README.md

metadata

language:
  - ar
  - en
tags:
  - sentence-transformers
  - sentence-similarity
  - feature-extraction
  - generated_from_trainer
  - dataset_size:34436
  - loss:MatryoshkaLoss
  - loss:CoSENTLoss
base_model: AhmedZaky1/DIMI-embedding-v2
widget:
  - source_sentence: الرجل يركب حصاناً
    sentences:
      - رجل يُبث الجبن الممزق على البيتزا
      - ar-ar
      - رجل يركب حصاناً
  - source_sentence: المرأة تقلي لحم خنزير مشوي
    sentences:
      - ar-ar
      - امرأة تطبخ لحم خنزير مخبوز
      - طائرة طيران تقلع
  - source_sentence: امرأة تحمل في ذراعها طفل كنغر
    sentences:
      - امرأة تعزف على الغيتار
      - ar-ar
      - امرأة تحمل و تحمل طفل كنغر
  - source_sentence: رجل يعزف على الناي
    sentences:
      - طائرة ستقلع
      - ar-ar
      - رجل يعزف على فرقة الخيزران
  - source_sentence: ثلاثة رجال يلعبون الشطرنج.
    sentences:
      - رجلين يلعبان الشطرنج
      - بعض الرجال يقاتلون
      - ar-ar
datasets:
  - silma-ai/silma-arabic-english-sts-dataset-v1.0
pipeline_tag: sentence-similarity
library_name: sentence-transformers
metrics:
  - pearson_cosine
  - spearman_cosine
model-index:
  - name: SentenceTransformer based on AhmedZaky1/DIMI-embedding-v2
    results:
      - task:
          type: semantic-similarity
          name: Semantic Similarity
        dataset:
          name: silma sts dev 768
          type: silma-sts-dev-768
        metrics:
          - type: pearson_cosine
            value: 0.8894298077237747
            name: Pearson Cosine
          - type: spearman_cosine
            value: 0.8357984695231979
            name: Spearman Cosine
      - task:
          type: semantic-similarity
          name: Semantic Similarity
        dataset:
          name: silma sts dev 512
          type: silma-sts-dev-512
        metrics:
          - type: pearson_cosine
            value: 0.8958835653694187
            name: Pearson Cosine
          - type: spearman_cosine
            value: 0.8394578198917563
            name: Spearman Cosine
      - task:
          type: semantic-similarity
          name: Semantic Similarity
        dataset:
          name: silma sts dev 256
          type: silma-sts-dev-256
        metrics:
          - type: pearson_cosine
            value: 0.9078743376141943
            name: Pearson Cosine
          - type: spearman_cosine
            value: 0.8470163055535588
            name: Spearman Cosine
      - task:
          type: semantic-similarity
          name: Semantic Similarity
        dataset:
          name: silma sts dev 128
          type: silma-sts-dev-128
        metrics:
          - type: pearson_cosine
            value: 0.9181556833949818
            name: Pearson Cosine
          - type: spearman_cosine
            value: 0.856188415278301
            name: Spearman Cosine
      - task:
          type: semantic-similarity
          name: Semantic Similarity
        dataset:
          name: silma sts dev 64
          type: silma-sts-dev-64
        metrics:
          - type: pearson_cosine
            value: 0.9066219844975816
            name: Pearson Cosine
          - type: spearman_cosine
            value: 0.8434430083292863
            name: Spearman Cosine
      - task:
          type: semantic-similarity
          name: Semantic Similarity
        dataset:
          name: sts17 ar test 768
          type: sts17-ar-test-768
        metrics:
          - type: pearson_cosine
            value: 0.8205269118955641
            name: Pearson Cosine
          - type: spearman_cosine
            value: 0.8258003312254673
            name: Spearman Cosine
      - task:
          type: semantic-similarity
          name: Semantic Similarity
        dataset:
          name: sts17 ar test 512
          type: sts17-ar-test-512
        metrics:
          - type: pearson_cosine
            value: 0.8193403796404517
            name: Pearson Cosine
          - type: spearman_cosine
            value: 0.8226611918447921
            name: Spearman Cosine
      - task:
          type: semantic-similarity
          name: Semantic Similarity
        dataset:
          name: sts17 ar test 256
          type: sts17-ar-test-256
        metrics:
          - type: pearson_cosine
            value: 0.8190666923783347
            name: Pearson Cosine
          - type: spearman_cosine
            value: 0.8245760514866052
            name: Spearman Cosine
      - task:
          type: semantic-similarity
          name: Semantic Similarity
        dataset:
          name: sts17 ar test 128
          type: sts17-ar-test-128
        metrics:
          - type: pearson_cosine
            value: 0.8114629254813825
            name: Pearson Cosine
          - type: spearman_cosine
            value: 0.8183273799928091
            name: Spearman Cosine
      - task:
          type: semantic-similarity
          name: Semantic Similarity
        dataset:
          name: sts17 ar test 64
          type: sts17-ar-test-64
        metrics:
          - type: pearson_cosine
            value: 0.796172574267003
            name: Pearson Cosine
          - type: spearman_cosine
            value: 0.8077141358495715
            name: Spearman Cosine

