12-H384-uncased on the ms-marco-shuffled dataset using the sentence-transformers library. It computes scores for pairs of texts, which can be used for semantic textual similarity, semantic search, paraphrase mining, text classification, clustering, and more.

Model Details

Model Description

• Model Type: Cross Encoder
• Base model: microsoft/MiniLM-L12-H384-uncased
• Maximum Sequence Length: 512 tokens
• Number of Output Labels: 1 label
• Training Dataset:
  • ms-marco-shuffled

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-msmarco-mse")
# Get scores for pairs of texts
pairs = [
    ['what is a electrophoresis apparatus', 'Gel electrophoresis is a method for separation and analysis of macromolecules (DNA, RNA and proteins) and their fragments, based on their size and charge.el electrophoresis of large DNA or RNA is usually done by agarose gel electrophoresis. See the  Chain termination method  page for an example of a polyacrylamide DNA sequencing gel. Characterization through ligand interaction of nucleic acids or fragments may be performed by mobility shift affinity electrophoresis.'],
    ['does creatine elevate creatinine levels', "Creatinine is produced from creatine, a molecule of major importance for energy production in muscles. Approximately 2% of the body's creatine is converted to creatinine every day. Creatinine is transported through the bloodstream to the kidneys."],
    ['how to get rid of caffeine in the body', 'In addition to quickly curing caffeine withdrawal headaches, caffeine may help cure regular headaches and even migraines. Some studies have shown that small doses of caffeine taken in conjunction with pain killers may help the body absorb the medication more quickly and cure the headache in a shorter period of time.'],
    ['define splanchnopleure', 'delineated, represented, delineate(verb) represented accurately or precisely. define, delineate(verb) show the form or outline of. The tree was clearly defined by the light; The camera could define the smallest object. specify, define, delineate, delimit, delimitate(verb) determine the essential quality of.'],
    ['how many calories does a glass of wine', 'A large glass of wine contains as many calories as an ice cream. We often drink wine with a meal. But did you know that a large glass of wine (250ml) with 13% ABV can add 228 calories to your dinner? Thatâ\x80\x99s similar to an ice cream or two fish fingers. A standard glass of red or white wine (175ml) with 13% ABV could also contain up to 160 calories, similar to a slice of Madeira cake.'],
]
scores = model.predict(pairs)
print(scores.shape)
# (5,)

# Or rank different texts based on similarity to a single text
ranks = model.rank(
    'what is a electrophoresis apparatus',
    [
        'Gel electrophoresis is a method for separation and analysis of macromolecules (DNA, RNA and proteins) and their fragments, based on their size and charge.el electrophoresis of large DNA or RNA is usually done by agarose gel electrophoresis. See the  Chain termination method  page for an example of a polyacrylamide DNA sequencing gel. Characterization through ligand interaction of nucleic acids or fragments may be performed by mobility shift affinity electrophoresis.',
        "Creatinine is produced from creatine, a molecule of major importance for energy production in muscles. Approximately 2% of the body's creatine is converted to creatinine every day. Creatinine is transported through the bloodstream to the kidneys.",
        'In addition to quickly curing caffeine withdrawal headaches, caffeine may help cure regular headaches and even migraines. Some studies have shown that small doses of caffeine taken in conjunction with pain killers may help the body absorb the medication more quickly and cure the headache in a shorter period of time.',
        'delineated, represented, delineate(verb) represented accurately or precisely. define, delineate(verb) show the form or outline of. The tree was clearly defined by the light; The camera could define the smallest object. specify, define, delineate, delimit, delimitate(verb) determine the essential quality of.',
        'A large glass of wine contains as many calories as an ice cream. We often drink wine with a meal. But did you know that a large glass of wine (250ml) with 13% ABV can add 228 calories to your dinner? Thatâ\x80\x99s similar to an ice cream or two fish fingers. A standard glass of red or white wine (175ml) with 13% ABV could also contain up to 160 calories, similar to a slice of Madeira cake.',
    ]
)
# [{'corpus_id': ..., 'score': ...}, {'corpus_id': ..., 'score': ...}, ...]

Evaluation

Metrics

Cross Encoder Reranking

• Datasets: NanoMSMARCO, NanoNFCorpus and NanoNQ
• Evaluated with CERerankingEvaluator

Metric	NanoMSMARCO	NanoNFCorpus	NanoNQ
map	0.5979 (+0.1083)	0.3464 (+0.0760)	0.6886 (+0.2679)
mrr@10	0.5893 (+0.1118)	0.6264 (+0.1266)	0.6962 (+0.2695)
ndcg@10	0.6585 (+0.1181)	0.3864 (+0.0613)	0.7366 (+0.2359)

Cross Encoder Nano BEIR

• Dataset: NanoBEIR_mean
• Evaluated with CENanoBEIREvaluator

Metric	Value
map	0.5443 (+0.1507)
mrr@10	0.6373 (+0.1693)
ndcg@10	0.5938 (+0.1385)

Training Details

Training Dataset

ms-marco-shuffled

• Dataset: ms-marco-shuffled at 0e80192
• Size: 1,990,000 training samples
• Columns: score, query, and passage
• Approximate statistics based on the first 1000 samples:

  	score	query	passage
  type	float	string	string
  details	min: -11.8 mean: 0.75 max: 11.16	min: 9 characters mean: 33.33 characters max: 123 characters	min: 53 characters mean: 348.8 characters max: 1016 characters

• Samples:

  score	query	passage
  6.732539335886638	what is shielding in welding	A benefit in using a shielding gas when welding is that there is no slag left on the weld that requires chipping and cleaning like that which is found on an arc weld. When a new wire welding machine is purchased, it does not come with a shielding gas tank. This must be purchased or rented from a gas supplier. Most welding supply stores also sell welding gasses and will be able to assist the buyer in a tank purchase.
  -5.769245758652687	what degree do you need for physical therapy	E. Medicaid covers occupational therapy, physical therapy and speech therapy services when provided to eligible Medicaid beneficiaries under age 21 in the Child Health Services (EPSDT) Program by qualified occupational, physical or speech therapy providers.
  9.033631960550943	cascade effect definition	In medicine, cascade effect may also refer to a chain of events initiated by an unnecessary test, an unexpected result, or patient or physician anxiety, which results in ill-advised tests or treatments that may cause harm to patients as the results are pursued.

