lohith-chanchu/reranker-gte-multilingual-reranker-base-custom-bce

Reranker trained on Custom Dataset

This is a Cross Encoder model finetuned from Alibaba-NLP/gte-multilingual-reranker-base 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: Alibaba-NLP/gte-multilingual-reranker-base
• Maximum Sequence Length: 8192 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("lohith-chanchu/reranker-gte-multilingual-reranker-base-custom-bce")
# Get scores for pairs of texts
pairs = [
    ['Gaskugelhahn, Gewinde, DN 32 Gaskugelhahn, zum manuellen Absperren, geeignet für Erdgas, PN 6, nach DIN EN 331, Gehäuse aus Pressmessing, in Durchgangsform, beid seits Gewindeanschluss, DIN-DVGW-zugelassen, DN 32, einschließlich Übergangsstücke sowie Verbindungs-, Dichtungs- und Befestigungsma terial', 'DITECH Gas-KH m gelbem Hebelgriff und vollem Durchgang 11/4"'],
    ['Gaskugelhahn, Gewinde, DN 40 jedoch DN 40', 'DITECH Gas-KH m gelbem Hebelgriff und vollem Durchgang 11/2"'],
    ['Gaskugelhahn, Gewinde, DN 50 jedoch DN 50', 'DITECH Gas-KH m gelbem Hebelgriff und vollem Durchgang 2"'],
    ['Doppelnippel, Stahl, DN 15, Montagehöhe bis 6,0 m Doppelnippel, aus Kohlenstoffstahl, für Rohrleitung aus mittelschwerem Stahlrohr DIN EN 10255, mit Außengewinde 1/2 , Montagehöhe üb er Gelände / Fußboden bis 6,0 m', 'HS Rohrdoppelnippel Nr. 23 schwarz 1/2" 100mm'],
    ['Doppelnippel, Stahl, DN 20, Montagehöhe bis 6,0 m jedoch Außengewinde 3/4', 'HS Rohrdoppelnippel Nr. 23 schwarz 3/4" 100mm'],
]
scores = model.predict(pairs)
print(scores.shape)
# (5,)

# Or rank different texts based on similarity to a single text
ranks = model.rank(
    'Gaskugelhahn, Gewinde, DN 32 Gaskugelhahn, zum manuellen Absperren, geeignet für Erdgas, PN 6, nach DIN EN 331, Gehäuse aus Pressmessing, in Durchgangsform, beid seits Gewindeanschluss, DIN-DVGW-zugelassen, DN 32, einschließlich Übergangsstücke sowie Verbindungs-, Dichtungs- und Befestigungsma terial',
    [
        'DITECH Gas-KH m gelbem Hebelgriff und vollem Durchgang 11/4"',
        'DITECH Gas-KH m gelbem Hebelgriff und vollem Durchgang 11/2"',
        'DITECH Gas-KH m gelbem Hebelgriff und vollem Durchgang 2"',
        'HS Rohrdoppelnippel Nr. 23 schwarz 1/2" 100mm',
        'HS Rohrdoppelnippel Nr. 23 schwarz 3/4" 100mm',
    ]
)
# [{'corpus_id': ..., 'score': ...}, {'corpus_id': ..., 'score': ...}, ...]

Evaluation

Metrics

Cross Encoder Reranking

• Dataset: custom-dev
• Evaluated with CrossEncoderRerankingEvaluator with these parameters:

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

Metric	Value
map	0.3148 (+0.1281)
mrr@10	0.3228 (+0.1424)
ndcg@10	0.3455 (+0.1352)

Training Details

Training Dataset

Unnamed Dataset

• Size: 447,164 training samples
• Columns: query, answer, and label
• Approximate statistics based on the first 1000 samples:

  	query	answer	label
  type	string	string	int
  details	min: 27 characters mean: 434.65 characters max: 2905 characters	min: 0 characters mean: 52.08 characters max: 81 characters	0: ~33.70% 1: ~66.30%

• Samples:

  query	answer	label
  Gaskugelhahn, Gewinde, DN 32 Gaskugelhahn, zum manuellen Absperren, geeignet für Erdgas, PN 6, nach DIN EN 331, Gehäuse aus Pressmessing, in Durchgangsform, beid seits Gewindeanschluss, DIN-DVGW-zugelassen, DN 32, einschließlich Übergangsstücke sowie Verbindungs-, Dichtungs- und Befestigungsma terial	DITECH Gas-KH m gelbem Hebelgriff und vollem Durchgang 11/4"	1
  Gaskugelhahn, Gewinde, DN 40 jedoch DN 40	DITECH Gas-KH m gelbem Hebelgriff und vollem Durchgang 11/2"	1
  Gaskugelhahn, Gewinde, DN 50 jedoch DN 50	DITECH Gas-KH m gelbem Hebelgriff und vollem Durchgang 2"	1

