bslm-triplet-2.6M-qp

This is a sentence-transformers bi-encoder continued from mjaliz/bslm-pair-206k and trained on 2,673,575 ecommerce (query, positive product title, hard negative product title) triplets with TripletLoss. It maps Persian ecommerce queries and product titles to a 1024-dimensional dense vector space for retrieval.

Model Details

Model Description

• Model Type: Sentence Transformer
• Base model: mjaliz/bslm-pair-206k
• Maximum Sequence Length: 256 tokens
• Output Dimensionality: 1024 dimensions
• Similarity Function: Cosine Similarity
• Supported Modality: Text
• Training Dataset:
  • json

Model Sources

• Documentation: Sentence Transformers Documentation
• Repository: Sentence Transformers on GitHub
• Hugging Face: Sentence Transformers on Hugging Face

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'transformer_task': 'feature-extraction', 'modality_config': {'text': {'method': 'forward', 'method_output_name': 'last_hidden_state'}}, 'module_output_name': 'token_embeddings', 'architecture': 'XLMRobertaModel'})
  (1): Pooling({'embedding_dimension': 1024, 'pooling_mode': 'mean', 'include_prompt': True})
)

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 SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("sentence_transformers_model_id")
# Run inference
sentences = [
    'پارچه چادری تور',
    'پارچه تور گیپور کارشده مجلسی با دوخت طلایی و نگین کریستالی',
    'کلمن شربت خوری پایه دار دیاموند ARCO مدل کریستالی مجلسی',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 1024]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities)
# tensor([[ 1.0000,  0.6841, -0.0943],
#         [ 0.6841,  1.0000,  0.0870],
#         [-0.0943,  0.0870,  1.0000]])

Evaluation

Metrics

Triplet

• Dataset: dev
• Evaluated with TripletEvaluator

Held-out hard-negative test set

• Dataset: data/triplets/test.jsonl
• Size: 27,300 triplets
• Evaluated with TripletEvaluator and cosine similarity.

Training Details

Training Dataset

json

• Dataset: json
• Size: 2,673,575 training samples
• Columns: anchor, positive, and negative
• Approximate statistics based on the first 100 samples:

  	anchor	positive	negative
  type	string	string	string
  modality	text	text	text
  details	min: 5 tokensmean: 8.45 tokensmax: 12 tokens	min: 12 tokensmean: 22.57 tokensmax: 32 tokens	min: 10 tokensmean: 21.52 tokensmax: 36 tokens

• Samples:

  anchor	positive	negative
  مانتو تابستانه سایزبزرگ	مانتو زنانه بلند کتان لمه مهتابان (Mahtaban) مدل قیطون‌دوزی شده سایزبزرگ	عروسک روسی خانم معلم با مانتو و شلوار رسمی، کتاب فارسی و کیف دستی
  مانتو تابستانه سایزبزرگ	مانتو زنانه بلند کتان لمه مهتابان (Mahtaban) مدل قیطون‌دوزی شده سایزبزرگ	مینی فیگور لگو هارلی کویین (Harley Quinn) دیسی طرح لوزی قرمز و مشکی
  شال توری بهاره	شال توری مشکی لبه گیپور دار	عروسک پولیشی Kuromi (کورومی) مدل لباس توری بنفش چشمک‌زن

• Loss: TripletLoss with these parameters:

  {
      "distance_metric": "TripletDistanceMetric.COSINE",
      "triplet_margin": 0.2
  }
  

Evaluation Dataset

json

• Dataset: json
• Size: 10,000 evaluation samples
• Columns: anchor, positive, and negative
• Approximate statistics based on the first 100 samples:

  	anchor	positive	negative
  type	string	string	string
  modality	text	text	text
  details	min: 6 tokensmean: 7.16 tokensmax: 9 tokens	min: 15 tokensmean: 22.46 tokensmax: 36 tokens	min: 13 tokensmean: 22.34 tokensmax: 32 tokens

• Samples:

  anchor	positive	negative
  وایت ویت	فلوئید ضد آفتاب Vitalayer (ویتالیر) Whitevit (وایت ویت) SPF50+ ضد لک و مقاوم در برابر تعریق	کاور محافظ موبایل AF038 طرح استتار Off-White (آف-وایت) برای هواوی Nova 7i/P40 Lite
  وایت ویت	فلوئید ضد آفتاب Vitalayer (ویتالیر) Whitevit (وایت ویت) SPF50+ ضد لک و مقاوم در برابر تعریق	پاکت ساندویچ Penguin (پنگوئن) 100 عددی یکبار مصرف مقاوم در برابر چربی
  وایت ویت	فلوئید ضد آفتاب Vitalayer (ویتالیر) Whitevit (وایت ویت) SPF50+ ضد لک و مقاوم در برابر تعریق	اسپری ضد آفتاب کودک Hydroderm (هیدرودرم) SPF 30 فیزیکال حجم 100 میلی لیتر

