SentenceTransformer based on distilbert/distilbert-base-uncased

This is a sentence-transformers model finetuned from distilbert/distilbert-base-uncased. 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: distilbert/distilbert-base-uncased
  • Maximum Sequence Length: 512 tokens
  • Output Dimensionality: 768 tokens
  • Similarity Function: Cosine Similarity

Model Sources

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'max_seq_length': 512, 'do_lower_case': False}) with Transformer model: DistilBertModel 
  (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("sentence_transformers_model_id")
# Run inference
sentences = [
    'T L F DUMMY CHEST LAT WIDEBAND 90 Deg Front 2020 CX482 G-S',
    'T R F DUMMY CHEST LAT WIDEBAND 90 Deg Front 2025 V363N G-S',
    'T R F DUMMY HEAD CG VERT WIDEBAND VIA Linear Impact Test 2021 C727 G-S',
]
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

Metric Value
pearson_cosine 0.2705
spearman_cosine 0.2799
pearson_manhattan 0.2287
spearman_manhattan 0.2535
pearson_euclidean 0.2302
spearman_euclidean 0.255
pearson_dot 0.2125
spearman_dot 0.1903
pearson_max 0.2705
spearman_max 0.2799

Semantic Similarity

Metric Value
pearson_cosine 0.2632
spearman_cosine 0.2722
pearson_manhattan 0.2177
spearman_manhattan 0.244
pearson_euclidean 0.2195
spearman_euclidean 0.2463
pearson_dot 0.2107
spearman_dot 0.1865
pearson_max 0.2632
spearman_max 0.2722

Training Details

Training Dataset

Unnamed Dataset

  • Size: 481,114 training samples
  • Columns: sentence1, sentence2, and score
  • Approximate statistics based on the first 1000 samples:
    sentence1 sentence2 score
    type string string float
    details
    • min: 16 tokens
    • mean: 32.14 tokens
    • max: 57 tokens
    • min: 17 tokens
    • mean: 32.62 tokens
    • max: 58 tokens
    • min: 0.0
    • mean: 0.45
    • max: 1.0
  • Samples:
    sentence1 sentence2 score
    T L C PLR SM SCS L2 HY REF 053 LAT WIDEBAND 75 Deg Oblique Left Side 10 in. Pole 2018 P558 G-S T PCM PWR POWER TO PCM VOLT 2 SEC WIDEBAND 75 Deg Oblique Left Side 10 in. Pole 2020 V363N VOLTS 0.5198143220305642
    T L F DUMMY L_FEMUR MX LOAD WIDEBAND 90 Deg Frontal Impact Simulation MY2025 U717 IN-LBS B L FRAME AT No 1 X MEM LAT WIDEBAND Inline 25% Left Front Offset Vehicle to Vehicle 2021 P702 G-S 0.5214072221695696
    T R F DOOR REAR OF SEAT H PT LAT WIDEBAND 75 Deg Oblique Right Side 10 in. Pole 2015 P552 G-S T SCS R2 HY BOS A12 008 TAP RIGHT C PILLAR VOLT WIDEBAND 30 Deg Front Angular Right 2021 CX727 VOLTS 0.322173496575591
  • Loss: CoSENTLoss with these parameters:
    {
        "scale": 20.0,
        "similarity_fct": "pairwise_cos_sim"
    }
    