SentenceTransformer based on AhmedZaky1/DIMI-embedding-v2

This is a sentence-transformers model finetuned from AhmedZaky1/DIMI-embedding-v2 on the silma-arabic-english-sts-dataset-v1.0 dataset. It maps sentences & paragraphs to a 768-dimensional dense vector space and can be used for semantic textual similarity, semantic search, paraphrase mining, text classification, clustering, and more.

Model Details

Model Description

• Model Type: Sentence Transformer
• Base model: AhmedZaky1/DIMI-embedding-v2
• Maximum Sequence Length: 8192 tokens
• Output Dimensionality: 768 dimensions
• Similarity Function: Cosine Similarity
• Training Dataset:
  • silma-arabic-english-sts-dataset-v1.0
• Languages: ar, en

Model Sources

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

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'max_seq_length': 8192, 'do_lower_case': False}) with Transformer model: NewModel 
  (1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': True, 'pooling_mode_mean_tokens': False, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False, '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("AhmedZaky1/DIMI-embedding-v2-silma-sts-matryoshka")
# Run inference
sentences = [
    'ثلاثة رجال يلعبون الشطرنج.',
    'رجلين يلعبان الشطرنج',
    'ar-ar',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 768]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities.shape)
# [3, 3]

Evaluation

Metrics

Semantic Similarity

• Datasets: silma-sts-dev-768, silma-sts-dev-512, silma-sts-dev-256, silma-sts-dev-128, silma-sts-dev-64, sts17-ar-test-768, sts17-ar-test-512, sts17-ar-test-256, sts17-ar-test-128 and sts17-ar-test-64
• Evaluated with EmbeddingSimilarityEvaluator

Metric	silma-sts-dev-768	silma-sts-dev-512	silma-sts-dev-256	silma-sts-dev-128	silma-sts-dev-64	sts17-ar-test-768	sts17-ar-test-512	sts17-ar-test-256	sts17-ar-test-128	sts17-ar-test-64
pearson_cosine	0.8894	0.8959	0.9079	0.9182	0.9066	0.8205	0.8193	0.8191	0.8115	0.7962
spearman_cosine	0.8358	0.8395	0.847	0.8562	0.8434	0.8258	0.8227	0.8246	0.8183	0.8077

Training Details

Training Dataset

silma-arabic-english-sts-dataset-v1.0

• Dataset: silma-arabic-english-sts-dataset-v1.0 at 1885690
• Size: 34,436 training samples
• Columns: sentence1, sentence2, score, and langs
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2	score	langs
  type	string	string	float	string
  details	min: 4 tokens mean: 9.68 tokens max: 26 tokens	min: 4 tokens mean: 9.68 tokens max: 26 tokens	min: 0.0 mean: 0.47 max: 1.0	min: 5 tokens mean: 5.0 tokens max: 5 tokens

• Samples:

  sentence1	sentence2	score	langs
  رجل يعزف على البيانو	امرأة تعزف على الكمان	0.2	ar-ar
  امرأة تعزف على الكمان	رجل يعزف على البيانو	0.2	ar-ar
  امرأة تعزف على الناي.	رجل يعزف على الغيتار	0.2	ar-ar

• Loss: MatryoshkaLoss with these parameters:

  {
      "loss": "CoSENTLoss",
      "matryoshka_dims": [
          768,
          512,
          256,
          128,
          64
      ],
      "matryoshka_weights": [
          1,
          1,
          1,
          1,
          1
      ],
      "n_dims_per_step": -1
  }
  