• Loss: BinaryCrossEntropyLoss with these parameters:

  {
      "activation_fn": "torch.nn.modules.linear.Identity",
      "pos_weight": 5
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: epoch
• per_device_train_batch_size: 100
• per_device_eval_batch_size: 100
• learning_rate: 2e-05
• num_train_epochs: 2
• 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: epoch
• prediction_loss_only: True
• per_device_train_batch_size: 100
• per_device_eval_batch_size: 100
• 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: 2
• 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}
• 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: 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: 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	custom-dev_ndcg@10
0.0002	1	1.5605	-
0.0224	100	0.9229	-
0.0447	200	0.4384	-
0.0671	300	0.3577	-
0.0894	400	0.3024	-
0.1118	500	0.267	-
0.1342	600	0.2393	-
0.1565	700	0.2228	-
0.1789	800	0.2196	-
0.2013	900	0.1812	-
0.2236	1000	0.2003	-
0.2460	1100	0.1756	-
0.2683	1200	0.1652	-
0.2907	1300	0.1529	-
0.3131	1400	0.1652	-
0.3354	1500	0.1327	-
0.3578	1600	0.1273	-
0.3801	1700	0.124	-
0.4025	1800	0.1371	-
0.4249	1900	0.1239	-
0.4472	2000	0.1252	-
0.4696	2100	0.115	-
0.4919	2200	0.116	-
0.5143	2300	0.1115	-
0.5367	2400	0.1157	-
0.5590	2500	0.1126	-
0.5814	2600	0.1071	-
0.6038	2700	0.1162	-
0.6261	2800	0.1088	-
0.6485	2900	0.1032	-
0.6708	3000	0.1086	-
0.6932	3100	0.0926	-
0.7156	3200	0.0846	-
0.7379	3300	0.0931	-
0.7603	3400	0.1053	-
0.7826	3500	0.0825	-
0.8050	3600	0.1116	-
0.8274	3700	0.0917	-
0.8497	3800	0.0907	-
0.8721	3900	0.0774	-
0.8945	4000	0.0789	-
0.9168	4100	0.0792	-
0.9392	4200	0.0933	-
0.9615	4300	0.0893	-
0.9839	4400	0.0993	-
1.0	4472	-	0.3409 (+0.1306)
1.0063	4500	0.0755	-
1.0286	4600	0.0551	-
1.0510	4700	0.0626	-
1.0733	4800	0.0694	-
1.0957	4900	0.0537	-
1.1181	5000	0.0557	-
1.1404	5100	0.0694	-
1.1628	5200	0.0621	-
1.1852	5300	0.0661	-
1.2075	5400	0.0494	-
1.2299	5500	0.0607	-
1.2522	5600	0.0561	-
1.2746	5700	0.0513	-
1.2970	5800	0.0617	-
1.3193	5900	0.0435	-
1.3417	6000	0.0659	-
1.3640	6100	0.0597	-
1.3864	6200	0.0668	-
1.4088	6300	0.0557	-
1.4311	6400	0.0566	-
1.4535	6500	0.0632	-
1.4758	6600	0.0573	-
1.4982	6700	0.0634	-
1.5206	6800	0.054	-
1.5429	6900	0.0392	-
1.5653	7000	0.046	-
1.5877	7100	0.0562	-
1.6100	7200	0.0443	-
1.6324	7300	0.0757	-
1.6547	7400	0.0555	-
1.6771	7500	0.0345	-
1.6995	7600	0.0525	-
1.7218	7700	0.0595	-
1.7442	7800	0.0561	-
1.7665	7900	0.0484	-
1.7889	8000	0.0465	-
1.8113	8100	0.0501	-
1.8336	8200	0.0411	-
1.8560	8300	0.0386	-
1.8784	8400	0.0477	-
1.9007	8500	0.0517	-
1.9231	8600	0.0338	-
1.9454	8700	0.0466	-
1.9678	8800	0.062	-
1.9902	8900	0.0647	-
2.0	8944	-	0.3455 (+0.1352)
-1	-1	-	0.3455 (+0.1352)

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.10.18
• Sentence Transformers: 5.1.0
• Transformers: 4.56.0
• PyTorch: 2.8.0+cu128
• Accelerate: 1.10.1
• Datasets: 4.0.0
• 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",
}