Evaluation Dataset

Unnamed Dataset

  • Size: 103,097 evaluation samples
  • Columns: sentence1, sentence2, and score
  • Approximate statistics based on the first 1000 samples:
    sentence1 sentence2 score
    type string string float
    details
    • min: 17 tokens
    • mean: 31.98 tokens
    • max: 56 tokens
    • min: 15 tokens
    • mean: 31.96 tokens
    • max: 58 tokens
    • min: 0.0
    • mean: 0.45
    • max: 1.0
  • Samples:
    sentence1 sentence2 score
    T R F DUMMY NECK UPPER MZ LOAD WIDEBAND 90 Deg Frontal Impact Simulation 2026 GENERIC IN-LBS T R ROCKER AT C PILLAR LAT WIDEBAND 90 Deg Front 2021 P702 G-S 0.5234504780172093
    T L ROCKER AT B_PILLAR VERT WIDEBAND 90 Deg Front 2024.5 P702 G-S T RCM BTWN SEATS LOW G Z RCM C1 LZ ALV RC7 003 VOLT WIDEBAND 75 Deg Oblique Left Side 10 in. Pole 2018 P558 VOLTS 0.36805699821563936
    T R FRAME AT C_PILLAR LONG WIDEBAND 90 Deg Left Side IIHS MDB to Vehicle 2024.5 P702 G-S T L F LAP BELT AT ANCHOR LOAD WIDEBAND 90 DEG / LEFT SIDE DECEL-3G 2021 P702 LBF 0.5309750606095435
  • Loss: CoSENTLoss with these parameters:
    {
        "scale": 20.0,
        "similarity_fct": "pairwise_cos_sim"
    }
    

Training Hyperparameters

Non-Default Hyperparameters

  • per_device_train_batch_size: 64
  • per_device_eval_batch_size: 64
  • num_train_epochs: 32
  • warmup_ratio: 0.1
  • fp16: True

All Hyperparameters

Click to expand
  • overwrite_output_dir: False
  • do_predict: False
  • 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
  • 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: 32
  • max_steps: -1
  • lr_scheduler_type: linear
  • 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
  • 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: 7
  • ddp_backend: None
  • tpu_num_cores: None
  • tpu_metrics_debug: False
  • debug: []
  • dataloader_drop_last: True
  • dataloader_num_workers: 0
  • 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_grad_ckpt': False}
  • fsdp_transformer_layer_cls_to_wrap: 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
  • 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
  • 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: False
  • include_tokens_per_second: False
  • neftune_noise_alpha: None
  • batch_sampler: batch_sampler
  • multi_dataset_batch_sampler: proportional

Training Logs

Epoch Step Training Loss loss sts-dev_spearman_cosine
1.0650 1000 7.6111 7.5503 0.4087
2.1299 2000 7.5359 7.5420 0.4448
3.1949 3000 7.5232 7.5292 0.4622
4.2599 4000 7.5146 7.5218 0.4779
5.3248 5000 7.5045 7.5200 0.4880
6.3898 6000 7.4956 7.5191 0.4934
7.4547 7000 7.4873 7.5170 0.4967
8.5197 8000 7.4781 7.5218 0.4931
9.5847 9000 7.4686 7.5257 0.4961
10.6496 10000 7.4596 7.5327 0.4884
11.7146 11000 7.4498 7.5403 0.4860
12.7796 12000 7.4386 7.5507 0.4735
13.8445 13000 7.4253 7.5651 0.4660
14.9095 14000 7.4124 7.5927 0.4467
15.9744 15000 7.3989 7.6054 0.4314
17.0394 16000 7.3833 7.6654 0.4163
18.1044 17000 7.3669 7.7186 0.3967
19.1693 18000 7.3519 7.7653 0.3779
20.2343 19000 7.3349 7.8356 0.3651
21.2993 20000 7.3191 7.8772 0.3495
22.3642 21000 7.3032 7.9346 0.3412
23.4292 22000 7.2873 7.9624 0.3231
24.4941 23000 7.2718 8.0169 0.3161
25.5591 24000 7.2556 8.0633 0.3050
26.6241 25000 7.2425 8.1021 0.2958
27.6890 26000 7.2278 8.1563 0.2954
28.7540 27000 7.2124 8.1955 0.2882
29.8190 28000 7.2014 8.2234 0.2821
30.8839 29000 7.1938 8.2447 0.2792
31.9489 30000 7.1811 8.2609 0.2799
32.0 30048 - - 0.2722

Framework Versions

  • Python: 3.10.6
  • Sentence Transformers: 3.0.0
  • Transformers: 4.35.0
  • PyTorch: 2.1.0a0+4136153
  • Accelerate: 0.30.1
  • Datasets: 2.14.1
  • Tokenizers: 0.14.1

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",
}

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},
}
Downloads last month
18
Safetensors
Model size
66.4M params
Tensor type
F32
·
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Model tree for gkudirka/crash_encoder1-sts

Finetuned
(7016)
this model

Evaluation results