yjoonjang/reranker-msmarco-v1.1-MiniLM-L12-H384-uncased-plistmle-tanh

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 dataset 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: microsoft/MiniLM-L12-H384-uncased
• Maximum Sequence Length: 512 tokens
• Number of Output Labels: 1 label
• Training Dataset:
  • ms_marco
• Language: en

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("yjoonjang/reranker-msmarco-v1.1-MiniLM-L12-H384-uncased-plistmle-tanh")
# 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

• 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	0.4793 (-0.0103)	0.3262 (+0.0652)	0.5706 (+0.1510)
mrr@10	0.4667 (-0.0108)	0.5391 (+0.0393)	0.5727 (+0.1460)
ndcg@10	0.5367 (-0.0038)	0.3374 (+0.0124)	0.6188 (+0.1182)

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	0.4587 (+0.0686)
mrr@10	0.5262 (+0.0582)
ndcg@10	0.4976 (+0.0423)

Training Details

Training Dataset

ms_marco

• Dataset: ms_marco at a47ee7a
• Size: 78,704 training samples
• Columns: query, docs, and labels
• Approximate statistics based on the first 1000 samples:

  	query	docs	labels
  type	string	list	list
  details	min: 11 characters mean: 33.21 characters max: 83 characters	min: 2 elements mean: 6.00 elements max: 10 elements	min: 2 elements mean: 6.00 elements max: 10 elements

• Samples:

  query	docs	labels
  what is dubbin made of	["A recipe for Dubbin. One of the disadvantages of living out in the sticks is that when and idea or urge pops into your head and you need to purchase something to complete it, you need to be patient. I had a bit of a browse on tinternet to find out if you could make it. turns out it is made from tallow, beeswax and fish or mink oil. Found a website where they make their own and got the recipe, tweaked it a bit and Bob's your uncle, got it made, Go Mel !!", 'I have successfully made my own Dubbin: I rendered sheep fat for the tallow (beautiful white stuff-also good for cooking and preparing cast iron cookware), and then I added Beeswax and Neatsfoot oil and a few other extras. 1 DUBBIN Can be used over any non-sealing finish, such as spirit dyes and water based dyes as well as water based inks. 2 DUBBIN allows leather to become supple without loosing its shape-it helps the leather to stay alive and always as beautiful as new.', "Dubbin is made from tallow, natural wax and OIL. The wax...	[1, 0, 0, 0, 0, ...]
  is chromogranin a a tumor marker	['Chromogranin A (CGA) is a protein found in and released from neuroendocrine cells. The Chromogranin A test is used as a tumor marker. It may be ordered in combination with or in place of serotonin to help as an aid to diagnose carcinoid syndrome. Typical carcinoid symptoms include: 1 Flushing. 2 Diarrhea. 3 Abdominal pain. 4 Wheezing. 5 Valvular heart disease. ', 'Chromogranin A. Chromogranins are a family of glycoproteins with dense-core secretory vesicles that are found in neuronal and endocrine tissues. 3 Despite certain limitations, CgA is currently the most useful circulating marker for carcinoid tumors and PNET 5 and is elevated in 60% to 100% of NET. 4.', '* a_mamdouh8@hotmail.com Abstract: Evaluation of Serum Chromogranin A as a Useful Tumor Marker for Diagnosis of Hepatocellular Carcinoma. Journal of American Science 2011; 7(1):999-1007]. (ISSN: 1545-1003). http://www.americanscience.org. ', 'Chromogranin A (CgA) is an acidic glycoprotein expressed in the secretory gran...	[1, 0, 0, 0, 0, ...]
  what is an MRA	['Magnetic Resonance Angiogram (MRA). Guide. A magnetic resonance angiogram (MRA) is a type of magnetic resonance imaging (MRI) scan that uses a magnetic field and pulses of radio wave energy to provide pictures of blood vessels inside the body.', 'Why It Is Done. A magnetic resonance angiogram (MRA) is done to look for: 1 A bulge (aneurysm), clot, or the buildup of fat and calcium deposits (stenosis caused by plaque) in the blood vessels leading to the brain. 2 An aneurysm or tear (dissection) in the aorta, which carries blood from the heart to the rest of the body.', 'MRA: The magnetic resonance angiogram, or MRA, is a noninvasive test that has demonstrated usefulness in defining the anatomy of blood vessels of certain size in the head and neck. MRA serves as a complement to traditional MRI scanning in evaluation of the brain and neck.', '2. MRA. An MRA is a troglodyte that writes “women’s rights --- ROFLOL!“ and “a woman is the useless skin around the vagina” on Urban Dictionary. ...	[1, 0, 0, 0, 0, ...]

