panalexeu/xlm-roberta-ua-distilled

SentenceTransformer based on FacebookAI/xlm-roberta-base

This is a sentence-transformers model finetuned from FacebookAI/xlm-roberta-base. 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.

👉 Check out the model on GitHub.

Model Details

Model Description

• Model Type: Sentence Transformer
• Base model: FacebookAI/xlm-roberta-base
• Maximum Sequence Length: 512 tokens
• Output Dimensionality: 768 dimensions
• Similarity Function: Cosine Similarity
• Training Dataset: parallel-sentences-talks, parallel-sentences-wikimatrix, parallel-sentences-tatoeba
• Language: Ukrainian, English
• License: MIT

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': 512, 'do_lower_case': False}) with Transformer model: 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("panalexeu/xlm-roberta-ua-distilled")
# Run inference
sentences = [
    "You'd better consult the doctor.",
    'Краще проконсультуйся у лікаря.',
    'Їх позначають як Aufklärungsfahrzeug 93 та Aufklärungsfahrzeug 97 відповідно.',
]
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

Knowledge Distillation

• Dataset: mse-en-ua
• Evaluated with MSEEvaluator

Metric	Value
negative_mse	-1.1089

Semantic Similarity

• Datasets: sts17-en-en, sts17-en-ua and sts17-ua-ua
• Evaluated with EmbeddingSimilarityEvaluator

Metric	sts17-en-en	sts17-en-ua	sts17-ua-ua
pearson_cosine	0.6785	0.5926	0.6159
spearman_cosine	0.7308	0.6198	0.6446

Training Details

Training Dataset

• Dataset: parallel-sentences-talks, parallel-sentences-wikimatrix, parallel-sentences-tatoeba
• Size: 523,982 training samples
• Columns: english, non_english, and label
• Approximate statistics based on the first 1000 samples:

  	english	non_english	label
  type	string	string	list
  details	min: 5 tokens mean: 21.11 tokens max: 254 tokens	min: 4 tokens mean: 23.15 tokens max: 293 tokens	size: 768 elements

• Samples:

  english	non_english	label
  Her real name is Lydia (リディア, Ridia), but she was mistaken for a boy and called Ricard.	Справжнє ім'я — Лідія, але її помилково сприйняли за хлопчика і назвали Рікард.	[0.15217968821525574, -0.17830222845077515, -0.12677159905433655, 0.22082313895225525, 0.40085524320602417, ...]
  (Applause) So he didn't just learn water.	(Аплодисменти) Він не тільки вивчив слово "вода".	[-0.1058148592710495, -0.08846072107553482, -0.2684604823589325, -0.105219267308712, 0.3050258755683899, ...]
  It is tightly integrated with SAM, the Storage and Archive Manager, and hence is often referred to as SAM-QFS.	Вона тісно інтегрована з SAM (Storage and Archive Manager), тому часто називається SAM-QFS.	[0.03270340710878372, -0.45798248052597046, -0.20090211927890778, 0.006579531356692314, -0.03178019821643829, ...]

• Loss: MSELoss

Evaluation Dataset

• Dataset: parallel-sentences-talks, parallel-sentences-wikimatrix, parallel-sentences-tatoeba
• Size: 3,838 evaluation samples
• Columns: english, non_english, and label
• Approximate statistics based on the first 1000 samples:

  	english	non_english	label
  type	string	string	list
  details	min: 5 tokens mean: 15.64 tokens max: 143 tokens	min: 5 tokens mean: 16.98 tokens max: 148 tokens	size: 768 elements

• Samples:

  english	non_english	label
  I have lost my wallet.	Я загубив гаманець.	[-0.11186987161636353, -0.03419225662946701, -0.31304317712783813, 0.0838347002863884, 0.108644500374794, ...]
  It's a pharmaceutical product.	Це фармацевтичний продукт.	[0.04133488982915878, -0.4182000756263733, -0.30786487460136414, -0.09351564198732376, -0.023946482688188553, ...]
  We've all heard of the Casual Friday thing.	Всі ми чули про «джинсову п’ятницю» (вільна форма одягу).	[-0.10697802156209946, 0.21002227067947388, -0.2513434886932373, -0.3718843460083008, 0.06871984899044037, ...]

