sts_v11 / README.md

seregadgl

Add new SentenceTransformer model

34d7252 verified 7 months ago

preview code

raw

history blame contribute delete

22.6 kB

metadata

tags:
  - sentence-transformers
  - sentence-similarity
  - feature-extraction
  - generated_from_trainer
  - dataset_size:111476
  - loss:CosineSimilarityLoss
base_model: sergeyzh/LaBSE-ru-sts
widget:
  - source_sentence: 'трюковый самокат plank 180 белый '
    sentences:
      - смарт-телевизор 75 sony kd-75x950h
      - самокат для трюков плэнк 1.80 м белый
      - xiaomi mi 11 8gb 128gb
  - source_sentence: 'вейп vaporesso xros '
    sentences:
      - садовая ограда классика 4 2 м белый
      - кухонные весы
      - электронная сигарета voopoo drag
  - source_sentence: серьги l atelier precieux 1628710
    sentences:
      - фильтр hepa для пылесоса варис st400
      - потолочная люстра майтон nostalgia ceiling chandelier mod048pl-06g
      - серьги atelier de bijoux 1628712
  - source_sentence: 'мобильный геймпад триггерами x2 '
    sentences:
      - электроскутер nitro pro milano 750w led
      - наушники без проводов мейзу ep52 lite
      - геймпад с функцией триггеров x2
  - source_sentence: комод 7 рисунком машинки 4 ящика
    sentences:
      - удлинитель far f 505 d lara выключателем 2 0м
      - беззеркальный фотоаппарат nikon z50 kit 16-50mm ilce-7cl красный
      - комод 8 с изображением супергероев 6 ящиков
datasets:
  - seregadgl/data_cross_gpt_139k
pipeline_tag: sentence-similarity
library_name: sentence-transformers
metrics:
  - cosine_accuracy
  - cosine_accuracy_threshold
  - cosine_f1
  - cosine_f1_threshold
  - cosine_precision
  - cosine_recall
  - cosine_ap
  - cosine_mcc
model-index:
  - name: SentenceTransformer based on sergeyzh/LaBSE-ru-sts
    results:
      - task:
          type: binary-classification
          name: Binary Classification
        dataset:
          name: eval
          type: eval
        metrics:
          - type: cosine_accuracy
            value: 0.9722640832436311
            name: Cosine Accuracy
          - type: cosine_accuracy_threshold
            value: 0.630459189414978
            name: Cosine Accuracy Threshold
          - type: cosine_f1
            value: 0.9724366041896361
            name: Cosine F1
          - type: cosine_f1_threshold
            value: 0.5821653008460999
            name: Cosine F1 Threshold
          - type: cosine_precision
            value: 0.9647847565278758
            name: Cosine Precision
          - type: cosine_recall
            value: 0.9802107980210798
            name: Cosine Recall
          - type: cosine_ap
            value: 0.9945729266353226
            name: Cosine Ap
          - type: cosine_mcc
            value: 0.9445047865635516
            name: Cosine Mcc

SentenceTransformer based on sergeyzh/LaBSE-ru-sts

This is a sentence-transformers model finetuned from sergeyzh/LaBSE-ru-sts on the data_cross_gpt_139k 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: sergeyzh/LaBSE-ru-sts
Maximum Sequence Length: 512 tokens
Output Dimensionality: 768 dimensions
Similarity Function: Cosine Similarity
Training Dataset:
- data_cross_gpt_139k

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): 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("seregadgl/sts_v11")
# Run inference
sentences = [
    'комод 7 рисунком машинки 4 ящика',
    'комод 8 с изображением супергероев 6 ящиков',
    'беззеркальный фотоаппарат nikon z50 kit 16-50mm ilce-7cl красный',
]
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

Binary Classification

Dataset: eval
Evaluated with BinaryClassificationEvaluator

Metric	Value
cosine_accuracy	0.9723
cosine_accuracy_threshold	0.6305
cosine_f1	0.9724
cosine_f1_threshold	0.5822
cosine_precision	0.9648
cosine_recall	0.9802
cosine_ap	0.9946
cosine_mcc	0.9445

Training Details

Training Dataset

data_cross_gpt_139k

Dataset: data_cross_gpt_139k at 9e1f5ca
Size: 111,476 training samples
Columns: sentence1, sentence2, and label
Approximate statistics based on the first 1000 samples:
sentence1 sentence2 label
type string string float
details
min: 3 tokens
mean: 14.84 tokens
max: 45 tokens

min: 4 tokens
mean: 15.64 tokens
max: 55 tokens

min: 0.0
mean: 0.47
max: 1.0

	sentence1	sentence2	label
type	string	string	float
details	min: 3 tokens mean: 14.84 tokens max: 45 tokens	min: 4 tokens mean: 15.64 tokens max: 55 tokens	min: 0.0 mean: 0.47 max: 1.0

Samples:

sentence1	sentence2	label
`нож кухонный 21см синий`	`кухонный нож 22см зелёный`	`0.0`
`блок питания универсальный для мерцающих флэш гирлянд rich led бахрома занавес нить белый`	`адаптер питания для мигающих led гирлянд "luminous decor" бахрома занавес нить зелёный`	`0.0`
`защитная пленка для apple iphone 6 прозрачная`	`protective film for apple iphone 6 transparent`	`1.0`

Loss: CosineSimilarityLoss with these parameters:

