leafxyz/main

SentenceTransformer based on prestoai/qwen3-embedding-0.6b-arabic-ecom

This is a sentence-transformers model finetuned from prestoai/qwen3-embedding-0.6b-arabic-ecom on the pairs_with_negatives and positives datasets. It maps sentences & paragraphs to a 1024-dimensional dense vector space and can be used for retrieval.

Model Details

Model Description

• Model Type: Sentence Transformer
• Base model: prestoai/qwen3-embedding-0.6b-arabic-ecom
• Maximum Sequence Length: 64 tokens
• Output Dimensionality: 1024 dimensions
• Similarity Function: Cosine Similarity
• Supported Modality: Text
• Training Datasets:
  • pairs_with_negatives
  • positives

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': 'Qwen3Model'})
  (1): Pooling({'embedding_dimension': 1024, 'pooling_mode': 'lasttoken', 'include_prompt': True})
  (2): Normalize({})
)

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("leafxyz/main")
# Run inference
queries = [
    'Instruct: Given an Arabic e-commerce search query, retrieve the product that best matches it\nQuery: شاحن ايفون BAVIN',
]
documents = [
    'كابل شحن ايفون  BAVIN - 2.4A CB-015',
    'كابل شحن تايب سي BAVIN -  2.4A',
    'مزيل طلاء اظافر  04 - Acetone',
]
query_embeddings = model.encode_query(queries)
document_embeddings = model.encode_document(documents)
print(query_embeddings.shape, document_embeddings.shape)
# [1, 1024] [3, 1024]

# Get the similarity scores for the embeddings
similarities = model.similarity(query_embeddings, document_embeddings)
print(similarities)
# tensor([[ 0.6069,  0.4563, -0.0245]])

Training Details

Training Datasets

pairs_with_negatives

• Dataset: pairs_with_negatives
• Size: 23,680 training samples
• Columns: anchor, positive, and negative
• Approximate statistics based on the first 1000 samples:

  	anchor	positive	negative
  type	string	string	string
  details	min: 23 tokens mean: 29.53 tokens max: 41 tokens	min: 3 tokens mean: 15.91 tokens max: 40 tokens	min: 3 tokens mean: 15.13 tokens max: 38 tokens

• Samples:

  anchor	positive	negative
  Instruct: Given an Arabic e-commerce search query, retrieve the product that best matches it Query: صدر دجاج كمون	صدر دجاج بالعظم	صدر دجاج مجمد - الاولى
  Instruct: Given an Arabic e-commerce search query, retrieve the product that best matches it Query: حليب	حليب الزهرات - 410 غ	بدلة نسائية - 051
  Instruct: Given an Arabic e-commerce search query, retrieve the product that best matches it Query: عصائر	عصير بيلو برتقال - 330 غ	D5699-زي تنكري

• Loss: MultipleNegativesRankingLoss with these parameters:

  {
      "scale": 20.0,
      "similarity_fct": "cos_sim",
      "gather_across_devices": false,
      "directions": [
          "query_to_doc"
      ],
      "partition_mode": "joint",
      "hardness_mode": null,
      "hardness_strength": 0.0
  }
  

positives

• Dataset: positives
• Size: 105,019 training samples
• Columns: anchor and positive
• Approximate statistics based on the first 1000 samples:

  	anchor	positive
  type	string	string
  details	min: 23 tokens mean: 29.41 tokens max: 41 tokens	min: 2 tokens mean: 13.66 tokens max: 36 tokens

• Samples:

  anchor	positive
  Instruct: Given an Arabic e-commerce search query, retrieve the product that best matches it Query: نحب مكرونة خرز	مكرونة خرز 2
  Instruct: Given an Arabic e-commerce search query, retrieve the product that best matches it Query: نحب حذاء افراح فضي	حذاء افراح - 5142wo
  Instruct: Given an Arabic e-commerce search query, retrieve the product that best matches it Query: بودرة بدون عطور	Baby Powder - Nunu

• Loss: MultipleNegativesRankingLoss with these parameters:

  {
      "scale": 20.0,
      "similarity_fct": "cos_sim",
      "gather_across_devices": false,
      "directions": [
          "query_to_doc"
      ],
      "partition_mode": "joint",
      "hardness_mode": null,
      "hardness_strength": 0.0
  }
  