• Loss: MSELoss

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• per_device_train_batch_size: 16
• per_device_eval_batch_size: 16
• gradient_accumulation_steps: 3
• num_train_epochs: 4
• warmup_ratio: 0.1

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: 3
• 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: 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}
• 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	mse-en-ua_negative_mse	sts17-en-en_spearman_cosine	sts17-en-ua_spearman_cosine	sts17-ua-ua_spearman_cosine
0.0938	1024	0.3281	0.0297	-2.9592	0.2325	0.1547	0.2265
0.1876	2048	0.1136	0.2042	-21.6693	0.0553	0.0429	0.2442
0.2814	3072	0.1008	0.0273	-2.7461	0.2666	0.0758	0.2613
0.3752	4096	0.0843	0.0243	-2.4623	0.2541	0.0012	0.3680
0.4690	5120	0.0756	0.0216	-2.2095	0.3933	0.2535	0.4342
0.5628	6144	0.0661	0.0187	-1.9539	0.5739	0.4222	0.5056
0.6566	7168	0.0579	0.0164	-1.7513	0.6184	0.4897	0.5826
0.7504	8192	0.0526	0.0153	-1.6546	0.6219	0.4568	0.5842
0.8442	9216	0.0488	0.0142	-1.5525	0.6160	0.5012	0.5884
0.9380	10240	0.046	0.0135	-1.4957	0.6361	0.5046	0.5969
1.0318	11264	0.0437	0.0130	-1.4506	0.6453	0.5093	0.5939
1.1256	12288	0.0419	0.0125	-1.4049	0.6403	0.5054	0.6020
1.2194	13312	0.0404	0.0122	-1.3794	0.6654	0.5442	0.6182
1.3132	14336	0.0394	0.0118	-1.3434	0.6800	0.5790	0.6291
1.4070	15360	0.0383	0.0115	-1.3184	0.6836	0.5805	0.6301
1.5008	16384	0.0375	0.0114	-1.3067	0.6742	0.5555	0.6055
1.5946	17408	0.0368	0.0111	-1.2864	0.6909	0.5765	0.6256
1.6884	18432	0.036	0.0109	-1.2633	0.6875	0.5801	0.6178
1.7822	19456	0.0353	0.0107	-1.2490	0.7060	0.5959	0.6322
1.8760	20480	0.035	0.0106	-1.2357	0.7127	0.6047	0.6389
1.9698	21504	0.0344	0.0105	-1.2265	0.7265	0.6233	0.6459
2.0636	22528	0.0335	0.0103	-1.2108	0.7184	0.6151	0.6438
2.1574	23552	0.0327	0.0103	-1.2101	0.7122	0.6074	0.6427
2.2512	24576	0.0324	0.0102	-1.1972	0.7232	0.6174	0.6447
2.3450	25600	0.0322	0.0100	-1.1813	0.7217	0.6166	0.6457
2.4388	26624	0.032	0.0099	-1.1745	0.7308	0.6272	0.6534
2.5326	27648	0.0316	0.0098	-1.1673	0.7289	0.6125	0.6441
2.6264	28672	0.0314	0.0098	-1.1622	0.7222	0.6105	0.6365
2.7202	29696	0.0312	0.0097	-1.1593	0.7175	0.6121	0.6348
2.8140	30720	0.0308	0.0096	-1.1457	0.7204	0.6044	0.6377
2.9078	31744	0.0307	0.0095	-1.1411	0.7230	0.6175	0.6353
3.0016	32768	0.0305	0.0095	-1.1414	0.7130	0.6052	0.6340
3.0954	33792	0.0296	0.0095	-1.1360	0.7234	0.6160	0.6411
3.1892	34816	0.0295	0.0094	-1.1317	0.7220	0.6131	0.6396
3.2830	35840	0.0294	0.0094	-1.1306	0.7315	0.6167	0.6505
3.3768	36864	0.0293	0.0094	-1.1263	0.7219	0.6089	0.6450
3.4706	37888	0.0292	0.0093	-1.1225	0.7236	0.6141	0.6451
3.5644	38912	0.0291	0.0093	-1.1204	0.7331	0.6179	0.6460
3.6582	39936	0.029	0.0092	-1.1147	0.7226	0.6127	0.6406
3.7520	40960	0.029	0.0092	-1.1118	0.7245	0.6184	0.6425
3.8458	41984	0.0289	0.0092	-1.1102	0.7279	0.6179	0.6465
3.9396	43008	0.0288	0.0092	-1.1099	0.7298	0.6191	0.6438
3.9997	43664	-	0.0092	-1.1089	0.7308	0.6198	0.6446

Framework Versions

• Python: 3.11.11
• Sentence Transformers: 3.4.1
• Transformers: 4.51.1
• PyTorch: 2.5.1+cu124
• Accelerate: 1.3.0
• Datasets: 3.5.0
• 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",
}

MSELoss

@inproceedings{reimers-2020-multilingual-sentence-bert,
    title = "Making Monolingual Sentence Embeddings Multilingual using Knowledge Distillation",
    author = "Reimers, Nils and Gurevych, Iryna",
    booktitle = "Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing",
    month = "11",
    year = "2020",
    publisher = "Association for Computational Linguistics",
    url = "https://arxiv.org/abs/2004.09813",
}