SentenceTransformer based on distilbert/distilbert-base-uncased

This is a sentence-transformers model finetuned from distilbert/distilbert-base-uncased on the wiki1m-for-simcse 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: distilbert/distilbert-base-uncased
Maximum Sequence Length: 32 tokens
Output Dimensionality: 768 dimensions
Similarity Function: Cosine Similarity
Training Dataset:
- wiki1m-for-simcse
Language: en

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': 32, 'do_lower_case': False, 'architecture': '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("tomaarsen/distilbert-base-uncased-stsb-simcse")
# Run inference
sentences = [
    'He attained the rank of rear admiral (shaojiang) in July 1999, and was promoted to the rank of vice admiral (zhongjiang) in July 2006.',
    'He attained the rank of rear admiral (shaojiang) in July 1999, and was promoted to the rank of vice admiral (zhongjiang) in July 2006.',
    'Kazakh playing cards differ from the classic French deck in that it has unique non-standard suits, i.e.',
]
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,  1.0000, -0.0107],
#         [ 1.0000,  1.0000, -0.0107],
#         [-0.0107, -0.0107,  1.0000]])

Evaluation

Metrics

Semantic Similarity

Datasets: sts-dev and sts-test
Evaluated with EmbeddingSimilarityEvaluator

Metric	sts-dev	sts-test
pearson_cosine	0.7683	0.746
spearman_cosine	0.7736	0.7419

Training Details

Training Dataset

wiki1m-for-simcse

Dataset: wiki1m-for-simcse at b20d549
Size: 985,723 training samples
Columns: sentence1 and sentence2
Approximate statistics based on the first 1000 samples:
sentence1 sentence2
type string string
details
min: 4 tokens
mean: 21.73 tokens
max: 32 tokens

min: 4 tokens
mean: 21.73 tokens
max: 32 tokens

	sentence1	sentence2
type	string	string
details	min: 4 tokens mean: 21.73 tokens max: 32 tokens	min: 4 tokens mean: 21.73 tokens max: 32 tokens

Samples:

sentence1	sentence2
`YMCA in South Australia`	`YMCA in South Australia`
`South Australia (SA) has a unique position in Australia's history as, unlike the other states which were founded as colonies, South Australia began as a self governing province Many were attracted to this and Adelaide and SA developed as an independent and free thinking state.`	`South Australia (SA) has a unique position in Australia's history as, unlike the other states which were founded as colonies, South Australia began as a self governing province Many were attracted to this and Adelaide and SA developed as an independent and free thinking state.`
`The compound of philosophical radicalism, evangelical religion and self reliant ability typical of its founders had given an equalitarian flavour to South Australian thinking from the beginning.`	`The compound of philosophical radicalism, evangelical religion and self reliant ability typical of its founders had given an equalitarian flavour to South Australian thinking from the beginning.`

Loss: MultipleNegativesRankingLoss with these parameters:

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

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: steps
per_device_train_batch_size: 128
num_train_epochs: 1
warmup_ratio: 0.1
warmup_steps: 0.1
fp16: True

All Hyperparameters

Click to expand

do_predict: False
eval_strategy: steps
prediction_loss_only: True
per_device_train_batch_size: 128
per_device_eval_batch_size: 8
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: 1.0
num_train_epochs: 1
max_steps: -1
lr_scheduler_type: linear
lr_scheduler_kwargs: None
warmup_ratio: 0.1
warmup_steps: 0.1
log_level: passive
log_level_replica: warning
log_on_each_node: True
logging_nan_inf_filter: True
enable_jit_checkpoint: False
save_on_each_node: False
save_only_model: False
restore_callback_states_from_checkpoint: False
use_cpu: False
seed: 42
data_seed: None
bf16: False
fp16: True
bf16_full_eval: False
fp16_full_eval: False
tf32: None
local_rank: -1
ddp_backend: None
debug: []
dataloader_drop_last: False
dataloader_num_workers: 0
dataloader_prefetch_factor: None
disable_tqdm: False
remove_unused_columns: True
label_names: None
load_best_model_at_end: False
ignore_data_skip: False
fsdp: []
fsdp_config: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}
accelerator_config: {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}
parallelism_config: None
deepspeed: None
label_smoothing_factor: 0.0
optim: adamw_torch_fused
optim_args: None
group_by_length: False
length_column_name: length
project: huggingface
trackio_space_id: trackio
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
push_to_hub: False
resume_from_checkpoint: None
hub_model_id: None
hub_strategy: every_save
hub_private_repo: None
hub_always_push: False
hub_revision: None
gradient_checkpointing: False
gradient_checkpointing_kwargs: None
include_for_metrics: []
eval_do_concat_batches: True
auto_find_batch_size: False
full_determinism: False
ddp_timeout: 1800
torch_compile: False
torch_compile_backend: None
torch_compile_mode: None
include_num_input_tokens_seen: no
neftune_noise_alpha: None
optim_target_modules: None
batch_eval_metrics: False
eval_on_start: False
use_liger_kernel: False
liger_kernel_config: None
eval_use_gather_object: False
average_tokens_across_devices: True
use_cache: False
prompts: None
batch_sampler: batch_sampler
multi_dataset_batch_sampler: proportional
router_mapping: {}
learning_rate_mapping: {}