Evaluation Datasets

pairs_with_negatives

• Dataset: pairs_with_negatives
• Size: 240 evaluation samples
• Columns: anchor, positive, and negative
• Approximate statistics based on the first 240 samples:

  	anchor	positive	negative
  type	string	string	string
  details	min: 24 tokens mean: 29.85 tokens max: 41 tokens	min: 3 tokens mean: 16.65 tokens max: 40 tokens	min: 3 tokens mean: 15.61 tokens max: 33 tokens

• Samples:

  anchor	positive	negative
  Instruct: Given an Arabic e-commerce search query, retrieve the product that best matches it Query: سماعات لاسلكية	سماعات بلوتوث - Moxom	كريم مزيل عرق بالجلسرين - Roncey
  Instruct: Given an Arabic e-commerce search query, retrieve the product that best matches it Query: نشتي تابل فلفل	تابل كارمنسيتا 4 انواع فلفل مطحنة - 145 غ	تابل كارمنسيتا فلفل اسود حب مطحنة -47 غ
  Instruct: Given an Arabic e-commerce search query, retrieve the product that best matches it Query: صلصة تريكي حدائق سويسرا 250	صلصة التريكي حدائق سويسرا - 250 مل	صلصة وسترشاير حدائق سويسرا - 250 مل

• Loss: MultipleNegativesRankingLoss with these parameters:

  {
      "scale": 20.0,
      "similarity_fct": "cos_sim",
      "gather_across_devices": false,
      "directions": [
          "query_to_doc"
      ],
      "partition_mode": "joint",
      "hardness_mode": null,
      "hardness_strength": 0.0
  }
  

positives

• Dataset: positives
• Size: 1,061 evaluation samples
• Columns: anchor and positive
• Approximate statistics based on the first 1000 samples:

  	anchor	positive
  type	string	string
  details	min: 23 tokens mean: 29.35 tokens max: 45 tokens	min: 3 tokens mean: 13.76 tokens max: 37 tokens

• Samples:

  anchor	positive
  Instruct: Given an Arabic e-commerce search query, retrieve the product that best matches it Query: كفر	كفر - 876
  Instruct: Given an Arabic e-commerce search query, retrieve the product that best matches it Query: بيرير شعير	بيرير شعير 330ملي
  Instruct: Given an Arabic e-commerce search query, retrieve the product that best matches it Query: مكيف هواء 01 1.5 طن	مكيف هواء - 01

• Loss: MultipleNegativesRankingLoss with these parameters:

  {
      "scale": 20.0,
      "similarity_fct": "cos_sim",
      "gather_across_devices": false,
      "directions": [
          "query_to_doc"
      ],
      "partition_mode": "joint",
      "hardness_mode": null,
      "hardness_strength": 0.0
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• per_device_train_batch_size: 16
• gradient_accumulation_steps: 2
• learning_rate: 3e-05
• num_train_epochs: 1
• warmup_steps: 0.05
• fp16: True

All Hyperparameters

Click to expand

• do_predict: False
• prediction_loss_only: True
• per_device_train_batch_size: 16
• per_device_eval_batch_size: 8
• gradient_accumulation_steps: 2
• eval_accumulation_steps: None
• torch_empty_cache_steps: None
• learning_rate: 3e-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: None
• warmup_ratio: None
• warmup_steps: 0.05
• log_level: passive
• log_level_replica: warning
• log_on_each_node: True
• logging_nan_inf_filter: True
• enable_jit_checkpoint: False
• save_on_each_node: False
• save_only_model: False
• restore_callback_states_from_checkpoint: False
• use_cpu: False
• seed: 42
• data_seed: None
• bf16: False
• fp16: True
• bf16_full_eval: False
• fp16_full_eval: False
• tf32: None
• local_rank: -1
• ddp_backend: None
• debug: []
• dataloader_drop_last: False
• dataloader_num_workers: 0
• dataloader_prefetch_factor: None
• disable_tqdm: False
• remove_unused_columns: True
• label_names: None
• load_best_model_at_end: False
• ignore_data_skip: False
• fsdp: []
• fsdp_config: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': 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
• deepspeed: None
• label_smoothing_factor: 0.0
• optim: adamw_torch_fused
• optim_args: None
• 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
• 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_for_metrics: []
• eval_do_concat_batches: True
• auto_find_batch_size: False
• full_determinism: False
• ddp_timeout: 1800
• torch_compile: False
• torch_compile_backend: None
• torch_compile_mode: None
• 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
• use_cache: False
• prompts: None
• batch_sampler: batch_sampler
• multi_dataset_batch_sampler: proportional
• router_mapping: {}
• learning_rate_mapping: {}

