This is a sentence-transformers model finetuned from hiieu/halong_embedding. 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: hiieu/halong_embedding
Maximum Sequence Length: 512 tokens
Output Dimensionality: 768 dimensions
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: 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})
  (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.

python

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("TuanNM171284/TuanNM171284-HaLong-embedding-medical-v3")
# Run inference
sentences = [
    'Nếu một con bò có hành vi bất thường, khó khăn khi di chuyển và bị giảm thể trọng, đó có thể là dấu hiệu của bệnh gì?',
    'Nếu một con bò có các biểu hiện như hành vi bất thường, khó khăn trong di chuyển và giảm thể trọng, đó có thể là dấu hiệu của bệnh bò điên (Bovine Spongiform Encephalopathy - BSE).',
    'Trong giai đoạn viêm tấy cấp tính khi bị giãn dây chằng, người bệnh tuyệt đối không được chườm nóng, xoa dầu nóng, rượu thuốc vì có thể làm tổn thương nặng hơn. Đồng thời, không vận động vùng tổn thương và không được tiêm kháng viêm trực tiếp vào vùng tổn thương.',
]
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

Information Retrieval

Dataset: dim_768
Evaluated with InformationRetrievalEvaluator with these parameters:
json
1{ 2 "truncate_dim": 768 3}

Metric	Value
cosine_accuracy@1	0.7276
cosine_accuracy@3	0.9563
cosine_accuracy@5	0.9785
cosine_accuracy@10	0.9936
cosine_precision@1	0.7276
cosine_precision@3	0.3188
cosine_precision@5	0.1957
cosine_precision@10	0.0994
cosine_recall@1	0.7276
cosine_recall@3	0.9563
cosine_recall@5	0.9785
cosine_recall@10	0.9936
cosine_ndcg@10	0.8834
cosine_mrr@10	0.8458
cosine_map@100	0.8461

Information Retrieval

Dataset: dim_512
Evaluated with InformationRetrievalEvaluator with these parameters:
json
1{ 2 "truncate_dim": 512 3}

Metric	Value
cosine_accuracy@1	0.7283
cosine_accuracy@3	0.9578
cosine_accuracy@5	0.9797
cosine_accuracy@10	0.9926
cosine_precision@1	0.7283
cosine_precision@3	0.3193
cosine_precision@5	0.1959
cosine_precision@10	0.0993
cosine_recall@1	0.7283
cosine_recall@3	0.9578
cosine_recall@5	0.9797
cosine_recall@10	0.9926
cosine_ndcg@10	0.8839
cosine_mrr@10	0.8467
cosine_map@100	0.8472

Information Retrieval

Dataset: dim_256
Evaluated with InformationRetrievalEvaluator with these parameters:
json
1{ 2 "truncate_dim": 256 3}

Metric	Value
cosine_accuracy@1	0.7336
cosine_accuracy@3	0.9644
cosine_accuracy@5	0.9831
cosine_accuracy@10	0.9938
cosine_precision@1	0.7336
cosine_precision@3	0.3215
cosine_precision@5	0.1966
cosine_precision@10	0.0994
cosine_recall@1	0.7336
cosine_recall@3	0.9644
cosine_recall@5	0.9831
cosine_recall@10	0.9938
cosine_ndcg@10	0.8881
cosine_mrr@10	0.8518
cosine_map@100	0.8522

Information Retrieval

Dataset: dim_128
Evaluated with InformationRetrievalEvaluator with these parameters:
json
1{ 2 "truncate_dim": 128 3}

Metric	Value
cosine_accuracy@1	0.733
cosine_accuracy@3	0.9616
cosine_accuracy@5	0.9817
cosine_accuracy@10	0.9941
cosine_precision@1	0.733
cosine_precision@3	0.3205
cosine_precision@5	0.1963
cosine_precision@10	0.0994
cosine_recall@1	0.733
cosine_recall@3	0.9616
cosine_recall@5	0.9817
cosine_recall@10	0.9941
cosine_ndcg@10	0.8873
cosine_mrr@10	0.8507
cosine_map@100	0.851

Information Retrieval

Dataset: dim_64
Evaluated with InformationRetrievalEvaluator with these parameters:
json
1{ 2 "truncate_dim": 64 3}

Metric	Value
cosine_accuracy@1	0.7273
cosine_accuracy@3	0.9585
cosine_accuracy@5	0.9802
cosine_accuracy@10	0.9924
cosine_precision@1	0.7273
cosine_precision@3	0.3195
cosine_precision@5	0.196
cosine_precision@10	0.0992
cosine_recall@1	0.7273
cosine_recall@3	0.9585
cosine_recall@5	0.9802
cosine_recall@10	0.9924
cosine_ndcg@10	0.8835
cosine_mrr@10	0.8462
cosine_map@100	0.8466

