alakxender/e5-dhivehi-qa-mnr

SentenceTransformer

This is a sentence-transformers model trained. 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
• Maximum Sequence Length: 256 tokens
• Output Dimensionality: 768 dimensions
• Similarity Function: Cosine Similarity

Fine-tuned Models

This model is part of a progressive series of sentence embedding models based on intfloat/multilingual-e5-base, fine-tuned specifically for Dhivehi language understanding.

Each stage leverages a targeted dataset to specialize the model for semantic similarity, question answering, and summarization tasks — improving performance for real-world Dhivehi NLP applications.

Stage	Task	Model	Dataset	Objective
0	Base	Multilingual Base	intfloat/multilingual-e5-base	—
1	Paraphrase Identification (MNR)	alakxender/e5-dhivehi-paws-mnr	alakxender/dhivehi-paws-labeled > label=1 Only	MultipleNegativesRankingLoss
2	Paraphrase Identification (Cosine)	alakxender/e5-dhivehi-paws-cos	alakxender/dhivehi-paws-labeled	CosineSimilarityLoss
3	Question → Passage Matching	alakxender/e5-dhivehi-qa-mnr	alakxender/dhivehi-qa-dataset	MultipleNegativesRankingLoss
4	News Title → Content	alakxender/e5-dhivehi-articles-mnr	alakxender/dhivehi-news-corpus	MultipleNegativesRankingLoss
5	Summary → Content	alakxender/e5-dhivehi-summaries-mnr	alakxender/dv-en-parallel-corpus-clean, alakxender/dv-summary-translation-corpus	MultipleNegativesRankingLoss

Each model builds upon the previous checkpoint, incrementally enhancing the semantic capabilities of the model for Dhivehi. The goal is to support high-quality sentence embeddings for a wide range of Dhivehi information retrieval and understanding tasks.

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': 256, 'do_lower_case': False, 'architecture': 'XLMRobertaModel'})
  (1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False, '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("alakxender/e5-dhivehi-qa-mnr")
# Run inference
sentences = [
    'query: އަކްރަމުގެ ދިރިއުޅުމަށް އައި ބަދަލުތަކަކީ ކޮބައިތޯ؟',
    'passage: މިގޮތުގެމަތިން އަކްރަމަށް ލިބުނު ޖަވާހިރުތައް ވިއްކައި، ލިބުނު ގިނަގުނަ ފައިސާއިން ޖަވާހިރުގެ ވިޔަފާރިފަށައި، ބައްޕައެކޭ އެއްފަދައިން މަޝްހޫރު ވިޔަފާރިވެރިއަކަށް ވެއްޖެއެވެ. އަދި އަކްރަމާއި ޢާއިލާގެ ދިރިއުޅުން ކުރިއެކޭވެސް އެއްފަދަ ތަނަވަސް ދިރިއުޅުމަކަށް ބަދަލުވެގެން ހިނގައްޖެއެވެ.',
    'passage: ސިވިލް ސަރވިސްގެ މުވައްޒަފުން މަޢުލޫމާތު ހާމަކުރުމުގައި ޢަމަލުކުރަންވާނީ ތިރީގައިވާ އުޞޫލުތަކުގެ މަތިންނެވެ. ޤާނޫނާ ގަވާއިދާ އެއްގޮތްވާގޮތުގެމަތިން، މަޢުލޫމާތު ދޭންޖެހޭ ކޮންމެ ޙާލަތެއްގައި ތެދު މަޢުލޫމާތު ދިނުން.',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 768]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities)
# tensor([[1.0000, 0.4978, 0.4952],
#         [0.4978, 1.0000, 0.8843],
#         [0.4952, 0.8843, 1.0000]])

Training Details

Training Dataset

• Size: 9,232 training samples
• Columns: sentence_0 and sentence_1
• Approximate statistics based on the first 1000 samples:

  	sentence_0	sentence_1
  type	string	string
  details	min: 9 tokens mean: 17.88 tokens max: 43 tokens	min: 9 tokens mean: 62.22 tokens max: 256 tokens