Training Logs

Click to expand

Epoch	Step	Training Loss	pairs with negatives loss	positives loss
0.0062	25	0.3026	-	-
0.0124	50	0.2774	-	-
0.0186	75	0.2329	-	-
0.0249	100	0.3215	-	-
0.0311	125	0.2840	-	-
0.0373	150	0.2273	-	-
0.0435	175	0.2653	-	-
0.0497	200	0.2954	0.4950	0.1348
0.0559	225	0.2065	-	-
0.0622	250	0.2829	-	-
0.0684	275	0.3483	-	-
0.0746	300	0.2630	-	-
0.0808	325	0.2213	-	-
0.0870	350	0.2100	-	-
0.0932	375	0.1968	-	-
0.0995	400	0.2757	0.4839	0.1207
0.1057	425	0.2429	-	-
0.1119	450	0.2036	-	-
0.1181	475	0.2323	-	-
0.1243	500	0.2300	-	-
0.1305	525	0.2574	-	-
0.1367	550	0.1876	-	-
0.1430	575	0.2610	-	-
0.1492	600	0.1821	0.4610	0.1138
0.1554	625	0.2224	-	-
0.1616	650	0.2094	-	-
0.1678	675	0.2207	-	-
0.1740	700	0.2337	-	-
0.1803	725	0.2254	-	-
0.1865	750	0.1587	-	-
0.1927	775	0.2680	-	-
0.1989	800	0.2329	0.4287	0.1062
0.2051	825	0.1858	-	-
0.2113	850	0.2216	-	-
0.2176	875	0.1858	-	-
0.2238	900	0.2059	-	-
0.2300	925	0.2509	-	-
0.2362	950	0.2116	-	-
0.2424	975	0.2355	-	-
0.2486	1000	0.2078	0.4243	0.0990
0.2548	1025	0.2096	-	-
0.2611	1050	0.1906	-	-
0.2673	1075	0.2119	-	-
0.2735	1100	0.2264	-	-
0.2797	1125	0.1925	-	-
0.2859	1150	0.1925	-	-
0.2921	1175	0.1872	-	-
0.2984	1200	0.2092	0.4205	0.0944
0.3046	1225	0.2271	-	-
0.3108	1250	0.1920	-	-
0.3170	1275	0.2476	-	-
0.3232	1300	0.2103	-	-
0.3294	1325	0.1690	-	-
0.3357	1350	0.1838	-	-
0.3419	1375	0.1581	-	-
0.3481	1400	0.2045	0.4076	0.0892
0.3543	1425	0.1498	-	-
0.3605	1450	0.2218	-	-
0.3667	1475	0.1973	-	-
0.3729	1500	0.1979	-	-
0.3792	1525	0.1658	-	-
0.3854	1550	0.2064	-	-
0.3916	1575	0.2152	-	-
0.3978	1600	0.2369	0.4015	0.0870
0.4040	1625	0.2124	-	-
0.4102	1650	0.1714	-	-
0.4165	1675	0.1537	-	-
0.4227	1700	0.1945	-	-
0.4289	1725	0.1481	-	-
0.4351	1750	0.1522	-	-
0.4413	1775	0.1983	-	-
0.4475	1800	0.1636	0.4106	0.0826
0.4538	1825	0.1199	-	-
0.4600	1850	0.1897	-	-
0.4662	1875	0.2022	-	-
0.4724	1900	0.1831	-	-
0.4786	1925	0.1973	-	-
0.4848	1950	0.1378	-	-
0.4910	1975	0.1843	-	-
0.4973	2000	0.1823	0.3930	0.0851
0.5035	2025	0.1733	-	-
0.5097	2050	0.1825	-	-
0.5159	2075	0.1704	-	-
0.5221	2100	0.1721	-	-
0.5283	2125	0.2249	-	-
0.5346	2150	0.1580	-	-
0.5408	2175	0.1572	-	-
0.5470	2200	0.1545	0.3999	0.0811
0.5532	2225	0.1840	-	-
0.5594	2250	0.1930	-	-
0.5656	2275	0.1541	-	-