{
    "loss_fct": "torch.nn.modules.loss.MSELoss"
}

Evaluation Dataset

data_cross_gpt_139k

Dataset: data_cross_gpt_139k at 9e1f5ca
Size: 27,870 evaluation samples
Columns: sentence1, sentence2, and label
Approximate statistics based on the first 1000 samples:
sentence1 sentence2 label
type string string float
details
min: 3 tokens
mean: 15.05 tokens
max: 58 tokens

min: 4 tokens
mean: 15.57 tokens
max: 53 tokens

min: 0.0
mean: 0.48
max: 1.0

	sentence1	sentence2	label
type	string	string	float
details	min: 3 tokens mean: 15.05 tokens max: 58 tokens	min: 4 tokens mean: 15.57 tokens max: 53 tokens	min: 0.0 mean: 0.48 max: 1.0

Samples:

sentence1	sentence2	label
`сумка дорожная складная полет оранжевая bradex td 0599`	`сумка для путешествий складная брадекс orange`	`1.0`
`наушники sennheiser hd 450bt белый`	`наушники сенхайзер hd 450bt white`	`1.0`
`перчатки stg al-05-1871 синие серые черные зеленыеполноразмерные xl`	`перчатки stg al-05-1871 blue gray black green full size xl`	`1.0`

Loss: CosineSimilarityLoss with these parameters:

{
    "loss_fct": "torch.nn.modules.loss.MSELoss"
}

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: steps
per_device_train_batch_size: 32
per_device_eval_batch_size: 32
learning_rate: 4.7459131195420915e-05
weight_decay: 0.03196240090522689
num_train_epochs: 2
warmup_ratio: 0.014344463935915175
fp16: 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: 32
per_device_eval_batch_size: 32
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: 4.7459131195420915e-05
weight_decay: 0.03196240090522689
adam_beta1: 0.9
adam_beta2: 0.999
adam_epsilon: 1e-08
max_grad_norm: 1.0
num_train_epochs: 2
max_steps: -1
lr_scheduler_type: linear
lr_scheduler_kwargs: {}
warmup_ratio: 0.014344463935915175
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: 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	eval_cosine_ap
0.0287	100	0.189	-	-
0.0574	200	0.0695	-	-
0.0861	300	0.067	-	-
0.1148	400	0.0643	-	-
0.1435	500	0.0594	0.0549	0.9862
0.1722	600	0.0565	-	-
0.2009	700	0.0535	-	-
0.2296	800	0.0506	-	-
0.2583	900	0.0549	-	-
0.2870	1000	0.0535	0.0451	0.9888
0.3157	1100	0.0492	-	-
0.3444	1200	0.0499	-	-
0.3731	1300	0.0486	-	-
0.4018	1400	0.0458	-	-
0.4305	1500	0.0458	0.0419	0.9877
0.4592	1600	0.0502	-	-
0.4879	1700	0.045	-	-
0.5166	1800	0.0435	-	-
0.5454	1900	0.0426	-	-
0.5741	2000	0.0422	0.0386	0.9906
0.6028	2100	0.0436	-	-
0.6315	2200	0.043	-	-
0.6602	2300	0.0432	-	-
0.6889	2400	0.0397	-	-
0.7176	2500	0.0394	0.0357	0.9903
0.7463	2600	0.039	-	-
0.7750	2700	0.0398	-	-
0.8037	2800	0.0394	-	-
0.8324	2900	0.0426	-	-
0.8611	3000	0.0345	0.0341	0.9921
0.8898	3100	0.0361	-	-
0.9185	3200	0.0365	-	-
0.9472	3300	0.0401	-	-
0.9759	3400	0.0391	-	-
1.0046	3500	0.0342	0.0310	0.9928
1.0333	3600	0.0267	-	-
1.0620	3700	0.0264	-	-
1.0907	3800	0.0263	-	-
1.1194	3900	0.0248	-	-
1.1481	4000	0.0282	0.0301	0.9928
1.1768	4100	0.0279	-	-
1.2055	4200	0.0258	-	-
1.2342	4300	0.0248	-	-
1.2629	4400	0.0289	-	-
1.2916	4500	0.0261	0.0291	0.9935
1.3203	4600	0.0262	-	-
1.3490	4700	0.0276	-	-
1.3777	4800	0.0256	-	-
1.4064	4900	0.0272	-	-
1.4351	5000	0.0283	0.0284	0.9939
1.4638	5100	0.0254	-	-
1.4925	5200	0.0252	-	-
1.5212	5300	0.0234	-	-
1.5499	5400	0.0228	-	-
1.5786	5500	0.0248	0.0277	0.9941
1.6073	5600	0.024	-	-
1.6361	5700	0.0225	-	-
1.6648	5800	0.0234	-	-
1.6935	5900	0.0226	-	-
1.7222	6000	0.0248	0.0265	0.9942
1.7509	6100	0.0247	-	-
1.7796	6200	0.0219	-	-
1.8083	6300	0.026	-	-
1.8370	6400	0.0209	-	-
1.8657	6500	0.0252	0.0262	0.9945
1.8944	6600	0.0218	-	-
1.9231	6700	0.0223	-	-
1.9518	6800	0.0228	-	-
1.9805	6900	0.0242	-	-
2.0	6968	-	0.0257	0.9946

Framework Versions

Python: 3.11.11
Sentence Transformers: 4.1.0
Transformers: 4.51.3
PyTorch: 2.6.0+cu124
Accelerate: 1.5.2
Datasets: 3.6.0
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",
}