Antix5/product-reranker-mmBERT-small

mmBERT-small reranker (LambdaLoss NDCG2++)

This is a Cross Encoder model finetuned from jhu-clsp/mmBERT-small on the product_similarity_dataset 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: jhu-clsp/mmBERT-small
• Maximum Sequence Length: 256 tokens
• Number of Output Labels: 1 label
• Training Dataset:
  • product_similarity_dataset
• Languages: multilingual, fr, de, zh, ru, pl, es, it, ja, ar, hi, pt, nl
• License: mit

Model Sources

• Documentation: Sentence Transformers Documentation
• Documentation: Cross Encoder Documentation
• Repository: Sentence Transformers on GitHub
• Hugging Face: Cross Encoders on Hugging Face

Warning : This model is just starting training, this is just a checkpoint

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("Antix5/product-reranker-mmBERT-small")
# Get scores for pairs of texts
pairs = [
    ['Milk Belgian Chocolate', 'Milk Chocolate Flavor'],
]
scores = model.predict(pairs)
print(scores.shape)
# (3,)

# Or rank different texts based on similarity to a single text
ranks = model.rank(
    '70 % Cacao Dark Chocolate With Coconut',
    [
        'DRK CHCLT BAR, COCONUT',
        'Coconut Cream Filled Dark Chocolate',
        'Blueberry & Dark Chocolate With Chia',
    ]
)
# [{'corpus_id': ..., 'score': ...}, {'corpus_id': ..., 'score': ...}, ...]

Evaluation

Metrics

Cross Encoder Reranking

• Dataset: rerank
• Evaluated with CrossEncoderRerankingEvaluator with these parameters:

  {
      "at_k": 10,
      "always_rerank_positives": false
  }
  

Metric	Value
map	0.9562 (-0.0359)
mrr@10	0.9561 (-0.0385)
ndcg@10	0.9656 (-0.0291)

Training Details

Training Dataset

product_similarity_dataset

• Dataset: product_similarity_dataset at 7aba3ef
• Size: 9,358 training samples
• Columns: query, documents, and scores
• Approximate statistics based on the first 1000 samples:

  	query	documents	scores
  type	string	list	list
  details	min: 6 characters mean: 57.18 characters max: 197 characters	size: 16 elements	size: 16 elements

• Samples:

  query	documents	scores
  Premier 26764 Car Spinner, Santa, 25 by 19-1/2-Inch	['Premier 26764 Tourbillon pour voiture, Santa, 25 x 19-1/2 pouces', 'BNTS, ЧИПСЫ ИЗ ФАСОЛИ NV И МОРСКАЯ СОЛЬ', 'Beanitos, Чипс из фасоли navy, Сыр на чо', 'K2 स्केट व्हील (4 का पैक)', 'BLST BALL МЯЧ ДЛЯ КИКА (2 ШТ.)', ...]	[1.0, 0.0, 0.0, 0.0, 0.0, ...]
  Juice Cocktail Blend From Concentrate, Apple Blueberry	['Mélange de cocktail de jus à base de concentré, pomme myrtille', 'Orange Juice From Concentrate With Pulp', 'Tropical Juice Splash From Concentrate', 'BLUEBERRY JUICE DRNK', 'APPLE NECTAR JUICE DRINK FROM CNCNTRT', ...]	[1.0, 0.4, 0.35, 0.65, 0.55, ...]
  Fruity Sour Strips Fruit-Flavored Chewy Candy	['Fruity Sour Strips Fruit-Flavored Chewy Candy', 'SR CANDIES, FRUIT SOUR', 'Fruit Candy, Fruit', 'FRT SNCK TUTTI FRUITY', 'Fruit Strips, Peach Passion', ...]	[1.0, 0.95, 0.7, 0.55, 0.9, ...]