• Loss: TripletLoss with these parameters:

  {
      "distance_metric": "TripletDistanceMetric.COSINE",
      "triplet_margin": 0.2
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• per_device_train_batch_size: 32
• num_train_epochs: 1.0
• learning_rate: 5e-06
• warmup_steps: 0.1
• weight_decay: 0.01
• bf16: True
• per_device_eval_batch_size: 256
• load_best_model_at_end: True
• dataloader_num_workers: 4

All Hyperparameters

• per_device_train_batch_size: 32
• num_train_epochs: 1.0
• max_steps: -1
• learning_rate: 5e-06
• lr_scheduler_type: linear
• lr_scheduler_kwargs: None
• warmup_steps: 0.1
• optim: adamw_torch_fused
• optim_args: None
• weight_decay: 0.01
• adam_beta1: 0.9
• adam_beta2: 0.999
• adam_epsilon: 1e-08
• optim_target_modules: None
• gradient_accumulation_steps: 1
• average_tokens_across_devices: True
• max_grad_norm: 1.0
• label_smoothing_factor: 0.0
• bf16: True
• fp16: False
• bf16_full_eval: False
• fp16_full_eval: False
• tf32: None
• gradient_checkpointing: False
• gradient_checkpointing_kwargs: None
• torch_compile: False
• torch_compile_backend: None
• torch_compile_mode: None
• use_liger_kernel: False
• liger_kernel_config: None
• use_cache: False
• neftune_noise_alpha: None
• torch_empty_cache_steps: None
• auto_find_batch_size: False
• log_on_each_node: True
• logging_nan_inf_filter: True
• include_num_input_tokens_seen: no
• log_level: passive
• log_level_replica: warning
• disable_tqdm: False
• project: huggingface
• trackio_space_id: None
• trackio_bucket_id: None
• trackio_static_space_id: None
• per_device_eval_batch_size: 256
• prediction_loss_only: True
• eval_on_start: False
• eval_do_concat_batches: True
• eval_use_gather_object: False
• eval_accumulation_steps: None
• include_for_metrics: []
• batch_eval_metrics: False
• save_only_model: False
• save_on_each_node: False
• enable_jit_checkpoint: False
• push_to_hub: False
• hub_private_repo: None
• hub_model_id: None
• hub_strategy: every_save
• hub_always_push: False
• hub_revision: None
• load_best_model_at_end: True
• ignore_data_skip: False
• restore_callback_states_from_checkpoint: False
• full_determinism: False
• seed: 42
• data_seed: None
• use_cpu: False
• 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
• dataloader_drop_last: True
• dataloader_num_workers: 4
• dataloader_pin_memory: True
• dataloader_persistent_workers: False
• dataloader_prefetch_factor: None
• remove_unused_columns: True
• label_names: None
• train_sampling_strategy: random
• length_column_name: length
• ddp_find_unused_parameters: None
• ddp_bucket_cap_mb: None
• ddp_broadcast_buffers: False
• ddp_static_graph: None
• ddp_backend: None
• ddp_timeout: 1800
• fsdp: None
• fsdp_config: None
• deepspeed: None
• debug: []
• skip_memory_metrics: True
• do_predict: False
• resume_from_checkpoint: None
• warmup_ratio: None
• local_rank: -1
• prompts: None
• batch_sampler: batch_sampler
• multi_dataset_batch_sampler: proportional
• router_mapping: {}
• learning_rate_mapping: {}

Training Logs

• The bold row denotes the saved checkpoint.

Training Time

• Training: 17.8 hours
• Evaluation: 10.1 minutes
• Total: 18.0 hours

Framework Versions

• Python: 3.11.14
• Sentence Transformers: 5.6.0
• Transformers: 5.12.1
• PyTorch: 2.12.1+cu130
• Accelerate: 1.14.0
• Datasets: 5.0.0
• 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",
}

TripletLoss

@misc{hermans2017defense,
    title={In Defense of the Triplet Loss for Person Re-Identification},
    author={Alexander Hermans and Lucas Beyer and Bastian Leibe},
    year={2017},
    eprint={1703.07737},
    archivePrefix={arXiv},
    primaryClass={cs.CV}
}