Upload folder using huggingface_hub

7775c00 verified about 1 month ago

37.5 kB

tags:
  - sentence-transformers
  - sentence-similarity
  - feature-extraction
  - generated_from_trainer
  - dataset_size:342061
  - loss:CachedMultipleNegativesRankingLoss
base_model: prestoai/qwen3-embedding-0.6b-arabic-ecom
widget:
  - source_sentence: >-
      Instruct: Given an Arabic e-commerce search query, retrieve the product
      that best matches it

      Query: نسونكس Spray
    sentences:
      - Nasonex - Nasal Spray
      - كابل شحن مايكرو XKIN - 2.4A
      - حلوى الشوكولاتة - Choco Lapki
  - source_sentence: >-
      Instruct: Given an Arabic e-commerce search query, retrieve the product
      that best matches it

      Query: مكرونة رقم 42
    sentences:
      - بنطلون رجالي - 0112
      - حقيبة حزام خصر - 4862
      - مكرونة الجيد معكوفة رقم 42 - 500 غ
  - source_sentence: >-
      Instruct: Given an Arabic e-commerce search query, retrieve the product
      that best matches it

      Query: سباغيتي
    sentences:
      - ملعب كرة قدم - DD18
      - مكرونة معكوفة - Favelli
      - مكرونة سباغيتي - Favelli
  - source_sentence: >-
      Instruct: Given an Arabic e-commerce search query, retrieve the product
      that best matches it

      Query: جبنة هواء
    sentences:
      - جبنة - ابو الولد
      - جبنة - Hawaa
      - كاني طعام كلاب البالغين دجاج - 3 ك
  - source_sentence: >-
      Instruct: Given an Arabic e-commerce search query, retrieve the product
      that best matches it

      Query: شاحن تايب سي للسيارة
    sentences:
      - شاحن سيارة قرين ليون مدخلين 36 وات مع كابل تايب سي - CBK
      - صوص المكرونة هاينز - 365 غ
      - بسكويت جولون بدون سكر شكلاتة ساندوتش
pipeline_tag: sentence-similarity
library_name: sentence-transformers

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: 128 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("sentence_transformers_model_id")
# Run inference
queries = [
    'Instruct: Given an Arabic e-commerce search query, retrieve the product that best matches it\nQuery: شاحن تايب سي للسيارة',
]
documents = [
    'شاحن سيارة قرين ليون مدخلين 36 وات مع كابل تايب سي - CBK',
    'بسكويت جولون بدون سكر شكلاتة ساندوتش',
    'صوص المكرونة هاينز - 365 غ',
]
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([[ 0.6054, -0.0610, -0.0953]])

Training Details

Training Datasets

pairs_with_negatives

Dataset: pairs_with_negatives
Size: 124,261 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: 29.44 tokens max: 42 tokens	min: 3 tokens mean: 15.95 tokens max: 49 tokens	min: 2 tokens mean: 14.87 tokens max: 40 tokens

Samples:

anchor	positive	negative
`Instruct: Given an Arabic e-commerce search query, retrieve the product that best matches it Query: واقي للوجه دهني`	`Anthelios Oil Control (Dry Touch) - La Roche Posay`	`Anthelios Invisible Mist (Dry Touch) - La Roche Posay`
`Instruct: Given an Arabic e-commerce search query, retrieve the product that best matches it Query: تن منارة زليتن`	`تن منارة زليتن بزيت دوار الشمس - 160 غ`	`تن فاني بزيت دوار الشمس - 160 غ`
`Instruct: Given an Arabic e-commerce search query, retrieve the product that best matches it Query: فانتا زجاجة صغيرة`	`مشروب فانتا برتقال زجاجة - 330 مل`	`مشروب فانتا - 1 ل (برتقال)`

Loss: CachedMultipleNegativesRankingLoss with these parameters:

{
    "scale": 20.0,
    "similarity_fct": "cos_sim",
    "mini_batch_size": 8,
    "gather_across_devices": false,
    "directions": [
        "query_to_doc"
    ],
    "partition_mode": "joint",
    "hardness_mode": null,
    "hardness_strength": 0.0
}

positives

Dataset: positives
Size: 217,800 training samples
Columns: anchor and positive
Approximate statistics based on the first 1000 samples:
anchor positive
type string string
details
min: 23 tokens
mean: 29.61 tokens
max: 52 tokens

min: 2 tokens
mean: 13.81 tokens
max: 41 tokens

	anchor	positive
type	string	string
details	min: 23 tokens mean: 29.61 tokens max: 52 tokens	min: 2 tokens mean: 13.81 tokens max: 41 tokens