• Loss: LambdaLoss with these parameters:

  {
      "weighting_scheme": "sentence_transformers.cross_encoder.losses.LambdaLoss.NDCGLoss2PPScheme",
      "k": null,
      "sigma": 1.0,
      "eps": 1e-10,
      "reduction_log": "binary",
      "activation_fn": "torch.nn.modules.linear.Identity",
      "mini_batch_size": null
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• learning_rate: 2e-05
• num_train_epochs: 1
• warmup_ratio: 0.1
• fp16: True
• load_best_model_at_end: True
• gradient_checkpointing: 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: 8
• per_device_eval_batch_size: 8
• 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: 42
• data_seed: None
• jit_mode_eval: False
• use_ipex: False
• bf16: False
• fp16: True
• 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
• 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: True
• 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: False
• 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: False
• prompts: None
• batch_sampler: batch_sampler
• multi_dataset_batch_sampler: proportional
• router_mapping: {}
• learning_rate_mapping: {}

Training Logs

Epoch	Step	Training Loss	rerank_ndcg@10
0.0009	1	0.1261	-
0.0171	20	0.1193	-
0.0342	40	0.0767	-
0.0513	60	0.0563	-
0.0684	80	0.055	-
0.0855	100	0.0546	-
0.1026	120	0.0483	-
0.1197	140	0.0489	-
0.1368	160	0.049	-
0.1538	180	0.0463	-
0.1709	200	0.046	0.9419 (-0.0528)
0.1880	220	0.0411	-
0.2051	240	0.0398	-
0.2222	260	0.0456	-
0.2393	280	0.0463	-
0.2564	300	0.043	-
0.2735	320	0.0447	-
0.2906	340	0.0419	-
0.3077	360	0.0403	-
0.3248	380	0.0429	-
0.3419	400	0.0423	0.9653 (-0.0294)
0.3590	420	0.0406	-
0.3761	440	0.041	-
0.3932	460	0.0427	-
0.4103	480	0.0376	-
0.4274	500	0.0408	-
0.4444	520	0.0394	-
0.4615	540	0.0423	-
0.4786	560	0.0403	-
0.4957	580	0.0336	-
0.5128	600	0.039	0.9668 (-0.0279)
0.5299	620	0.0389	-
0.5470	640	0.0376	-
0.5641	660	0.0422	-
0.5812	680	0.0406	-
0.5983	700	0.037	-
0.6154	720	0.0368	-
0.6325	740	0.0365	-
0.6496	760	0.0356	-
0.6667	780	0.0359	-
0.6838	800	0.0368	0.9646 (-0.0301)
0.7009	820	0.0342	-
0.7179	840	0.0376	-
0.7350	860	0.036	-
0.7521	880	0.0331	-
0.7692	900	0.0341	-
0.7863	920	0.0372	-
0.8034	940	0.0361	-
0.8205	960	0.0352	-
0.8376	980	0.0351	-
0.8547	1000	0.0348	0.9620 (-0.0327)
0.8718	1020	0.0341	-
0.8889	1040	0.0354	-
0.9060	1060	0.035	-
0.9231	1080	0.0325	-
0.9402	1100	0.038	-
0.9573	1120	0.0376	-
0.9744	1140	0.0335	-
0.9915	1160	0.0375	-
-1	-1	-	0.9656 (-0.0291)

Framework Versions

• Python: 3.12.11
• Sentence Transformers: 5.1.0
• Transformers: 4.56.1
• PyTorch: 2.8.0+cu126
• Accelerate: 1.10.1
• Datasets: 2.20.0
• Tokenizers: 0.22.0

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",
}

LambdaLoss

@inproceedings{wang2018lambdaloss,
  title={The LambdaLoss Framework for Ranking Metric Optimization},
  author={Wang, Xuanhui and Li, Cheng and Golbandi, Nadav and Bendersky, Michael and Najork, Marc},
  booktitle={Proceedings of the 27th ACM international conference on information and knowledge management},
  pages={1313--1322},
  year={2018}
}