This is a sentence-transformers model finetuned from BookingCare/multilingual-e5-base-v2. 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: BookingCare/multilingual-e5-base-v2
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: 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("BookingCare/multilingual-base-e5-v3.1")
# Run inference
sentences = [
    'Có cách nào để cải thiện môi trường làm việc độc hại không?',
    ' Tương tự như chất độc trong không khí,\ncó thể gây hại cho sức khỏe tinh thần và thể chất của người lao động. Nếu bạn tiếp tục làm việc quá lâu, nó có thể dẫn đến mức độ căng thẳng cao, lòng tự trọng bị tụt giảm và bệnh lý trầm cảm. môi trường làm việc độc hại Nếu sự vấn đề đến từ lãnh đạo hoặc tư duy của công ty, bạn sẽ không thể làm được gì nhiều để cải thiện, tuy nhiên nếu vấn đề chỉ đến từ 1 hoặc 2 người, bạn có thể thảo luận với người quản lý đáng tin cậy hoặc nói chuyện với bộ phận nhân sự (HR). Sau đó, công ty có thể thuê trợ giúp từ bên ngoài như thông qua chương trình hỗ trợ nhân viên (EAP) để giúp giải quyết vấn đề. Nếu không có sự lựa chọn nào ngoài việc ở lại lúc này, hãy thử đặt mình vào một vỏ bọc nhỏ, cố gắng tránh mọi thị phi và giữ an tĩnh cho riêng mình. Tập trung vào các mục tiêu bên ngoài công việc và bắt đầu lập kế hoạch để thoát ra ngoài.',
    ' Chấn thương đầu, cổ, tủy sống rất nguy hiểm vì có thể gây mất vận động (liệt),\nhôn mê\nvà tử vong.\nChấn thương tủy sống\nlà nguyên nhân tổn thương thần kinh và gây ra\nkhó thở\n. hôn mê Chấn thương tủy sống khó thở  Người bệnh bị chấn thương đầu, cổ, tủy sống cần được vận chuyển hết sức thận trọng. Bởi bất cứ vận động nào không phù hợp cũng có thể làm chấn thương nặng thêm như liệt tay hoặc chân. Nếu người bệnh không tỉnh, cần thực hiện hỗ trợ sự sống cơ bản.',
]
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: healthcare-dev
Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.8483
cosine_accuracy@3	0.9266
cosine_accuracy@5	0.9465
cosine_accuracy@10	0.9639
cosine_precision@1	0.8483
cosine_precision@3	0.3089
cosine_precision@5	0.1893
cosine_precision@10	0.0964
cosine_recall@1	0.8483
cosine_recall@3	0.9266
cosine_recall@5	0.9465
cosine_recall@10	0.9639
cosine_ndcg@10	0.9104
cosine_mrr@10	0.8928
cosine_map@100	0.8943
dot_accuracy@1	0.8483
dot_accuracy@3	0.9266
dot_accuracy@5	0.9465
dot_accuracy@10	0.9639
dot_precision@1	0.8483
dot_precision@3	0.3089
dot_precision@5	0.1893
dot_precision@10	0.0964
dot_recall@1	0.8483
dot_recall@3	0.9266
dot_recall@5	0.9465
dot_recall@10	0.9639
dot_ndcg@10	0.9104
dot_mrr@10	0.8928
dot_map@100	0.8943

Information Retrieval

Dataset: healthcare-test
Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.6714
cosine_accuracy@3	0.8209
cosine_accuracy@5	0.865
cosine_accuracy@10	0.8996
cosine_precision@1	0.6714
cosine_precision@3	0.2736
cosine_precision@5	0.173
cosine_precision@10	0.09
cosine_recall@1	0.6714
cosine_recall@3	0.8209
cosine_recall@5	0.865
cosine_recall@10	0.8996
cosine_ndcg@10	0.7892
cosine_mrr@10	0.7533
cosine_map@100	0.7563
dot_accuracy@1	0.6714
dot_accuracy@3	0.8209
dot_accuracy@5	0.865
dot_accuracy@10	0.8996
dot_precision@1	0.6714
dot_precision@3	0.2736
dot_precision@5	0.173
dot_precision@10	0.09
dot_recall@1	0.6714
dot_recall@3	0.8209
dot_recall@5	0.865
dot_recall@10	0.8996
dot_ndcg@10	0.7892
dot_mrr@10	0.7533
dot_map@100	0.7563

Training Details

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: steps
per_device_train_batch_size: 5
per_device_eval_batch_size: 6
learning_rate: 2e-05
num_train_epochs: 2
warmup_ratio: 0.1
fp16: True
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: 5
per_device_eval_batch_size: 6
per_gpu_train_batch_size: None
per_gpu_eval_batch_size: None
gradient_accumulation_steps: 1
eval_accumulation_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: 2
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: 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}
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
batch_sampler: no_duplicates
multi_dataset_batch_sampler: proportional