Evaluation Dataset

silma-arabic-english-sts-dataset-v1.0

• Dataset: silma-arabic-english-sts-dataset-v1.0 at 1885690
• Size: 100 evaluation samples
• Columns: sentence1, sentence2, score, and langs
• Approximate statistics based on the first 100 samples:

  	sentence1	sentence2	score	langs
  type	string	string	float	string
  details	min: 5 tokens mean: 9.49 tokens max: 19 tokens	min: 5 tokens mean: 9.49 tokens max: 19 tokens	min: 0.1 mean: 0.74 max: 1.0	min: 5 tokens mean: 5.0 tokens max: 5 tokens

• Samples:

  sentence1	sentence2	score	langs
  طائرة ستقلع	طائرة طيران تقلع	1.0	ar-ar
  طائرة طيران تقلع	طائرة ستقلع	1.0	ar-ar
  رجل يعزف على ناي كبير	رجل يعزف على الناي	0.76	ar-ar

• Loss: MatryoshkaLoss with these parameters:

  {
      "loss": "CoSENTLoss",
      "matryoshka_dims": [
          768,
          512,
          256,
          128,
          64
      ],
      "matryoshka_weights": [
          1,
          1,
          1,
          1,
          1
      ],
      "n_dims_per_step": -1
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• per_device_train_batch_size: 16
• per_device_eval_batch_size: 16
• num_train_epochs: 4
• warmup_ratio: 0.1
• save_only_model: True
• fp16: True
• load_best_model_at_end: 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: 16
• per_device_eval_batch_size: 16
• 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: 5e-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: 4
• 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: True
• 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}
• tp_size: 0
• 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}
• deepspeed: None
• label_smoothing_factor: 0.0
• optim: adamw_torch
• 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
• 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
• eval_use_gather_object: False
• average_tokens_across_devices: False
• prompts: None
• batch_sampler: batch_sampler
• multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	Training Loss	Validation Loss	silma-sts-dev-768_spearman_cosine	silma-sts-dev-512_spearman_cosine	silma-sts-dev-256_spearman_cosine	silma-sts-dev-128_spearman_cosine	silma-sts-dev-64_spearman_cosine	sts17-ar-test-768_spearman_cosine	sts17-ar-test-512_spearman_cosine	sts17-ar-test-256_spearman_cosine	sts17-ar-test-128_spearman_cosine	sts17-ar-test-64_spearman_cosine
0.0929	100	39.5796	45.0982	0.7199	0.7173	0.7292	0.7433	0.7196	-	-	-	-	-
0.1857	200	31.3305	29.9877	0.7233	0.7248	0.7344	0.7337	0.7192	-	-	-	-	-
0.2786	300	27.7756	31.4644	0.7288	0.7268	0.7331	0.7388	0.7169	-	-	-	-	-
0.3714	400	27.7405	33.3315	0.7172	0.7168	0.7341	0.7349	0.7219	-	-	-	-	-
0.4643	500	27.1884	30.4957	0.7469	0.7428	0.7475	0.7547	0.7426	-	-	-	-	-
0.5571	600	27.0428	29.5877	0.7133	0.7138	0.7380	0.7549	0.7533	-	-	-	-	-
0.6500	700	26.7957	30.3813	0.7520	0.7430	0.7570	0.7604	0.7647	-	-	-	-	-
0.7428	800	26.2667	30.6293	0.7323	0.7333	0.7558	0.7609	0.7479	-	-	-	-	-
0.8357	900	25.9412	29.8621	0.7730	0.7732	0.7913	0.8117	0.7797	-	-	-	-	-
0.9285	1000	25.7816	31.7315	0.7856	0.7918	0.7916	0.8025	0.8048	-	-	-	-	-
1.0214	1100	25.1666	31.6311	0.7651	0.7668	0.7673	0.7826	0.7846	-	-	-	-	-
1.1142	1200	24.7681	32.3005	0.7719	0.7892	0.7941	0.8022	0.7939	-	-	-	-	-
