CrossEncoder based on cross-encoder/ms-marco-MiniLM-L12-v2

This is a Cross Encoder model finetuned from cross-encoder/ms-marco-MiniLM-L12-v2 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: cross-encoder/ms-marco-MiniLM-L12-v2
• Maximum Sequence Length: 512 tokens
• Number of Output Labels: 1 label

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("cross_encoder_model_id")
# Get scores for pairs of texts
pairs = [
    ['the system must identify any risk profile that has expired and is currently marked as overdue to ensure ongoing suitability compliance.', "so, like, your portfolio risk profile is out of date, and i've got a flag here saying it needs renewal before we can do any new trades."],
    ['to identify risk misalignment trades, the system must flag a risk mismatch whenever the product risk rating exceeds the client risk profile.', "so, it's a solid choice, but i gotta mention, there's a bit of a risk mismatch between the fund's rating and your own suitability score, so it's a bit of a hurdle."],
    ['the system identifies an execution only wrapper when the order initiation confirms that this trade is performed on an execution only basis with no advice given.', "so... uh... let's just do it, but it's execution only, you know? no advice was provided, so you're on your own with the strategy on this one, i'm so rushed."],
    ['the system must identify any risk profile that has expired and is currently marked as overdue to ensure ongoing suitability compliance.', "hey, um, checking the dashboard here and it says your prp is overdue, you know, we haven't updated it in a bit and it's flagged."],
    ['to identify risk misalignment trades, the system must flag a risk mismatch whenever the product risk rating exceeds the client risk profile.', "don't worry about the specifics right now the main thing is getting the allocation because it's oversubscribed so can i confirm the trade"],
]
scores = model.predict(pairs)
print(scores.shape)
# (5,)

# Or rank different texts based on similarity to a single text
ranks = model.rank(
    'the system must identify any risk profile that has expired and is currently marked as overdue to ensure ongoing suitability compliance.',
    [
        "so, like, your portfolio risk profile is out of date, and i've got a flag here saying it needs renewal before we can do any new trades.",
        "so, it's a solid choice, but i gotta mention, there's a bit of a risk mismatch between the fund's rating and your own suitability score, so it's a bit of a hurdle.",
        "so... uh... let's just do it, but it's execution only, you know? no advice was provided, so you're on your own with the strategy on this one, i'm so rushed.",
        "hey, um, checking the dashboard here and it says your prp is overdue, you know, we haven't updated it in a bit and it's flagged.",
        "don't worry about the specifics right now the main thing is getting the allocation because it's oversubscribed so can i confirm the trade",
    ]
)
# [{'corpus_id': ..., 'score': ...}, {'corpus_id': ..., 'score': ...}, ...]

Evaluation

Metrics

Cross Encoder Classification

• Dataset: compliance-eval
• Evaluated with CrossEncoderClassificationEvaluator

Metric	Value
accuracy	0.9636
accuracy_threshold	-1.7519
f1	0.9663
f1_threshold	-2.8692
precision	0.9556
recall	0.9773
average_precision	0.994

Training Details

Training Dataset

Unnamed Dataset

• Size: 1,485 training samples
• Columns: sentence1, sentence2, and label
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2	label
  type	string	string	float
  details	min: 135 characters mean: 302.95 characters max: 725 characters	min: 97 characters mean: 179.3 characters max: 463 characters	min: 0.0 mean: 0.49 max: 1.0

• Samples:

  sentence1	sentence2	label
  the rm must use the instrument_code to identify the soft lock disclosure and inform the client that 'this fund has a soft lock-up duration of xx months. you will be subjected to an early redemption charge of x% by the fund house if you were to redeem the fund within the soft lock-up period.' and, if applicable, that 'the fund is currently still within the soft lock-up period. should you wish to proceed with the redemption, you will incur an early redemption charge of x% by the fund house.'	there's a bit of a soft lock on this one, you know, if you take the money out too soon there's a small charge, but it's no big deal.	0.0
  the system identifies an execution only wrapper when the order initiation confirms that this trade is performed on an execution only basis with no advice given.	i can't believe how expensive flights have become lately, it's just ridiculous. let's just go ahead with that stock buy, i'll put it through as we discussed earlier, it’s a simple execution for us.	0.0
  for a client initiated (ci) wrapper where the order initiation is 'client initiated', the bank must confirm that 'this trade is based on your initiated interest in underlying and product type' or 'this trade is based on your initiated interest in underlying or product type'.	exactly, i-i see what you mean, and since you're the one who initiated this conversation about the emerging markets fund, i'll just log that as your interest. did you ever get that classic car fixed up?	1.0