Samples:

anchor	positive
`Instruct: Given an Arabic e-commerce search query, retrieve the product that best matches it Query: نحب جبنة القرية`	`ميرسين جبنة القرية 200 جم`
`Instruct: Given an Arabic e-commerce search query, retrieve the product that best matches it Query: كابل شحن مايكرو Moxom A2.4`	`كابل شحن مايكرو Moxom - A2.4`
`Instruct: Given an Arabic e-commerce search query, retrieve the product that best matches it Query: dry idea`	`Dry idea (powder fresh)`

Loss: CachedMultipleNegativesRankingLoss with these parameters:

{
    "scale": 20.0,
    "similarity_fct": "cos_sim",
    "mini_batch_size": 8,
    "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: 1,256 evaluation 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: 29.56 tokens max: 40 tokens	min: 3 tokens mean: 16.04 tokens max: 43 tokens	min: 3 tokens mean: 15.05 tokens max: 33 tokens

Samples:

anchor	positive	negative
`Instruct: Given an Arabic e-commerce search query, retrieve the product that best matches it Query: هيبوتك بوزن`	`عطر Hypnotic Poison - PERFECTO COLLECTION`	`عطر Poison Girl - Dior`
`Instruct: Given an Arabic e-commerce search query, retrieve the product that best matches it Query: شاحن مايكرو 2.4A`	`شحن مايكرو Smila - 2.4A`	`شحن تايب سي Smila - 2.4A`
`Instruct: Given an Arabic e-commerce search query, retrieve the product that best matches it Query: تمر صعيدي`	`تمر صعيدي مشفوط`	`تمر قصيم مشفوط`

Loss: CachedMultipleNegativesRankingLoss with these parameters:

{
    "scale": 20.0,
    "similarity_fct": "cos_sim",
    "mini_batch_size": 8,
    "gather_across_devices": false,
    "directions": [
        "query_to_doc"
    ],
    "partition_mode": "joint",
    "hardness_mode": null,
    "hardness_strength": 0.0
}

positives

Dataset: positives
Size: 2,200 evaluation samples
Columns: anchor and positive
Approximate statistics based on the first 1000 samples:
anchor positive
type string string
details
min: 24 tokens
mean: 29.27 tokens
max: 42 tokens

min: 3 tokens
mean: 13.61 tokens
max: 44 tokens

	anchor	positive
type	string	string
details	min: 24 tokens mean: 29.27 tokens max: 42 tokens	min: 3 tokens mean: 13.61 tokens max: 44 tokens

Samples:

anchor	positive
`Instruct: Given an Arabic e-commerce search query, retrieve the product that best matches it Query: حاملة أدوات القطط`	`حاملة أدوات القطة - AA04`
`Instruct: Given an Arabic e-commerce search query, retrieve the product that best matches it Query: كريم شمس أطفال`	`واقي شمس كريمي - Chicco`
`Instruct: Given an Arabic e-commerce search query, retrieve the product that best matches it Query: كانديسارتان هيدروكلوروثيازيد`	`Candesartan and Hydrochlorothiazide 16mg/12.5mg`

Loss: CachedMultipleNegativesRankingLoss with these parameters:

{
    "scale": 20.0,
    "similarity_fct": "cos_sim",
    "mini_batch_size": 8,
    "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: 32
learning_rate: 0.0001
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: 32
per_device_eval_batch_size: 8
gradient_accumulation_steps: 1
eval_accumulation_steps: None
torch_empty_cache_steps: None
learning_rate: 0.0001
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
0.0023	25	0.4772	-	-
0.0047	50	0.4593	-	-
0.0070	75	0.3945	-	-
0.0094	100	0.3752	-	-
0.0117	125	0.4048	-	-
0.0140	150	0.4563	-	-
0.0164	175	0.3492	-	-
0.0187	200	0.4171	0.3915	0.1481
0.0210	225	0.4297	-	-
0.0234	250	0.4365	-	-
0.0257	275	0.4344	-	-
0.0281	300	0.4184	-	-
0.0304	325	0.4198	-	-
0.0327	350	0.4293	-	-
0.0351	375	0.4759	-	-
0.0374	400	0.3312	0.3695	0.1180
0.0398	425	0.3887	-	-
0.0421	450	0.4402	-	-
0.0444	475	0.4105	-	-
0.0468	500	0.3923	-	-
0.0491	525	0.3163	-	-
0.0514	550	0.3565	-	-
0.0538	575	0.3707	-	-
0.0561	600	0.3008	0.3388	0.1086
0.0585	625	0.3594	-	-
0.0608	650	0.3936	-	-
0.0631	675	0.3207	-	-
0.0655	700	0.3371	-	-
0.0678	725	0.3385	-	-
0.0702	750	0.2718	-	-
0.0725	775	0.4429	-	-
0.0748	800	0.2684	0.3453	0.1043
0.0772	825	0.2539	-	-
0.0795	850	0.3239	-	-
0.0818	875	0.2944	-	-
0.0842	900	0.3067	-	-
0.0865	925	0.3113	-	-
0.0889	950	0.3387	-	-
0.0912	975	0.2735	-	-
0.0935	1000	0.2985	0.3211	0.0891
0.0959	1025	0.3553	-	-
0.0982	1050	0.2568	-	-
0.1006	1075	0.3447	-	-
0.1029	1100	0.3239	-	-
0.1052	1125	0.3015	-	-
0.1076	1150	0.3865	-	-
0.1099	1175	0.2982	-	-
0.1122	1200	0.3105	0.3232	0.0829
0.1146	1225	0.2964	-	-
0.1169	1250	0.2417	-	-
0.1193	1275	0.2686	-	-
0.1216	1300	0.2932	-	-
0.1239	1325	0.2383	-	-
0.1263	1350	0.3108	-	-
0.1286	1375	0.3216	-	-
0.1310	1400	0.2083	0.3091	0.0894
0.1333	1425	0.2933	-	-
0.1356	1450	0.2038	-	-
0.1380	1475	0.2515	-	-
0.1403	1500	0.2643	-	-
0.1426	1525	0.2484	-	-
0.1450	1550	0.3216	-	-
0.1473	1575	0.3265	-	-
0.1497	1600	0.2626	0.3166	0.0775
0.1520	1625	0.2811	-	-
0.1543	1650	0.2792	-	-
0.1567	1675	0.2888	-	-
0.1590	1700	0.3243	-	-
0.1614	1725	0.2318	-	-
0.1637	1750	0.2943	-	-
0.1660	1775	0.2494	-	-
0.1684	1800	0.3478	0.3113	0.0751
0.1707	1825	0.3265	-	-
0.1730	1850	0.2933	-	-
0.1754	1875	0.2671	-	-
0.1777	1900	0.2927	-	-
0.1801	1925	0.2939	-	-
0.1824	1950	0.2356	-	-
0.1847	1975	0.2413	-	-
0.1871	2000	0.2026	0.2921	0.0650
0.1894	2025	0.2663	-	-
0.1918	2050	0.2438	-	-
0.1941	2075	0.2321	-	-
0.1964	2100	0.2482	-	-
0.1988	2125	0.3000	-	-
0.2011	2150	0.1990	-	-
0.2034	2175	0.2393	-	-
0.2058	2200	0.2370	0.2844	0.0670
0.2081	2225	0.2131	-	-
0.2105	2250	0.2548	-	-
0.2128	2275	0.3016	-	-
0.2151	2300	0.1959	-	-
0.2175	2325	0.2604	-	-
0.2198	2350	0.3141	-	-
0.2221	2375	0.2729	-	-
0.2245	2400	0.2492	0.2854	0.0640
0.2268	2425	0.2326	-	-
0.2292	2450	0.2850	-	-
0.2315	2475	0.2393	-	-
0.2338	2500	0.2748	-	-
0.2362	2525	0.2289	-	-