• Loss: ListMLELoss with these parameters:

  {
      "lambda_weight": "sentence_transformers.cross_encoder.losses.ListMLELoss.ListMLELambdaWeight",
      "activation_fct": "torch.nn.modules.activation.Tanh",
      "mini_batch_size": 16,
      "respect_input_order": true
  }
  

Evaluation Dataset

ms_marco

• Dataset: ms_marco at a47ee7a
• Size: 1,000 evaluation samples
• Columns: query, docs, and labels
• Approximate statistics based on the first 1000 samples:

  	query	docs	labels
  type	string	list	list
  details	min: 11 characters mean: 33.85 characters max: 109 characters	min: 3 elements mean: 6.50 elements max: 10 elements	min: 3 elements mean: 6.50 elements max: 10 elements

• Samples:

  query	docs	labels
  what is an xc mountain bike	['A cross-country mountain biker on a trail in Utah. Cross-country (XC) cycling is the most common discipline of mountain biking. Cross-country cycling became an Olympic sport in 1996 and is the only form of mountain biking practiced at the Olympics. Cross-country bicycles are some of the lightest mountain bikes, typically between 7 and 16 kilograms (15 and 35 lb). They usually feature suspension forks in front and sometimes have suspension in the rear', 'All Mountain, or Enduro bikes, are very similar to XC Trail bikes, but will have stronger frames, and a bit more travel in the suspension. Most of these bike will be in the full suspension category, and will have around 140-160 mm travel in them. The cold hard reality of modern mountain biking is that there as many bike types as there are riding disciplines. For people who are really into riding, and for whom riding is a way of life, it really isn’t uncommon for them to own 2 or 3 different types of mountain bike.', "Posts. 12. XC cou...	[1, 0, 0, 0, 0, ...]
  when did colorado become a state	["Colorado Became a State August 1, 1876. The 1846-1848 war between Mexico and the United States ended in the Treaty of Guadalupe Hidalgo (1848). The impetus for the organization of the Colorado territory was the discovery of gold. On 28 February 1861, the U.S. government organized the Territory of Colorado. (Colorado City and Golden served as the territory's capital, before Denver was declared the capital in 1867.). Colorado became a state on 1 August 1876. Due to the expansion of the railroads across the plains and into the mountains, and the subsequent increase in economic linkages, the state's population quickly grew", "The impetus for the organization of the Colorado territory was the discovery of gold. On 28 February 1861, the U.S. government organized the Territory of Colorado. (Colorado City and Golden served as the territory's capital, before Denver was declared the capital in 1867.) Colorado became a state on 1 August 1876.", 'It took sixteen years, four Colorado votes, three...	[1, 0, 0, 0, 0, ...]
  how much for roth ira value	['For details, see more on Roth IRA conversions). For 2015, you can contribute the maximum $5,500 to a Roth IRA ($6,500 if you are age 50 or older by the end of the year) if you are single or the single head of a household and your modified adjusted gross income (MAGI) is less than $164,000.', 'Calculate your after tax deposit amount and how much you will save in a Roth IRA. Add this retirement calculator to your site to keep your customers coming back. This Roth IRA calculator will integrate right into your existing website automatically.', 'The current balance of your Roth IRA. Annual contribution. The amount you will contribute to your Roth IRA each year. This calculator assumes that you make your contribution at the beginning of each year. The maximum annual IRA contribution of $5,500 is unchanged for 2015. It is important to note that this is the maximum total contributed to all of your IRA accounts.', 'Contribution Limits. As of 2013 you can contribute up to $5,500 each year to y...	[1, 0, 0, 0, 0, ...]

