tomaarsen/reranker-MiniLM-L12-gooaq-bce

MiniLM-L12-H384-uncased trained on GooAQ

This is a Cross Encoder model finetuned from microsoft/MiniLM-L12-H384-uncased 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
• 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-MiniLM-L12-gooaq-bce")
# Get scores for pairs of texts
pairs = [
    ['what is the remote desktop connection broker?', 'A remote desktop connection broker is software that allows clients to access various types of server-hosted desktops and applications. ... Load balancing the servers that host the desktops. Managing desktop images. Redirecting multimedia processing to the client.'],
    ['what is the remote desktop connection broker?', 'Remote Desktop Connection (RDC, also called Remote Desktop, formerly Microsoft Terminal Services Client, mstsc or tsclient) is the client application for RDS. It allows a user to remotely log into a networked computer running the terminal services server.'],
    ['what is the remote desktop connection broker?', "['Click the Start menu on your PC and search for Remote Desktop Connection.', 'Launch Remote Desktop Connection and click on Show Options.', 'Select the Local Resources tab and click More.', 'Under Drives, check the box for your C: drive or the drives that contain the files you will transfer and click OK.']"],
    ['what is the remote desktop connection broker?', "['Press the MENU button on your remote.', 'Select Parental Favs & Setup > System Setup > Remote or Remote Control > Program Remote.', 'Choose the device (TV, audio, DVD) you want to program. ... ', 'Follow the on-screen instructions to finish programming your remote.']"],
    ['what is the remote desktop connection broker?', "['Press MENU on your remote.', 'Select Parental Favs & Setup > System Setup > Remote or Remote Control > Program Remote.', 'Choose the device (TV, audio, DVD) you wish to program. ... ', 'Follow the on-screen prompts to complete the programming.']"],
]
scores = model.predict(pairs)
print(scores.shape)
# (5,)

# Or rank different texts based on similarity to a single text
ranks = model.rank(
    'what is the remote desktop connection broker?',
    [
        'A remote desktop connection broker is software that allows clients to access various types of server-hosted desktops and applications. ... Load balancing the servers that host the desktops. Managing desktop images. Redirecting multimedia processing to the client.',
        'Remote Desktop Connection (RDC, also called Remote Desktop, formerly Microsoft Terminal Services Client, mstsc or tsclient) is the client application for RDS. It allows a user to remotely log into a networked computer running the terminal services server.',
        "['Click the Start menu on your PC and search for Remote Desktop Connection.', 'Launch Remote Desktop Connection and click on Show Options.', 'Select the Local Resources tab and click More.', 'Under Drives, check the box for your C: drive or the drives that contain the files you will transfer and click OK.']",
        "['Press the MENU button on your remote.', 'Select Parental Favs & Setup > System Setup > Remote or Remote Control > Program Remote.', 'Choose the device (TV, audio, DVD) you want to program. ... ', 'Follow the on-screen instructions to finish programming your remote.']",
        "['Press MENU on your remote.', 'Select Parental Favs & Setup > System Setup > Remote or Remote Control > Program Remote.', 'Choose the device (TV, audio, DVD) you wish to program. ... ', 'Follow the on-screen prompts to complete the programming.']",
    ]
)
# [{'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	0.6856 (+0.1545)
mrr@10	0.6830 (+0.1591)
ndcg@10	0.7314 (+0.1402)

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	0.4320 (-0.0576)	0.3503 (+0.0894)	0.5234 (+0.1038)
mrr@10	0.4205 (-0.0570)	0.5706 (+0.0708)	0.5284 (+0.1018)
ndcg@10	0.5022 (-0.0382)	0.3846 (+0.0596)	0.5854 (+0.0847)

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	0.4353 (+0.0452)
mrr@10	0.5065 (+0.0385)
ndcg@10	0.4907 (+0.0354)

Training Details

Training Dataset

Unnamed Dataset

• Size: 578,402 training samples
• Columns: question, answer, and label
• Approximate statistics based on the first 1000 samples:

  	question	answer	label
  type	string	string	int
  details	min: 18 characters mean: 42.66 characters max: 73 characters	min: 51 characters mean: 252.61 characters max: 368 characters	0: ~82.90% 1: ~17.10%

• Samples:

  question	answer	label
  what is the remote desktop connection broker?	A remote desktop connection broker is software that allows clients to access various types of server-hosted desktops and applications. ... Load balancing the servers that host the desktops. Managing desktop images. Redirecting multimedia processing to the client.	1
  what is the remote desktop connection broker?	Remote Desktop Connection (RDC, also called Remote Desktop, formerly Microsoft Terminal Services Client, mstsc or tsclient) is the client application for RDS. It allows a user to remotely log into a networked computer running the terminal services server.	0
  what is the remote desktop connection broker?	['Click the Start menu on your PC and search for Remote Desktop Connection.', 'Launch Remote Desktop Connection and click on Show Options.', 'Select the Local Resources tab and click More.', 'Under Drives, check the box for your C: drive or the drives that contain the files you will transfer and click OK.']	0