1.2071	1300	24.8771	32.1761	0.7660	0.7744	0.7807	0.7884	0.7841	-	-	-	-	-
1.2999	1400	24.9063	33.2694	0.7646	0.7644	0.7884	0.7906	0.7886	-	-	-	-	-
1.3928	1500	24.7283	32.4350	0.7935	0.7974	0.8071	0.8112	0.8062	-	-	-	-	-
1.4856	1600	24.4217	34.1219	0.7781	0.7754	0.7739	0.7916	0.7889	-	-	-	-	-
1.5785	1700	24.4923	33.1239	0.7636	0.7709	0.7882	0.7991	0.7913	-	-	-	-	-
1.6713	1800	24.0844	33.5233	0.7785	0.7832	0.7880	0.7977	0.8014	-	-	-	-	-
1.7642	1900	24.1453	35.4602	0.7795	0.7816	0.8053	0.8115	0.7944	-	-	-	-	-
1.8570	2000	24.2271	36.2812	0.8003	0.8009	0.8008	0.8102	0.8009	-	-	-	-	-
1.9499	2100	23.7371	37.0276	0.7769	0.7866	0.7918	0.7926	0.7832	-	-	-	-	-
2.0427	2200	23.3566	34.5721	0.7931	0.8017	0.8020	0.8159	0.8027	-	-	-	-	-
2.1356	2300	23.2523	35.5316	0.7931	0.7981	0.7896	0.8157	0.8142	-	-	-	-	-
2.2284	2400	23.0447	36.6811	0.7973	0.7962	0.7935	0.8030	0.8037	-	-	-	-	-
2.3213	2500	22.9782	37.5482	0.8121	0.8185	0.8200	0.8293	0.8244	-	-	-	-	-
2.4141	2600	22.9119	37.2809	0.8077	0.8116	0.8113	0.8333	0.8151	-	-	-	-	-
2.5070	2700	23.1302	37.7993	0.8255	0.8304	0.8310	0.8376	0.8303	-	-	-	-	-
2.5998	2800	22.9941	38.8005	0.8182	0.8214	0.8143	0.8193	0.8155	-	-	-	-	-
2.6927	2900	22.8876	36.2524	0.8201	0.8222	0.8194	0.8347	0.8260	-	-	-	-	-
2.7855	3000	22.5304	38.1523	0.8195	0.8280	0.8356	0.8545	0.8394	-	-	-	-	-
2.8784	3100	22.446	39.4876	0.8242	0.8246	0.8319	0.8483	0.8397	-	-	-	-	-
2.9712	3200	22.5077	39.1910	0.8231	0.8249	0.8334	0.8475	0.8372	-	-	-	-	-
3.0641	3300	21.9675	36.4245	0.8408	0.8425	0.8456	0.8619	0.8577	-	-	-	-	-
3.1569	3400	21.9361	36.7119	0.8344	0.8405	0.8460	0.8656	0.8644	-	-	-	-	-
3.2498	3500	21.7747	37.7140	0.8279	0.8353	0.8414	0.8510	0.8446	-	-	-	-	-
3.3426	3600	21.8649	38.9102	0.8298	0.8331	0.8456	0.8494	0.8447	-	-	-	-	-
3.4355	3700	21.794	37.4385	0.8278	0.8328	0.8377	0.8442	0.8373	-	-	-	-	-
3.5283	3800	21.7968	37.0225	0.8352	0.8501	0.8540	0.8722	0.8553	-	-	-	-	-
3.6212	3900	21.5941	37.5736	0.8344	0.8515	0.8511	0.8643	0.8587	-	-	-	-	-
3.7140	4000	21.8181	37.4984	0.8340	0.8440	0.8470	0.8607	0.8484	-	-	-	-	-
3.8069	4100	21.7035	37.9701	0.8346	0.8394	0.8436	0.8615	0.8479	-	-	-	-	-
3.8997	4200	21.398	38.1567	0.8349	0.8365	0.8470	0.8572	0.8405	-	-	-	-	-
3.9926	4300	21.6518	38.3515	0.8358	0.8395	0.8470	0.8562	0.8434	-	-	-	-	-
4.0	4308	-	-	-	-	-	-	-	0.8258	0.8227	0.8246	0.8183	0.8077

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.12.7
• Sentence Transformers: 3.3.1
• Transformers: 4.51.3
• PyTorch: 2.6.0+cu124
• Accelerate: 1.4.0
• Datasets: 3.3.2
• Tokenizers: 0.21.1

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",
}

MatryoshkaLoss

@misc{kusupati2024matryoshka,
    title={Matryoshka Representation Learning},
    author={Aditya Kusupati and Gantavya Bhatt and Aniket Rege and Matthew Wallingford and Aditya Sinha and Vivek Ramanujan and William Howard-Snyder and Kaifeng Chen and Sham Kakade and Prateek Jain and Ali Farhadi},
    year={2024},
    eprint={2205.13147},
    archivePrefix={arXiv},
    primaryClass={cs.LG}
}

CoSENTLoss

@online{kexuefm-8847,
    title={CoSENT: A more efficient sentence vector scheme than Sentence-BERT},
    author={Su Jianlin},
    year={2022},
    month={Jan},
    url={https://kexue.fm/archives/8847},
}