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

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("tomaarsen/reranker-msmarco-v1.1-MiniLM-L12-H384-uncased-plistmle")
# 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	0.4975 (+0.0079)	0.3252 (+0.0642)	0.5857 (+0.1661)
mrr@10	0.4843 (+0.0068)	0.5679 (+0.0681)	0.5922 (+0.1655)
ndcg@10	0.5506 (+0.0102)	0.3756 (+0.0505)	0.6570 (+0.1563)

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	0.4695 (+0.0794)
mrr@10	0.5481 (+0.0801)
ndcg@10	0.5277 (+0.0724)

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: 9 characters mean: 33.73 characters max: 119 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 to avoid during early pregnancy	['Although caffeine does not come under the category of foods to avoid during early pregnancy, pregnant women are advised to limit their caffeine consumption. Caffeine can be found in tea, coffee, soft drinks, chocolate etc.', 'Learn what foods to eat and what to avoid during pregnancy to ensure a healthy environment for your unborn baby! As a concerned parent, you want to do everything possible to ensure the well being and safety of your baby.', 'To stay safe, also avoid these foods during your pregnancy. Meats. 1 Cold cuts, deli meats, hot dogs, and other ready-to-eat meats. ( 2 You can safely eat these if they are heated to steaming and served hot.). 3 Pre-stuffed, fresh, turkey or chicken. 4 Steak tartare or any raw meat. 5 Rare cuts of meat and undercooked meats.', 'Raw and undercooked meat is another among foods to avoid during first trimester. Make sure that the meat is well cooked and consume while it is still hot. It would be good to avoid processed meat since pregnant wom...	[1, 0, 0, 0, 0, ...]
  where is bells creek	['Simpsonville, SC Real Estate. #facebook# Bells Creek is a small neighborhood in Simpsonville, SC, close to Bells Crossing Elementary. Located near Woodruff Rd., I-85 and I-385, the upscale Bells Creek homes are typically on large lots with mature trees. Bells Creek amenities include a pool and cabana. Bells Creek real estate prices average $220,000.', '#facebook# Bells Creek is a small neighborhood in Simpsonville, SC, close to Bells Crossing Elementary. Located near Woodruff Rd., I-85 and I-385, the upscale Bells Creek homes are typically on large lots with mature trees. Bells Creek amenities include a pool and cabana. Bells Creek real estate prices average $220,000.', "Welcome to The Overlook at Bells Creek, an exclusive Eastwood Homes' Greenville area community only minutes away from Five Forks in Simpsonville, SC.", 'Property Details. Property details for 213 Bells Creek Dr, Simpsonville, SC 29681. This Single Family Home is located at Bells Creek in Simpsonville, South Carolina. The home provides approximately 2074 square feet of living space. This property features 4 bedrooms. There are 3 bathrooms. 213 Bells Creek Dr, Simpsonville, SC 29681 falls within the Greenville county lines. This home sold for $180,000 on Dec 17, 2014. Similar homes in the area are priced around $187,091.']	[1, 1, 0, 0]
  how long does it take to hatch geese eggs in an incubator	['Geese take 31 days of incubation for a goose egg to hatch. Whether underneath its parents, or in an incubator, the incubation time is the same.', 'While chicken eggs take 21 days, for example, geese can take between 30 and 35 days and need a higher humidity level. Goslings are also more likely to hatch if the eggs are sprayed with water every day between days 6 and about 25, whereas chicken eggs need to be kept in humid conditions, but dry.', 'Incubation Duration. Incubating goose eggs should be done for a period of about 28 days for smaller breeds, and up to 35 days for larger breeds before pipping begins. Once goose eggs begin hatching, the process can take up to three days before they are completely out of their shell.', "It takes 21 days to incubate the egg. the 21st day is the hatching day. if the eggs are mail order don't count the day the eggs arrive if the temperature is below 55 degrees f … . on the 21st day the eggs will hatch at all different times.", '5. Wait until the eg...	[1, 0, 0, 0, 0, ...]

• Loss: ListMLELoss with these parameters:

  {
      "lambda_weight": "sentence_transformers.cross_encoder.losses.ListMLELoss.ListMLELambdaWeight",
      "activation_fct": "torch.nn.modules.linear.Identity",
      "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: 12 characters mean: 33.41 characters max: 94 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
  who does glenn beck want for speaker of the house	["	. Nebraska Senator Ben Sasse spoke to Glenn Beck Wednesday about Sasse's idea to make Arthur Brooks, head of the American Enterprise Institute, the next Speaker of the House. There’s nothing in the Constitution that requires you to have a Speaker who is an elected member of Congress, Sasse said.", 'And the guy that kept coming to my mind as I was watching Sunday Night Football is Arthur Brooks, the head of AEI. And so I think that the House Republicans should think about going outside the box. There’s nothing in the Constitution that requires you to have a Speaker who is an elected member of Congress.', 'Share on Facebook Share on Twitter. Conservative radio host Glenn Beck went off on Republican leadership after the House passed a budget compromise Thursday, calling House Speaker John Boehner “worthless” and Senate Minority Leader Mitch McConnell a “liar.”.', '“I think John Boehner is one of the prime examples of worthless, worthless Republicans,” Beck said Thursday on Mark Levin’s...
  how long do you have to keep former employee records	['Employee Contracts. If you have employment contracts with your employees, then you should maintain these contracts for at least 10 years. According to Financial Web, you should “err on the side of caution” and maintain employee records for a longer period than you think you may need, in case a legal issue arises.', 'Small business expert Rieva Lesonsky suggests you keep employment records for a minimum of two years and for up to seven years. She says that most states have a two-year statute of limitations on lawsuit filings by former employees, so you want to make sure you have documents at hand if this occurs.', 'Since employees may come and go, you may wonder how long you should hang on to the employee records. The Internal Revenue Service (IRS) weighs in on records pertaining to employee taxes, such as payroll, but the other records depend on what types of records you have for employees.', 'Effective January 1, 2013, California law provides that current and former employees (or a ...	[1, 0, 0, 0, 0, ...]
  what year was velcro invented	['Velcro was invented by George de Mestral a Swiss electrical engineer in 1941. This idea of inventing Velcro came to him when one day he returned after a walk from the hills and found cockleburs stuck to his clothes and his dog’s fur. George noticed its natural hook and loop quality and started making a fabric fastener on the same quality.', 'Velcro, which was invented by a Swiss Electrical Engineer George de Mestral, comprises two layers and when both these sides are hard-pressed together, they assist in fixing two surfaces. The thought to invent Velcro hits Mestral’s mind in the year 1941 after coming back from a hunting tour with his dog. ', 'In 1958, de Mestral filed for a patent application for his hook-and-loop fastener in Switzerland, which was granted in 1961. The term Velcro is a registered trademark of Velcro Industries B.V. Velcro Industries is a privately held worldwide corporation manufacturing consumer and industrial products. Among them is a series of mechanical-based f...	[1, 0, 0, 0, 0, ...]

• Loss: ListMLELoss with these parameters:

  {
      "lambda_weight": "sentence_transformers.cross_encoder.losses.ListMLELoss.ListMLELambdaWeight",
      "activation_fct": "torch.nn.modules.linear.Identity",
      "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	-	-	0.0344 (-0.5060)	0.2073 (-0.1178)	0.0336 (-0.4671)	0.0918 (-0.3636)
0.0002	1	1412.3083	-	-	-	-	-
0.0508	250	887.7485	-	-	-	-	-
0.1016	500	853.8898	903.5635	0.2242 (-0.3163)	0.2467 (-0.0783)	0.3585 (-0.1421)	0.2765 (-0.1789)
0.1525	750	867.3723	-	-	-	-	-
0.2033	1000	851.3223	880.1996	0.4790 (-0.0614)	0.3435 (+0.0184)	0.5945 (+0.0938)	0.4723 (+0.0170)
0.2541	1250	840.5654	-	-	-	-	-
0.3049	1500	836.1076	872.8075	0.5189 (-0.0216)	0.3394 (+0.0143)	0.6097 (+0.1091)	0.4893 (+0.0339)
0.3558	1750	853.3524	-	-	-	-	-
0.4066	2000	859.1896	872.7851	0.5453 (+0.0049)	0.3638 (+0.0387)	0.6322 (+0.1315)	0.5137 (+0.0584)
0.4574	2250	816.2849	-	-	-	-	-
0.5082	2500	832.0728	866.5376	0.5428 (+0.0023)	0.3737 (+0.0487)	0.6384 (+0.1378)	0.5183 (+0.0629)
0.5591	2750	825.9285	-	-	-	-	-
0.6099	3000	809.4326	865.0468	0.5319 (-0.0085)	0.3488 (+0.0238)	0.6320 (+0.1313)	0.5042 (+0.0489)
0.6607	3250	807.3669	-	-	-	-	-
0.7115	3500	828.0153	869.0601	0.5479 (+0.0075)	0.3690 (+0.0440)	0.6495 (+0.1488)	0.5221 (+0.0668)
0.7624	3750	841.2574	-	-	-	-	-
0.8132	4000	814.0583	865.1564	0.5406 (+0.0001)	0.3571 (+0.0320)	0.6519 (+0.1513)	0.5165 (+0.0612)
0.8640	4250	814.6952	-	-	-	-	-
0.9148	4500	825.9762	864.4775	0.5506 (+0.0102)	0.3756 (+0.0505)	0.6570 (+0.1563)	0.5277 (+0.0724)
0.9656	4750	821.2723	-	-	-	-	-
-1	-1	-	-	0.5506 (+0.0102)	0.3756 (+0.0505)	0.6570 (+0.1563)	0.5277 (+0.0724)

• The bold row denotes the saved checkpoint.

Environmental Impact

Carbon emissions were measured using CodeCarbon.

• Energy Consumed: 0.244 kWh
• Carbon Emitted: 0.095 kg of CO2
• Hours Used: 0.917 hours

Training Hardware

• On Cloud: No
• GPU Model: 1 x NVIDIA GeForce RTX 3090
• CPU Model: 13th Gen Intel(R) Core(TM) i7-13700K
• RAM Size: 31.78 GB

Framework Versions

• Python: 3.11.6
• Sentence Transformers: 3.5.0.dev0
• Transformers: 4.49.0
• PyTorch: 2.6.0+cu124
• Accelerate: 1.5.1
• Datasets: 3.3.2
• 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",
}

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}
}