0.5719	2300	0.1881	-	-
0.5781	2325	0.1268	-	-
0.5843	2350	0.1661	-	-
0.5905	2375	0.1613	-	-
0.5967	2400	0.1662	0.4001	0.0775
0.6029	2425	0.2088	-	-
0.6091	2450	0.1483	-	-
0.6154	2475	0.1487	-	-
0.6216	2500	0.1636	-	-
0.6278	2525	0.1632	-	-
0.6340	2550	0.1878	-	-
0.6402	2575	0.2039	-	-
0.6464	2600	0.1754	0.3891	0.0750
0.6527	2625	0.1898	-	-
0.6589	2650	0.1902	-	-
0.6651	2675	0.1900	-	-
0.6713	2700	0.1797	-	-
0.6775	2725	0.1806	-	-
0.6837	2750	0.1478	-	-
0.6900	2775	0.1667	-	-
0.6962	2800	0.1471	0.3865	0.0731
0.7024	2825	0.1262	-	-
0.7086	2850	0.1921	-	-
0.7148	2875	0.1559	-	-
0.7210	2900	0.1194	-	-
0.7273	2925	0.1569	-	-
0.7335	2950	0.1295	-	-
0.7397	2975	0.1444	-	-
0.7459	3000	0.1667	0.3877	0.0723
0.7521	3025	0.1636	-	-
0.7583	3050	0.1669	-	-
0.7645	3075	0.1424	-	-
0.7708	3100	0.1300	-	-
0.7770	3125	0.1584	-	-
0.7832	3150	0.1857	-	-
0.7894	3175	0.1787	-	-
0.7956	3200	0.1710	0.3773	0.0718
0.8018	3225	0.1579	-	-
0.8081	3250	0.1875	-	-
0.8143	3275	0.1241	-	-
0.8205	3300	0.1421	-	-
0.8267	3325	0.1767	-	-
0.8329	3350	0.1506	-	-
0.8391	3375	0.1519	-	-
0.8454	3400	0.1404	0.3784	0.0710
0.8516	3425	0.1597	-	-
0.8578	3450	0.1692	-	-
0.8640	3475	0.1806	-	-
0.8702	3500	0.1635	-	-
0.8764	3525	0.1924	-	-
0.8826	3550	0.1493	-	-
0.8889	3575	0.1444	-	-
0.8951	3600	0.1664	0.3793	0.0710
0.9013	3625	0.1612	-	-
0.9075	3650	0.1231	-	-
0.9137	3675	0.1670	-	-
0.9199	3700	0.1440	-	-
0.9262	3725	0.1956	-	-
0.9324	3750	0.1827	-	-
0.9386	3775	0.1263	-	-
0.9448	3800	0.1520	0.3762	0.0699
0.9510	3825	0.1412	-	-
0.9572	3850	0.1800	-	-
0.9635	3875	0.1551	-	-
0.9697	3900	0.1345	-	-
0.9759	3925	0.1644	-	-
0.9821	3950	0.0972	-	-
0.9883	3975	0.1284	-	-
0.9945	4000	0.1377	0.3746	0.0697

Training Time

• Training: 1.5 hours

Framework Versions

• Python: 3.12.13
• Sentence Transformers: 5.4.1
• Transformers: 5.0.0
• PyTorch: 2.10.0+cu128
• Accelerate: 1.13.0
• 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",
}

MultipleNegativesRankingLoss

@misc{oord2019representationlearningcontrastivepredictive,
      title={Representation Learning with Contrastive Predictive Coding},
      author={Aaron van den Oord and Yazhe Li and Oriol Vinyals},
      year={2019},
      eprint={1807.03748},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/1807.03748},
}