0.2385	2550	0.2486	-	-
0.2409	2575	0.2846	-	-
0.2432	2600	0.2027	0.2798	0.0586
0.2455	2625	0.2336	-	-
0.2479	2650	0.2207	-	-
0.2502	2675	0.2357	-	-
0.2525	2700	0.2132	-	-
0.2549	2725	0.2152	-	-
0.2572	2750	0.2046	-	-
0.2596	2775	0.1824	-	-
0.2619	2800	0.2406	0.2775	0.0624
0.2642	2825	0.2240	-	-
0.2666	2850	0.2538	-	-
0.2689	2875	0.1901	-	-
0.2713	2900	0.2792	-	-
0.2736	2925	0.2489	-	-
0.2759	2950	0.2371	-	-
0.2783	2975	0.2170	-	-
0.2806	3000	0.2408	0.2762	0.0570
0.2829	3025	0.1543	-	-
0.2853	3050	0.2858	-	-
0.2876	3075	0.2290	-	-
0.2900	3100	0.3003	-	-
0.2923	3125	0.2143	-	-
0.2946	3150	0.2486	-	-
0.2970	3175	0.2412	-	-
0.2993	3200	0.2683	0.2755	0.0600
0.3017	3225	0.2805	-	-
0.3040	3250	0.1677	-	-
0.3063	3275	0.2121	-	-
0.3087	3300	0.1980	-	-
0.3110	3325	0.2685	-	-
0.3133	3350	0.2513	-	-
0.3157	3375	0.2237	-	-
0.3180	3400	0.2345	0.2724	0.0598
0.3204	3425	0.1962	-	-
0.3227	3450	0.2863	-	-
0.3250	3475	0.2069	-	-
0.3274	3500	0.2614	-	-
0.3297	3525	0.2608	-	-
0.3321	3550	0.1719	-	-
0.3344	3575	0.1922	-	-
0.3367	3600	0.2148	0.2833	0.0574
0.3391	3625	0.2505	-	-
0.3414	3650	0.2786	-	-
0.3437	3675	0.2462	-	-
0.3461	3700	0.2319	-	-
0.3484	3725	0.2415	-	-
0.3508	3750	0.2363	-	-
0.3531	3775	0.1833	-	-
0.3554	3800	0.2266	0.2764	0.0558
0.3578	3825	0.1910	-	-
0.3601	3850	0.2883	-	-
0.3625	3875	0.1867	-	-
0.3648	3900	0.2211	-	-
0.3671	3925	0.2308	-	-
0.3695	3950	0.2339	-	-
0.3718	3975	0.2290	-	-
0.3741	4000	0.2557	0.2681	0.0557
0.3765	4025	0.2158	-	-
0.3788	4050	0.2279	-	-
0.3812	4075	0.2348	-	-
0.3835	4100	0.3228	-	-
0.3858	4125	0.2015	-	-
0.3882	4150	0.1773	-	-
0.3905	4175	0.2194	-	-
0.3929	4200	0.2076	0.2610	0.0545
0.3952	4225	0.2413	-	-
0.3975	4250	0.2106	-	-
0.3999	4275	0.2011	-	-
0.4022	4300	0.2752	-	-
0.4045	4325	0.2190	-	-
0.4069	4350	0.2234	-	-
0.4092	4375	0.2648	-	-
0.4116	4400	0.2325	0.2550	0.0508
0.4139	4425	0.3020	-	-
0.4162	4450	0.2301	-	-
0.4186	4475	0.1998	-	-
0.4209	4500	0.2532	-	-
0.4233	4525	0.2002	-	-
0.4256	4550	0.2243	-	-
0.4279	4575	0.2274	-	-
0.4303	4600	0.1878	0.2571	0.0476
0.4326	4625	0.2324	-	-
0.4349	4650	0.1931	-	-
0.4373	4675	0.2420	-	-
0.4396	4700	0.2707	-	-
0.4420	4725	0.2175	-	-
0.4443	4750	0.2280	-	-
0.4466	4775	0.2171	-	-
0.4490	4800	0.2313	0.2441	0.0505
0.4513	4825	0.2048	-	-
0.4537	4850	0.2170	-	-
0.4560	4875	0.2318	-	-
0.4583	4900	0.2193	-	-
0.4607	4925	0.2338	-	-
0.4630	4950	0.2273	-	-
0.4653	4975	0.2223	-	-
0.4677	5000	0.2221	0.2421	0.0469

Training Time

Training: 5.4 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",
}

CachedMultipleNegativesRankingLoss

@misc{gao2021scaling,
    title={Scaling Deep Contrastive Learning Batch Size under Memory Limited Setup},
    author={Luyu Gao and Yunyi Zhang and Jiawei Han and Jamie Callan},
    year={2021},
    eprint={2101.06983},
    archivePrefix={arXiv},
    primaryClass={cs.LG}
}