• Loss: ListMLELoss with these parameters:

  {
      "lambda_weight": "sentence_transformers.cross_encoder.losses.ListMLELoss.ListMLELambdaWeight",
      "activation_fct": "torch.nn.modules.activation.Tanh",
      "mini_batch_size": 16,
      "respect_input_order": true
  }
  

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
• 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: 0
• 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_R100_ndcg@10	NanoNFCorpus_R100_ndcg@10	NanoNQ_R100_ndcg@10	NanoBEIR_R100_mean_ndcg@10
-1	-1	-	-	0.0669 (-0.4735)	0.2492 (-0.0759)	0.0338 (-0.4669)	0.1166 (-0.3387)
0.0002	1	1172.8087	-	-	-	-	-
0.0508	250	927.3161	-	-	-	-	-
0.1016	500	922.4274	900.2140	0.1417 (-0.3987)	0.2186 (-0.1064)	0.3470 (-0.1536)	0.2358 (-0.2196)
0.1525	750	887.1161	-	-	-	-	-
0.2033	1000	883.4986	890.5686	0.3210 (-0.2194)	0.3210 (-0.0040)	0.5384 (+0.0377)	0.3935 (-0.0619)
0.2541	1250	893.9945	-	-	-	-	-
0.3049	1500	882.1399	886.1149	0.4382 (-0.1023)	0.3402 (+0.0151)	0.6326 (+0.1319)	0.4703 (+0.0149)
0.3558	1750	893.6919	-	-	-	-	-
0.4066	2000	860.9903	883.2151	0.4323 (-0.1081)	0.3507 (+0.0256)	0.5919 (+0.0913)	0.4583 (+0.0029)
0.4574	2250	864.9159	-	-	-	-	-
0.5082	2500	879.1681	880.0072	0.4801 (-0.0603)	0.3309 (+0.0058)	0.5657 (+0.0651)	0.4589 (+0.0035)
0.5591	2750	871.9338	-	-	-	-	-
0.6099	3000	876.5861	876.0945	0.5367 (-0.0038)	0.3374 (+0.0124)	0.6188 (+0.1182)	0.4976 (+0.0423)
0.6607	3250	869.3333	-	-	-	-	-
0.7115	3500	877.5201	878.9338	0.4719 (-0.0685)	0.3416 (+0.0166)	0.5921 (+0.0914)	0.4685 (+0.0132)
0.7624	3750	874.4561	-	-	-	-	-
0.8132	4000	869.9234	875.8646	0.5169 (-0.0236)	0.3306 (+0.0055)	0.5979 (+0.0973)	0.4818 (+0.0264)
0.8640	4250	876.4072	-	-	-	-	-
0.9148	4500	873.4344	875.6801	0.5085 (-0.0319)	0.3286 (+0.0036)	0.6254 (+0.1248)	0.4875 (+0.0322)
0.9656	4750	858.7703	-	-	-	-	-
-1	-1	-	-	0.5367 (-0.0038)	0.3374 (+0.0124)	0.6188 (+0.1182)	0.4976 (+0.0423)

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.11.11
• Sentence Transformers: 3.5.0.dev0
• Transformers: 4.49.0
• PyTorch: 2.6.0+cu124
• Accelerate: 1.5.2
• Datasets: 3.4.0
• 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",
}

ListMLELoss

@inproceedings{lan2013position,
    title={Position-aware ListMLE: a sequential learning process for ranking},
    author={Lan, Yanyan and Guo, Jiafeng and Cheng, Xueqi and Liu, Tie-Yan},
    booktitle={Proceedings of the Twenty-Ninth Conference on Uncertainty in Artificial Intelligence},
    pages={333--342},
    year={2013}
}