Training Logs

Click to expand

Epoch	Step	Training Loss	sts-dev_spearman_cosine	sts-test_spearman_cosine
-1	-1	-	0.6728	0.5705
0.0101	78	0.0417	-	-
0.0203	156	0.0023	-	-
0.0304	234	0.0009	-	-
0.0405	312	0.0003	-	-
0.0506	390	0.0003	-	-
0.0608	468	0.0001	-	-
0.0709	546	0.0002	-	-
0.0810	624	0.0001	-	-
0.0912	702	0.0000	-	-
0.1001	771	-	0.7898	-
0.1013	780	0.0000	-	-
0.1114	858	0.0001	-	-
0.1215	936	0.0000	-	-
0.1317	1014	0.0001	-	-
0.1418	1092	0.0002	-	-
0.1519	1170	0.0000	-	-
0.1621	1248	0.0000	-	-
0.1722	1326	0.0000	-	-
0.1823	1404	0.0000	-	-
0.1924	1482	0.0000	-	-
0.2002	1542	-	0.7828	-
0.2026	1560	0.0000	-	-
0.2127	1638	0.0002	-	-
0.2228	1716	0.0000	-	-
0.2330	1794	0.0000	-	-
0.2431	1872	0.0000	-	-
0.2532	1950	0.0002	-	-
0.2633	2028	0.0000	-	-
0.2735	2106	0.0000	-	-
0.2836	2184	0.0002	-	-
0.2937	2262	0.0000	-	-
0.3004	2313	-	0.7998	-
0.3039	2340	0.0000	-	-
0.3140	2418	0.0000	-	-
0.3241	2496	0.0000	-	-
0.3342	2574	0.0000	-	-
0.3444	2652	0.0002	-	-
0.3545	2730	0.0000	-	-
0.3646	2808	0.0001	-	-
0.3748	2886	0.0001	-	-
0.3849	2964	0.0001	-	-
0.3950	3042	0.0000	-	-
0.4005	3084	-	0.7713	-
0.4051	3120	0.0000	-	-
0.4153	3198	0.0004	-	-
0.4254	3276	0.0000	-	-
0.4355	3354	0.0001	-	-
0.4457	3432	0.0002	-	-
0.4558	3510	0.0000	-	-
0.4659	3588	0.0002	-	-
0.4760	3666	0.0002	-	-
0.4862	3744	0.0000	-	-
0.4963	3822	0.0000	-	-
0.5006	3855	-	0.7824	-
0.5064	3900	0.0006	-	-
0.5166	3978	0.0000	-	-
0.5267	4056	0.0000	-	-
0.5368	4134	0.0002	-	-
0.5469	4212	0.0000	-	-
0.5571	4290	0.0000	-	-
0.5672	4368	0.0000	-	-
0.5773	4446	0.0000	-	-
0.5875	4524	0.0000	-	-
0.5976	4602	0.0001	-	-
0.6007	4626	-	0.7806	-
0.6077	4680	0.0003	-	-
0.6178	4758	0.0007	-	-
0.6280	4836	0.0000	-	-
0.6381	4914	0.0000	-	-
0.6482	4992	0.0000	-	-
0.6584	5070	0.0000	-	-
0.6685	5148	0.0002	-	-
0.6786	5226	0.0002	-	-
0.6887	5304	0.0000	-	-
0.6989	5382	0.0004	-	-
0.7008	5397	-	0.7509	-
0.7090	5460	0.0000	-	-
0.7191	5538	0.0000	-	-
0.7293	5616	0.0000	-	-
0.7394	5694	0.0000	-	-
0.7495	5772	0.0002	-	-
0.7596	5850	0.0000	-	-
0.7698	5928	0.0003	-	-
0.7799	6006	0.0000	-	-
0.7900	6084	0.0003	-	-
0.8002	6162	0.0000	-	-
0.8009	6168	-	0.7714	-
0.8103	6240	0.0000	-	-
0.8204	6318	0.0000	-	-
0.8305	6396	0.0000	-	-
0.8407	6474	0.0001	-	-
0.8508	6552	0.0000	-	-
0.8609	6630	0.0000	-	-
0.8711	6708	0.0000	-	-
0.8812	6786	0.0003	-	-
0.8913	6864	0.0002	-	-
0.9011	6939	-	0.7711	-
0.9014	6942	0.0000	-	-
0.9116	7020	0.0000	-	-
0.9217	7098	0.0000	-	-
0.9318	7176	0.0000	-	-
0.9420	7254	0.0000	-	-
0.9521	7332	0.0000	-	-
0.9622	7410	0.0000	-	-
0.9723	7488	0.0000	-	-
0.9825	7566	0.0000	-	-
0.9926	7644	0.0000	-	-
-1	-1	-	0.7736	0.7419

Framework Versions

Python: 3.11.6
Sentence Transformers: 5.3.0.dev0
Transformers: 5.0.1.dev0
PyTorch: 2.10.0+cu126
Accelerate: 1.12.0
Datasets: 4.3.0
Tokenizers: 0.22.2

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}
}