• Loss: BinaryCrossEntropyLoss with these parameters:

  {
      "activation_fn": "torch.nn.modules.linear.Identity",
      "pos_weight": null
  }
  

Evaluation Dataset

Unnamed Dataset

• Size: 165 evaluation samples
• Columns: sentence1, sentence2, and label
• Approximate statistics based on the first 165 samples:

  	sentence1	sentence2	label
  type	string	string	float
  details	min: 135 characters mean: 302.44 characters max: 725 characters	min: 97 characters mean: 178.02 characters max: 631 characters	min: 0.0 mean: 0.53 max: 1.0

• Samples:

  sentence1	sentence2	label
  the system must identify any risk profile that has expired and is currently marked as overdue to ensure ongoing suitability compliance.	so, like, your portfolio risk profile is out of date, and i've got a flag here saying it needs renewal before we can do any new trades.	1.0
  to identify risk misalignment trades, the system must flag a risk mismatch whenever the product risk rating exceeds the client risk profile.	so, it's a solid choice, but i gotta mention, there's a bit of a risk mismatch between the fund's rating and your own suitability score, so it's a bit of a hurdle.	1.0
  the system identifies an execution only wrapper when the order initiation confirms that this trade is performed on an execution only basis with no advice given.	so... uh... let's just do it, but it's execution only, you know? no advice was provided, so you're on your own with the strategy on this one, i'm so rushed.	1.0

• Loss: BinaryCrossEntropyLoss with these parameters:

  {
      "activation_fn": "torch.nn.modules.linear.Identity",
      "pos_weight": null
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• per_device_train_batch_size: 16
• per_device_eval_batch_size: 16
• learning_rate: 2e-05
• warmup_ratio: 0.1
• 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: 3
• 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: 42
• data_seed: None
• jit_mode_eval: False
• bf16: False
• 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}
• parallelism_config: None
• deepspeed: None
• label_smoothing_factor: 0.0
• optim: adamw_torch_fused
• optim_args: None
• adafactor: False
• group_by_length: False
• length_column_name: length
• project: huggingface
• trackio_space_id: trackio
• 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
• hub_revision: None
• 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
• include_tokens_per_second: False
• include_num_input_tokens_seen: no
• neftune_noise_alpha: None
• optim_target_modules: None
• batch_eval_metrics: False
• eval_on_start: False
• use_liger_kernel: False
• liger_kernel_config: None
• eval_use_gather_object: False
• average_tokens_across_devices: True
• prompts: None
• batch_sampler: batch_sampler
• multi_dataset_batch_sampler: proportional
• router_mapping: {}
• learning_rate_mapping: {}

Training Logs

Epoch	Step	Training Loss	Validation Loss	compliance-eval_average_precision
0.1075	10	1.9119	1.1985	0.6783
0.2151	20	0.9675	1.0970	0.6914
0.3226	30	0.7458	0.4725	0.8480
0.4301	40	0.5308	0.4431	0.8849
0.5376	50	0.3888	0.4183	0.9097
0.6452	60	0.3477	0.3472	0.9325
0.7527	70	0.3082	0.3005	0.9524
0.8602	80	0.3364	0.2682	0.9647
0.9677	90	0.3069	0.2345	0.9804
1.0753	100	0.2636	0.1847	0.9886
1.1828	110	0.2577	0.1793	0.9847
1.2903	120	0.1793	0.1940	0.9826
1.3978	130	0.19	0.2333	0.9794
1.5054	140	0.1788	0.1615	0.9858
1.6129	150	0.1277	0.1576	0.9862
1.7204	160	0.1851	0.1399	0.9903
1.8280	170	0.1652	0.1056	0.9947
1.9355	180	0.085	0.1077	0.9949
2.043	190	0.1111	0.0943	0.9955
2.1505	200	0.09	0.1137	0.9955
2.2581	210	0.1136	0.1222	0.9934
2.3656	220	0.0703	0.1155	0.9937
2.4731	230	0.0866	0.1147	0.9935
2.5806	240	0.1104	0.1089	0.9943
2.6882	250	0.1523	0.1141	0.9940
2.7957	260	0.1189	0.1297	0.9943
2.9032	270	0.0479	0.1365	0.9940

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.12.12
• Sentence Transformers: 5.2.0
• Transformers: 4.57.3
• PyTorch: 2.9.0+cu126
• Accelerate: 1.12.0
• Datasets: 4.0.0
• Tokenizers: 0.22.2

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",
}