whitemouse84/LaBSE-en-ru-distilled-each-third-layer

SentenceTransformer based on cointegrated/LaBSE-en-ru

This is a sentence-transformers model finetuned from cointegrated/LaBSE-en-ru. 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: cointegrated/LaBSE-en-ru
• Maximum Sequence Length: 512 tokens
• Output Dimensionality: 768 tokens
• Similarity Function: Cosine Similarity

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: BertModel 
  (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): Dense({'in_features': 768, 'out_features': 768, 'bias': True, 'activation_function': 'torch.nn.modules.activation.Tanh'})
  (3): 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("whitemouse84/LaBSE-en-ru-distilled-each-third-layer")
# Run inference
sentences = [
    'See Name section.',
    'Ms. Packard is the voice of the female blood elf in the video game World of Warcraft.',
    'Yeah, people who might not be hungry.',
]
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

• Dataset: sts-dev
• Evaluated with EmbeddingSimilarityEvaluator

Metric	Value
pearson_cosine	0.5305
spearman_cosine	0.6347
pearson_manhattan	0.5553
spearman_manhattan	0.6389
pearson_euclidean	0.55
spearman_euclidean	0.6347
pearson_dot	0.5305
spearman_dot	0.6347
pearson_max	0.5553
spearman_max	0.6389

Knowledge Distillation

• Evaluated with MSEEvaluator

Metric	Value
negative_mse	-0.0063

Semantic Similarity

• Dataset: sts-test
• Evaluated with EmbeddingSimilarityEvaluator

Metric	Value
pearson_cosine	0.5043
spearman_cosine	0.5986
pearson_manhattan	0.5227
spearman_manhattan	0.5984
pearson_euclidean	0.5227
spearman_euclidean	0.5986
pearson_dot	0.5043
spearman_dot	0.5986
pearson_max	0.5227
spearman_max	0.5986

Training Details

Training Dataset

Unnamed Dataset

• Size: 10,975,066 training samples
• Columns: sentence and label
• Approximate statistics based on the first 1000 samples:

  	sentence	label
  type	string	list
  details	min: 6 tokens mean: 26.93 tokens max: 139 tokens	size: 768 elements

• Samples:

  sentence	label
  It is based on the Java Persistence API (JPA), but it does not strictly follow the JSR 338 Specification, as it implements different design patterns and technologies.	[-0.012331949546933174, -0.04570527374744415, -0.024963658303022385, -0.03620213270187378, 0.022556383162736893, ...]
  Покупаем вторичное сырье в Каунасе (Переработка вторичного сырья) - Алфенас АНД КО, ЗАО на Bizorg.	[-0.07498518377542496, -0.01913534104824066, -0.01797042042016983, 0.048263177275657654, -0.00016611881437711418, ...]
  At the Equal Justice Conference ( EJC ) held in March 2001 in San Diego , LSC and the Project for the Future of Equal Justice held the second Case Management Software pre-conference .	[0.03870972990989685, -0.0638347640633583, -0.01696585863828659, -0.043612319976091385, -0.048241738229990005, ...]

• Loss: MSELoss

Evaluation Dataset

Unnamed Dataset

• Size: 10,000 evaluation samples
• Columns: sentence and label
• Approximate statistics based on the first 1000 samples:

  	sentence	label
  type	string	list
  details	min: 5 tokens mean: 24.18 tokens max: 111 tokens	size: 768 elements

• Samples:

  sentence	label
  The Canadian Canoe Museum is a museum dedicated to canoes located in Peterborough, Ontario, Canada.	[-0.05444105342030525, -0.03650881350040436, -0.041163671761751175, -0.010616903193295002, -0.04094529151916504, ...]
  И мне нравилось, что я одновременно зарабатываю и смотрю бои».	[-0.03404555842280388, 0.028203096240758896, -0.056121889501810074, -0.0591997392475605, -0.05523117259144783, ...]
  Ну, а на следующий день, разумеется, Президент Кеннеди объявил блокаду Кубы, и наши корабли остановили у кубинских берегов направлявшийся на Кубу российский корабль, и у него на борту нашли ракеты.	[-0.008193841204047203, 0.00694894278421998, -0.03027420863509178, -0.03290146216750145, 0.01425305474549532, ...]

• Loss: MSELoss

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• per_device_train_batch_size: 64
• per_device_eval_batch_size: 64
• learning_rate: 0.0001
• num_train_epochs: 1
• warmup_ratio: 0.1
• 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: 64
• per_device_eval_batch_size: 64
• 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: 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: {}
• 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: 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}
• 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: False
• hub_always_push: False
• gradient_checkpointing: False
• gradient_checkpointing_kwargs: None
• include_inputs_for_metrics: False
• 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
• dispatch_batches: None
• split_batches: 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
• eval_use_gather_object: False
• batch_sampler: batch_sampler
• multi_dataset_batch_sampler: proportional