• Loss: BinaryCrossEntropyLoss with these parameters:

  {
      "activation_fn": "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	-	0.1548 (-0.4365)	0.0475 (-0.4929)	0.2762 (-0.0489)	0.0485 (-0.4521)	0.1241 (-0.3313)
0.0001	1	1.0439	-	-	-	-	-
0.0221	200	1.1645	-	-	-	-	-
0.0443	400	1.0837	-	-	-	-	-
0.0664	600	0.8732	-	-	-	-	-
0.0885	800	0.7901	-	-	-	-	-
0.1106	1000	0.755	0.6710 (+0.0798)	0.5150 (-0.0254)	0.3164 (-0.0086)	0.6085 (+0.1079)	0.4800 (+0.0246)
0.1328	1200	0.7095	-	-	-	-	-
0.1549	1400	0.7094	-	-	-	-	-
0.1770	1600	0.6715	-	-	-	-	-
0.1992	1800	0.6583	-	-	-	-	-
0.2213	2000	0.6865	0.6994 (+0.1082)	0.5033 (-0.0372)	0.3608 (+0.0357)	0.6058 (+0.1052)	0.4900 (+0.0346)
0.2434	2200	0.6392	-	-	-	-	-
0.2655	2400	0.6403	-	-	-	-	-
0.2877	2600	0.6538	-	-	-	-	-
0.3098	2800	0.6273	-	-	-	-	-
0.3319	3000	0.6091	0.7033 (+0.1121)	0.4779 (-0.0625)	0.3369 (+0.0119)	0.5859 (+0.0852)	0.4669 (+0.0115)
0.3541	3200	0.6244	-	-	-	-	-
0.3762	3400	0.6246	-	-	-	-	-
0.3983	3600	0.6222	-	-	-	-	-
0.4204	3800	0.5986	-	-	-	-	-
0.4426	4000	0.622	0.7252 (+0.1339)	0.5538 (+0.0133)	0.3718 (+0.0468)	0.5965 (+0.0959)	0.5074 (+0.0520)
0.4647	4200	0.5742	-	-	-	-	-
0.4868	4400	0.6171	-	-	-	-	-
0.5090	4600	0.6023	-	-	-	-	-
0.5311	4800	0.5988	-	-	-	-	-
0.5532	5000	0.5693	0.7248 (+0.1336)	0.5174 (-0.0231)	0.3631 (+0.0381)	0.5575 (+0.0569)	0.4793 (+0.0240)
0.5753	5200	0.5783	-	-	-	-	-
0.5975	5400	0.5866	-	-	-	-	-
0.6196	5600	0.543	-	-	-	-	-
0.6417	5800	0.57	-	-	-	-	-
0.6639	6000	0.5662	0.7273 (+0.1361)	0.5148 (-0.0256)	0.3644 (+0.0393)	0.5754 (+0.0748)	0.4849 (+0.0295)
0.6860	6200	0.5605	-	-	-	-	-
0.7081	6400	0.5836	-	-	-	-	-
0.7303	6600	0.5703	-	-	-	-	-
0.7524	6800	0.5732	-	-	-	-	-
0.7745	7000	0.5679	0.7306 (+0.1394)	0.5185 (-0.0219)	0.3767 (+0.0517)	0.5826 (+0.0820)	0.4926 (+0.0372)
0.7966	7200	0.5454	-	-	-	-	-
0.8188	7400	0.5471	-	-	-	-	-
0.8409	7600	0.5592	-	-	-	-	-
0.8630	7800	0.5545	-	-	-	-	-
0.8852	8000	0.5477	0.7314 (+0.1402)	0.5022 (-0.0382)	0.3846 (+0.0596)	0.5854 (+0.0847)	0.4907 (+0.0354)
0.9073	8200	0.5411	-	-	-	-	-
0.9294	8400	0.5299	-	-	-	-	-
0.9515	8600	0.5677	-	-	-	-	-
0.9737	8800	0.5202	-	-	-	-	-
0.9958	9000	0.5211	0.7311 (+0.1399)	0.5090 (-0.0315)	0.3735 (+0.0484)	0.5923 (+0.0916)	0.4916 (+0.0362)
-1	-1	-	0.7314 (+0.1402)	0.5022 (-0.0382)	0.3846 (+0.0596)	0.5854 (+0.0847)	0.4907 (+0.0354)

• The bold row denotes the saved checkpoint.

Environmental Impact

Carbon emissions were measured using CodeCarbon.

• Energy Consumed: 0.143 kWh
• Carbon Emitted: 0.056 kg of CO2
• Hours Used: 0.391 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",
}