Training Details

Training Dataset

Unnamed Dataset

Size: 5,808 training samples
Columns: anchor and positive
Approximate statistics based on the first 1000 samples:
anchor positive
type string string
details
min: 8 tokens
mean: 23.52 tokens
max: 83 tokens
min: 10 tokens
mean: 71.2 tokens
max: 170 tokens

Samples:

anchor	positive
`Bệnh Addison là gì?`	`Bệnh Addison, còn được gọi là suy thượng thận nguyên phát, là tình trạng tuyến thượng thận không sản xuất đủ các hormone cortisol và aldosterone.`
`Nguyên nhân chính gây ra bệnh Addison là gì?`	`Hầu hết các trường hợp bệnh Addison là do bệnh tự miễn, khi hệ thống miễn dịch tấn công nhầm tuyến thượng thận. Các nguyên nhân khác bao gồm nhiễm trùng kéo dài như bệnh lao, HIV, nhiễm nấm, và các tế bào ung thư lây lan đến tuyến thượng thận.`
`Tôi thường xuyên mệt mỏi, yếu cơ, sụt cân và da bị sạm màu. Đây có thể là triệu chứng của bệnh gì?`	`Các triệu chứng như mệt mỏi mãn tính, yếu cơ, giảm cân, và da sẫm màu (nám, sạm đen, tàn nhang) có thể là dấu hiệu của bệnh Addison.`

Loss: MatryoshkaLoss with these parameters:
json
1{ 2 "loss": "MultipleNegativesRankingLoss", 3 "matryoshka_dims": [ 4 768, 5 512, 6 256, 7 128, 8 64 9 ], 10 "matryoshka_weights": [ 11 1, 12 1, 13 1, 14 1, 15 1 16 ], 17 "n_dims_per_step": -1 18}

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: steps
per_device_eval_batch_size: 4
gradient_accumulation_steps: 4
learning_rate: 2e-05
lr_scheduler_type: cosine
warmup_ratio: 0.1
bf16: True
load_best_model_at_end: True
optim: adamw_torch_fused
batch_sampler: no_duplicates