Training Logs

Click to expand

Epoch	Step	Training Loss	loss	negative_mse	sts-dev_spearman_cosine	sts-test_spearman_cosine
0	0	-	-	-0.2381	0.4206	-
0.0058	1000	0.0014	-	-	-	-
0.0117	2000	0.0009	-	-	-	-
0.0175	3000	0.0007	-	-	-	-
0.0233	4000	0.0006	-	-	-	-
0.0292	5000	0.0005	0.0004	-0.0363	0.6393	-
0.0350	6000	0.0004	-	-	-	-
0.0408	7000	0.0004	-	-	-	-
0.0467	8000	0.0003	-	-	-	-
0.0525	9000	0.0003	-	-	-	-
0.0583	10000	0.0003	0.0002	-0.0207	0.6350	-
0.0641	11000	0.0003	-	-	-	-
0.0700	12000	0.0003	-	-	-	-
0.0758	13000	0.0002	-	-	-	-
0.0816	14000	0.0002	-	-	-	-
0.0875	15000	0.0002	0.0002	-0.0157	0.6328	-
0.0933	16000	0.0002	-	-	-	-
0.0991	17000	0.0002	-	-	-	-
0.1050	18000	0.0002	-	-	-	-
0.1108	19000	0.0002	-	-	-	-
0.1166	20000	0.0002	0.0001	-0.0132	0.6317	-
0.1225	21000	0.0002	-	-	-	-
0.1283	22000	0.0002	-	-	-	-
0.1341	23000	0.0002	-	-	-	-
0.1400	24000	0.0002	-	-	-	-
0.1458	25000	0.0002	0.0001	-0.0118	0.6251	-
0.1516	26000	0.0002	-	-	-	-
0.1574	27000	0.0002	-	-	-	-
0.1633	28000	0.0002	-	-	-	-
0.1691	29000	0.0002	-	-	-	-
0.1749	30000	0.0002	0.0001	-0.0109	0.6304	-
0.1808	31000	0.0002	-	-	-	-
0.1866	32000	0.0002	-	-	-	-
0.1924	33000	0.0002	-	-	-	-
0.1983	34000	0.0001	-	-	-	-
0.2041	35000	0.0001	0.0001	-0.0102	0.6280	-
0.2099	36000	0.0001	-	-	-	-
0.2158	37000	0.0001	-	-	-	-
0.2216	38000	0.0001	-	-	-	-
0.2274	39000	0.0001	-	-	-	-
0.2333	40000	0.0001	0.0001	-0.0098	0.6272	-
0.2391	41000	0.0001	-	-	-	-
0.2449	42000	0.0001	-	-	-	-
0.2507	43000	0.0001	-	-	-	-
0.2566	44000	0.0001	-	-	-	-
0.2624	45000	0.0001	0.0001	-0.0093	0.6378	-
0.2682	46000	0.0001	-	-	-	-
0.2741	47000	0.0001	-	-	-	-
0.2799	48000	0.0001	-	-	-	-
0.2857	49000	0.0001	-	-	-	-
0.2916	50000	0.0001	0.0001	-0.0089	0.6325	-
0.2974	51000	0.0001	-	-	-	-
0.3032	52000	0.0001	-	-	-	-
0.3091	53000	0.0001	-	-	-	-
0.3149	54000	0.0001	-	-	-	-
0.3207	55000	0.0001	0.0001	-0.0087	0.6328	-
0.3266	56000	0.0001	-	-	-	-
0.3324	57000	0.0001	-	-	-	-
0.3382	58000	0.0001	-	-	-	-
0.3441	59000	0.0001	-	-	-	-
0.3499	60000	0.0001	0.0001	-0.0085	0.6357	-
0.3557	61000	0.0001	-	-	-	-
0.3615	62000	0.0001	-	-	-	-
0.3674	63000	0.0001	-	-	-	-
0.3732	64000	0.0001	-	-	-	-
0.3790	65000	0.0001	0.0001	-0.0083	0.6366	-
0.3849	66000	0.0001	-	-	-	-
0.3907	67000	0.0001	-	-	-	-
0.3965	68000	0.0001	-	-	-	-
0.4024	69000	0.0001	-	-	-	-
0.4082	70000	0.0001	0.0001	-0.0080	0.6325	-
0.4140	71000	0.0001	-	-	-	-
0.4199	72000	0.0001	-	-	-	-
0.4257	73000	0.0001	-	-	-	-
0.4315	74000	0.0001	-	-	-	-
0.4374	75000	0.0001	0.0001	-0.0078	0.6351	-
0.4432	76000	0.0001	-	-	-	-
0.4490	77000	0.0001	-	-	-	-
0.4548	78000	0.0001	-	-	-	-
0.4607	79000	0.0001	-	-	-	-
0.4665	80000	0.0001	0.0001	-0.0077	0.6323	-
0.4723	81000	0.0001	-	-	-	-
0.4782	82000	0.0001	-	-	-	-
0.4840	83000	0.0001	-	-	-	-
0.4898	84000	0.0001	-	-	-	-
0.4957	85000	0.0001	0.0001	-0.0076	0.6316	-
0.5015	86000	0.0001	-	-	-	-
0.5073	87000	0.0001	-	-	-	-
0.5132	88000	0.0001	-	-	-	-
0.5190	89000	0.0001	-	-	-	-
0.5248	90000	0.0001	0.0001	-0.0074	0.6306	-