• Loss: MSELoss

Evaluation Dataset

ms-marco-shuffled

• Dataset: ms-marco-shuffled at 0e80192
• Size: 10,000 evaluation samples
• Columns: score, query, and passage
• Approximate statistics based on the first 1000 samples:

  	score	query	passage
  type	float	string	string
  details	min: -11.86 mean: 0.72 max: 11.07	min: 10 characters mean: 33.83 characters max: 101 characters	min: 50 characters mean: 343.73 characters max: 929 characters

• Samples:

  score	query	passage
  4.691008905569713	what is a electrophoresis apparatus	Gel electrophoresis is a method for separation and analysis of macromolecules (DNA, RNA and proteins) and their fragments, based on their size and charge.el electrophoresis of large DNA or RNA is usually done by agarose gel electrophoresis. See the Chain termination method page for an example of a polyacrylamide DNA sequencing gel. Characterization through ligand interaction of nucleic acids or fragments may be performed by mobility shift affinity electrophoresis.
  0.7860534191131592	does creatine elevate creatinine levels	Creatinine is produced from creatine, a molecule of major importance for energy production in muscles. Approximately 2% of the body's creatine is converted to creatinine every day. Creatinine is transported through the bloodstream to the kidneys.
  -1.2669222354888916	how to get rid of caffeine in the body	In addition to quickly curing caffeine withdrawal headaches, caffeine may help cure regular headaches and even migraines. Some studies have shown that small doses of caffeine taken in conjunction with pain killers may help the body absorb the medication more quickly and cure the headache in a shorter period of time.

• Loss: MSELoss

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• per_device_train_batch_size: 64
• per_device_eval_batch_size: 64
• learning_rate: 8e-06
• 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: 8e-06
• 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	Validation Loss	NanoMSMARCO_ndcg@10	NanoNFCorpus_ndcg@10	NanoNQ_ndcg@10	NanoBEIR_mean_ndcg@10
-1	-1	-	-	0.0219 (-0.5185)	0.2538 (-0.0712)	0.0498 (-0.4509)	0.1085 (-0.3469)
0.0000	1	64.054	-	-	-	-	-
0.0322	1000	55.8586	-	-	-	-	-
0.0643	2000	31.6183	-	-	-	-	-
0.0965	3000	13.1762	-	-	-	-	-
0.1286	4000	6.1773	-	-	-	-	-
0.1608	5000	4.2945	3.4889	0.6180 (+0.0776)	0.3893 (+0.0643)	0.7144 (+0.2137)	0.5739 (+0.1185)
0.1930	6000	3.6451	-	-	-	-	-
0.2251	7000	3.3041	-	-	-	-	-
0.2573	8000	2.9813	-	-	-	-	-
0.2894	9000	2.8473	-	-	-	-	-
0.3216	10000	2.6852	2.6960	0.6124 (+0.0720)	0.3992 (+0.0742)	0.7315 (+0.2309)	0.5811 (+0.1257)
0.3538	11000	2.6128	-	-	-	-	-
0.3859	12000	2.5252	-	-	-	-	-
0.4181	13000	2.461	-	-	-	-	-
0.4502	14000	2.3625	-	-	-	-	-
0.4824	15000	2.2746	2.0279	0.6397 (+0.0993)	0.3963 (+0.0713)	0.7369 (+0.2363)	0.5910 (+0.1356)
0.5146	16000	2.2551	-	-	-	-	-
0.5467	17000	2.2193	-	-	-	-	-
0.5789	18000	2.2099	-	-	-	-	-
0.6111	19000	2.1277	-	-	-	-	-
0.6432	20000	2.0969	1.9564	0.6468 (+0.1063)	0.3936 (+0.0685)	0.7391 (+0.2385)	0.5932 (+0.1378)
0.6754	21000	2.0624	-	-	-	-	-
0.7075	22000	2.0565	-	-	-	-	-
0.7397	23000	2.0226	-	-	-	-	-
0.7719	24000	1.9583	-	-	-	-	-
0.8040	25000	2.0048	1.8239	0.6575 (+0.1171)	0.3884 (+0.0634)	0.7339 (+0.2333)	0.5933 (+0.1379)
0.8362	26000	1.9861	-	-	-	-	-
0.8683	27000	1.9675	-	-	-	-	-
0.9005	28000	1.9531	-	-	-	-	-
0.9327	29000	1.9139	-	-	-	-	-
0.9648	30000	1.9224	1.7848	0.6585 (+0.1181)	0.3864 (+0.0613)	0.7366 (+0.2359)	0.5938 (+0.1385)
0.9970	31000	1.9059	-	-	-	-	-
-1	-1	-	-	0.6585 (+0.1181)	0.3864 (+0.0613)	0.7366 (+0.2359)	0.5938 (+0.1385)

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.11.10
• Sentence Transformers: 3.5.0.dev0
• Transformers: 4.49.0.dev0
• PyTorch: 2.6.0.dev20241112+cu121
• Accelerate: 1.2.0
• Datasets: 3.2.0
• 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",
}