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
eval_strategy: steps
prediction_loss_only: True
per_device_train_batch_size: 8
per_device_eval_batch_size: 4
per_gpu_train_batch_size: None
per_gpu_eval_batch_size: None
gradient_accumulation_steps: 4
eval_accumulation_steps: None
torch_empty_cache_steps: None
learning_rate: 2e-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: 3
max_steps: -1
lr_scheduler_type: cosine
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: True
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: 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_fused
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: no_duplicates
multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	Training Loss	dim_768_cosine_ndcg@10	dim_512_cosine_ndcg@10	dim_256_cosine_ndcg@10	dim_128_cosine_ndcg@10	dim_64_cosine_ndcg@10
0.0551	10	0.9863	0.8053	0.8005	0.7821	0.7483	0.6675
0.1102	20	1.046	0.8175	0.8137	0.8027	0.7784	0.7122
0.1653	30	1.045	0.8269	0.8241	0.8172	0.8010	0.7524
0.2204	40	1.1269	0.8329	0.8309	0.8275	0.8147	0.7808
0.2755	50	0.5781	0.8390	0.8370	0.8357	0.8271	0.7991
0.3306	60	0.319	0.8391	0.8377	0.8379	0.8321	0.8101
0.3857	70	0.1891	0.8460	0.8454	0.8447	0.8414	0.8279
0.4408	80	0.2646	0.8514	0.8518	0.8506	0.8486	0.8335
0.4959	90	0.2737	0.8542	0.8544	0.8541	0.8514	0.8372
0.5510	100	0.1422	0.8565	0.8574	0.8564	0.8525	0.8366
0.6061	110	0.3984	0.8556	0.8555	0.8551	0.8520	0.8382
0.6612	120	0.443	0.8589	0.8593	0.8606	0.8565	0.8440
0.7163	130	0.1714	0.8584	0.8598	0.8610	0.8578	0.8477
0.7713	140	0.0658	0.8577	0.8583	0.8597	0.8573	0.8496
0.8264	150	0.2713	0.8597	0.8604	0.8618	0.8606	0.8519
0.8815	160	0.6141	0.8627	0.8638	0.8653	0.8623	0.8537
0.9366	170	0.741	0.8711	0.8709	0.8695	0.8651	0.8509
0.9917	180	0.1569	0.8694	0.8692	0.8689	0.8669	0.8545
1.0441	190	0.042	0.8638	0.8638	0.8657	0.8651	0.8567
1.0992	200	0.0586	0.8581	0.8587	0.8629	0.8610	0.8533
1.1543	210	0.2068	0.8616	0.8634	0.8666	0.8665	0.8588
1.2094	220	0.1943	0.8733	0.8741	0.8773	0.8747	0.8664
1.2645	230	0.0132	0.8664	0.8675	0.8704	0.8705	0.8629
1.3196	240	0.1768	0.8681	0.8696	0.8725	0.8727	0.8654
1.3747	250	0.1628	0.8772	0.8771	0.8797	0.8790	0.8716
1.4298	260	0.0547	0.8749	0.8756	0.8784	0.8783	0.8721
1.4848	270	0.0532	0.8794	0.8800	0.8826	0.8808	0.8743
1.5399	280	0.0991	0.8779	0.8785	0.8817	0.8808	0.8747
1.5950	290	0.0678	0.8763	0.8767	0.8801	0.8794	0.8735
1.6501	300	0.0354	0.8760	0.8768	0.8800	0.8793	0.8732
1.7052	310	0.0463	0.8787	0.8796	0.8825	0.8823	0.8767
1.7603	320	0.0391	0.8791	0.8799	0.8834	0.8826	0.8780
1.8154	330	0.0855	0.8777	0.8781	0.8813	0.8810	0.8763
1.8705	340	0.0339	0.8782	0.8785	0.8820	0.8816	0.8774
1.9256	350	0.0305	0.8760	0.8771	0.8814	0.8813	0.8772
1.9807	360	0.082	0.8784	0.8791	0.8831	0.8826	0.8791
2.0331	370	0.0791	0.8781	0.8793	0.8832	0.8832	0.8801
2.0882	380	0.0336	0.8799	0.8805	0.8845	0.8845	0.8818
2.1433	390	0.0374	0.8788	0.8799	0.8843	0.8842	0.8813
2.1983	400	0.0232	0.8779	0.8791	0.8833	0.8824	0.8808
2.2534	410	0.0276	0.8776	0.8791	0.8822	0.8822	0.8804
2.3085	420	0.0442	0.8777	0.8791	0.8830	0.8818	0.8798
2.3636	430	0.0323	0.8786	0.8798	0.8838	0.8827	0.8803
2.4187	440	0.0284	0.8823	0.8828	0.8862	0.8856	0.8826
2.4738	450	0.0275	0.8840	0.8848	0.8885	0.8874	0.8839
2.5289	460	0.0267	0.8842	0.8847	0.8892	0.8878	0.8837
2.5840	470	0.023	0.8838	0.8846	0.8889	0.8879	0.8839
2.6391	480	0.2072	0.8838	0.8843	0.8885	0.8879	0.8838
2.6942	490	0.0506	0.8836	0.8841	0.8882	0.8877	0.8836
2.7493	500	0.0368	0.8834	0.8838	0.8883	0.8872	0.8834
2.8044	510	0.0205	0.8832	0.8838	0.8885	0.8876	0.8836
2.8595	520	0.022	0.8831	0.8836	0.8880	0.8875	0.8833
2.9146	530	0.0167	0.8836	0.8840	0.8882	0.8876	0.8839
2.9697	540	0.0654	0.8834	0.8839	0.8881	0.8873	0.8835

Framework Versions

Python: 3.11.13
Sentence Transformers: 4.1.0
Transformers: 4.52.4
PyTorch: 2.6.0+cu124
Accelerate: 1.8.1
Datasets: 3.6.0
Tokenizers: 0.21.2

Citation

BibTeX

Sentence Transformers

bibtex

@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",
}

MatryoshkaLoss

bibtex

@misc{kusupati2024matryoshka,
    title={Matryoshka Representation Learning},
    author={Aditya Kusupati and Gantavya Bhatt and Aniket Rege and Matthew Wallingford and Aditya Sinha and Vivek Ramanujan and William Howard-Snyder and Kaifeng Chen and Sham Kakade and Prateek Jain and Ali Farhadi},
    year={2024},
    eprint={2205.13147},
    archivePrefix={arXiv},
    primaryClass={cs.LG}
}

MultipleNegativesRankingLoss

bibtex

@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}
}

	anchor	positive
type	string	string
details	min: 8 tokens mean: 23.52 tokens max: 83 tokens	min: 10 tokens mean: 71.2 tokens max: 170 tokens