0.5307	91000	0.0001	-	-	-	-
0.5365	92000	0.0001	-	-	-	-
0.5423	93000	0.0001	-	-	-	-
0.5481	94000	0.0001	-	-	-	-
0.5540	95000	0.0001	0.0001	-0.0073	0.6305	-
0.5598	96000	0.0001	-	-	-	-
0.5656	97000	0.0001	-	-	-	-
0.5715	98000	0.0001	-	-	-	-
0.5773	99000	0.0001	-	-	-	-
0.5831	100000	0.0001	0.0001	-0.0072	0.6333	-
0.5890	101000	0.0001	-	-	-	-
0.5948	102000	0.0001	-	-	-	-
0.6006	103000	0.0001	-	-	-	-
0.6065	104000	0.0001	-	-	-	-
0.6123	105000	0.0001	0.0001	-0.0071	0.6351	-
0.6181	106000	0.0001	-	-	-	-
0.6240	107000	0.0001	-	-	-	-
0.6298	108000	0.0001	-	-	-	-
0.6356	109000	0.0001	-	-	-	-
0.6415	110000	0.0001	0.0001	-0.0070	0.6330	-
0.6473	111000	0.0001	-	-	-	-
0.6531	112000	0.0001	-	-	-	-
0.6589	113000	0.0001	-	-	-	-
0.6648	114000	0.0001	-	-	-	-
0.6706	115000	0.0001	0.0001	-0.0070	0.6336	-
0.6764	116000	0.0001	-	-	-	-
0.6823	117000	0.0001	-	-	-	-
0.6881	118000	0.0001	-	-	-	-
0.6939	119000	0.0001	-	-	-	-
0.6998	120000	0.0001	0.0001	-0.0069	0.6305	-
0.7056	121000	0.0001	-	-	-	-
0.7114	122000	0.0001	-	-	-	-
0.7173	123000	0.0001	-	-	-	-
0.7231	124000	0.0001	-	-	-	-
0.7289	125000	0.0001	0.0001	-0.0068	0.6362	-
0.7348	126000	0.0001	-	-	-	-
0.7406	127000	0.0001	-	-	-	-
0.7464	128000	0.0001	-	-	-	-
0.7522	129000	0.0001	-	-	-	-
0.7581	130000	0.0001	0.0001	-0.0067	0.6340	-
0.7639	131000	0.0001	-	-	-	-
0.7697	132000	0.0001	-	-	-	-
0.7756	133000	0.0001	-	-	-	-
0.7814	134000	0.0001	-	-	-	-
0.7872	135000	0.0001	0.0001	-0.0067	0.6365	-
0.7931	136000	0.0001	-	-	-	-
0.7989	137000	0.0001	-	-	-	-
0.8047	138000	0.0001	-	-	-	-
0.8106	139000	0.0001	-	-	-	-
0.8164	140000	0.0001	0.0001	-0.0066	0.6339	-
0.8222	141000	0.0001	-	-	-	-
0.8281	142000	0.0001	-	-	-	-
0.8339	143000	0.0001	-	-	-	-
0.8397	144000	0.0001	-	-	-	-
0.8456	145000	0.0001	0.0001	-0.0066	0.6352	-
0.8514	146000	0.0001	-	-	-	-
0.8572	147000	0.0001	-	-	-	-
0.8630	148000	0.0001	-	-	-	-
0.8689	149000	0.0001	-	-	-	-
0.8747	150000	0.0001	0.0001	-0.0065	0.6357	-
0.8805	151000	0.0001	-	-	-	-
0.8864	152000	0.0001	-	-	-	-
0.8922	153000	0.0001	-	-	-	-
0.8980	154000	0.0001	-	-	-	-
0.9039	155000	0.0001	0.0001	-0.0065	0.6336	-
0.9097	156000	0.0001	-	-	-	-
0.9155	157000	0.0001	-	-	-	-
0.9214	158000	0.0001	-	-	-	-
0.9272	159000	0.0001	-	-	-	-
0.9330	160000	0.0001	0.0001	-0.0064	0.6334	-
0.9389	161000	0.0001	-	-	-	-
0.9447	162000	0.0001	-	-	-	-
0.9505	163000	0.0001	-	-	-	-
0.9563	164000	0.0001	-	-	-	-
0.9622	165000	0.0001	0.0001	-0.0064	0.6337	-
0.9680	166000	0.0001	-	-	-	-
0.9738	167000	0.0001	-	-	-	-
0.9797	168000	0.0001	-	-	-	-
0.9855	169000	0.0001	-	-	-	-
0.9913	170000	0.0001	0.0001	-0.0063	0.6347	-
0.9972	171000	0.0001	-	-	-	-
1.0	171486	-	-	-	-	0.5986

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.10.14
• Sentence Transformers: 3.0.1
• Transformers: 4.44.0
• PyTorch: 2.4.0
• Accelerate: 0.33.0
• Datasets: 2.20.0
• Tokenizers: 0.19.1

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",
}

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",
}