• Samples:

  sentence_0	sentence_1
  query: ދީނީ ތަގްރީރުކުރުމާއި ދީނީ ނަސޭހަތް ދިނުން ކަމަށް ބުނެފައިވަނީ ކޮން ކަމަކަށް؟	passage: "ދީނީ ތަގްރީރުކުރުމާއި ދީނީ ނަސޭހަތް ދިނުން" ކަމަށް ބުނެފައި އެވަނީ، ހަބަރުފަތުރާ ވަސީލަތަކުން ނުވަތަ ހާންމު މީހުން ޖަމާވެފައިވާ ތަނެއްގައި ދީނީ ތަގްރީރުކުރުމާއި ދީނީ ނަސޭހަތް ދިނުމަށެވެ. ދީނީ އެއްބައިވަންތަކަން ހިމާޔަތް ކުރުމުގެ ގާނޫނު (ގާނޫނު ނަންބަރު 6/1994) އަށް 1 ވަނަ އިސްލާހު ގެނައުމުގެ ގާނޫނު (ގާނޫނު ނަންބަރު 8/2014) ގެ 7 ވަނަ މާއްދާގެ (ނ) ގައި
  query: ގާނޫނީ ގޮތުން ގައިދީން ކަމަށް ބުނެފައިވަނީ ކޮން ބައެއް؟	passage: "ގައިދީ" ނުވަތަ "ގައިދީން" ކަމަށް ބުނެފައިއެވަނީ، ޖަލުގައި ތިބޭ މީހުންގެ ތެރެއިން ހުކުމެއް ތަންފީޒުކުރަމުންދާ މީހުންނާއި އަރުވާލަން ނުވަތަ ގޭބަންދަށް ކޯޓުން ހުކުމްކޮށްފައި ތިބޭ މީހުންނަށެވެ. ޖަލުތަކާއި ޕެރޯލްގެ ގާނޫނު (ގާނޫނު ނަންބަރު 14/2013) ގެ 155 ވަނަ މާއްދާގެ (ލ) ގައި
  query: އިސްލާމްދީނުގައި ޢިލްމު އުނގެނުމަކީ ކޮބައިތޯ؟	passage: ކީރިތި ރަސޫލާ އަންގަވައިފައިވަނީ، [طلب العلم فريضة على كلّ مسلم ومسلمة] [ޢިލްމު އުނގެނުމަކީ ކޮންމެ މުސްލިމް ފިރިހެނަކާއި ކޮންމެ މުސްލިމް އަންހެނެއްގެ މައްޗަށް އޮތް ވާޖިބެކޭ، ފަރުޟެކޭ.]

• Loss: MultipleNegativesRankingLoss with these parameters:

  {
      "scale": 20.0,
      "similarity_fct": "cos_sim"
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• per_device_train_batch_size: 4
• per_device_eval_batch_size: 4
• multi_dataset_batch_sampler: round_robin

All Hyperparameters

Click to expand

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: no
• prediction_loss_only: True
• per_device_train_batch_size: 4
• per_device_eval_batch_size: 4
• 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
• num_train_epochs: 3
• max_steps: -1
• lr_scheduler_type: linear
• lr_scheduler_kwargs: {}
• warmup_ratio: 0.0
• 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: 42
• data_seed: None
• jit_mode_eval: False
• use_ipex: False
• bf16: False
• 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: 0
• dataloader_prefetch_factor: None
• past_index: -1
• disable_tqdm: False
• remove_unused_columns: True
• label_names: None
• load_best_model_at_end: False
• 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}
• 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: round_robin
• router_mapping: {}
• learning_rate_mapping: {}

Training Logs

Epoch	Step	Training Loss
0.2166	500	1.2589
0.4333	1000	1.0569
0.6499	1500	1.0562
0.8666	2000	0.9848
1.0832	2500	0.9565
1.2998	3000	0.9604
1.5165	3500	0.958
1.7331	4000	0.9232
1.9497	4500	0.9202
2.1664	5000	0.8813
2.3830	5500	0.893
2.5997	6000	0.8324
2.8163	6500	0.7792

Framework Versions

• Python: 3.9.21
• Sentence Transformers: 5.0.0
• Transformers: 4.52.4
• PyTorch: 2.5.1+cu124
• Accelerate: 1.3.0
• Datasets: 3.2.0
• Tokenizers: 0.21.0

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",
}

MultipleNegativesRankingLoss

@misc{henderson2017efficient,
    title={Efficient Natural Language Response Suggestion for Smart Reply},
    author={Matthew Henderson and Rami Al-Rfou and Brian Strope and Yun-hsuan Sung and Laszlo Lukacs and Ruiqi Guo and Sanjiv Kumar and Balint Miklos and Ray Kurzweil},
    year={2017},
    eprint={1705.00652},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}