Training Logs

Click to expand

Epoch	Step	Training Loss	Validation Loss	healthcare-dev_cosine_map@100	healthcare-test_cosine_map@100
0	0	-	-	0.8140	0.6266
0.0126	100	0.1461	0.1289	0.8342	-
0.0251	200	0.1063	0.1130	0.8448	-
0.0377	300	0.1015	0.1008	0.8536	-
0.0502	400	0.086	0.0937	0.8586	-
0.0628	500	0.0824	0.0895	0.8654	-
0.0753	600	0.1008	0.0872	0.8669	-
0.0879	700	0.0755	0.0930	0.8658	-
0.1004	800	0.0968	0.0923	0.8683	-
0.1130	900	0.1011	0.0889	0.8677	-
0.1255	1000	0.0943	0.0805	0.8706	-
0.1381	1100	0.0668	0.0782	0.8660	-
0.1507	1200	0.0746	0.0814	0.8738	-
0.1632	1300	0.0825	0.0768	0.8728	-
0.1758	1400	0.0851	0.0860	0.8660	-
0.1883	1500	0.1029	0.0736	0.8752	-
0.2009	1600	0.071	0.0805	0.8760	-
0.2134	1700	0.081	0.0717	0.8731	-
0.2260	1800	0.0767	0.0698	0.8744	-
0.2385	1900	0.0895	0.0795	0.8705	-
0.2511	2000	0.0666	0.0740	0.8701	-
0.2637	2100	0.0791	0.0702	0.8733	-
0.2762	2200	0.0779	0.0797	0.8767	-
0.2888	2300	0.0812	0.0739	0.8790	-
0.3013	2400	0.0492	0.0754	0.8798	-
0.3139	2500	0.0442	0.0850	0.8722	-
0.3264	2600	0.0652	0.0901	0.8717	-
0.3390	2700	0.0579	0.0865	0.8733	-
0.3515	2800	0.0543	0.0945	0.8742	-
0.3641	2900	0.0639	0.0950	0.8678	-
0.3766	3000	0.0587	0.0824	0.8775	-
0.3892	3100	0.078	0.0864	0.8675	-
0.4018	3200	0.091	0.0686	0.8763	-
0.4143	3300	0.0763	0.0780	0.8734	-
0.4269	3400	0.0552	0.0842	0.8668	-
0.4394	3500	0.0549	0.0748	0.8748	-
0.4520	3600	0.0642	0.0755	0.8790	-
0.4645	3700	0.0796	0.0815	0.8650	-
0.4771	3800	0.0949	0.0755	0.8642	-
0.4896	3900	0.0783	0.0691	0.8698	-
0.5022	4000	0.0534	0.0655	0.8822	-
0.5148	4100	0.0453	0.0709	0.8742	-
0.5273	4200	0.0498	0.0612	0.8838	-
0.5399	4300	0.0903	0.0619	0.8795	-
0.5524	4400	0.0667	0.0712	0.8825	-
0.5650	4500	0.0364	0.0962	0.8722	-
0.5775	4600	0.0502	0.0706	0.8790	-
0.5901	4700	0.0685	0.0672	0.8788	-
0.6026	4800	0.0675	0.0695	0.8768	-
0.6152	4900	0.083	0.0680	0.8787	-
0.6277	5000	0.0598	0.0715	0.8769	-
0.6403	5100	0.0548	0.0710	0.8744	-
0.6529	5200	0.0682	0.0679	0.8855	-
0.6654	5300	0.0378	0.0779	0.8809	-
0.6780	5400	0.0274	0.0711	0.8864	-
0.6905	5500	0.0635	0.0699	0.8842	-
0.7031	5600	0.0681	0.0563	0.8867	-
0.7156	5700	0.0389	0.0595	0.8806	-
0.7282	5800	0.0419	0.0586	0.8796	-
0.7407	5900	0.0306	0.0520	0.8837	-
0.7533	6000	0.0418	0.0622	0.8759	-
0.7659	6100	0.0459	0.0691	0.8770	-
0.7784	6200	0.0616	0.0679	0.8818	-
0.7910	6300	0.0541	0.0658	0.8888	-
0.8035	6400	0.0742	0.0767	0.8890	-
0.8161	6500	0.0531	0.0675	0.8904	-
0.8286	6600	0.0513	0.0720	0.8909	-
0.8412	6700	0.0505	0.0722	0.8897	-
0.8537	6800	0.0451	0.0705	0.8895	-
0.8663	6900	0.0456	0.0704	0.8892	-
0.8788	7000	0.0506	0.0668	0.8901	-
0.8914	7100	0.0424	0.0556	0.8903	-
0.9040	7200	0.036	0.0602	0.8890	-
0.9165	7300	0.0545	0.0656	0.8886	-
0.9291	7400	0.0604	0.0695	0.8863	-
0.9416	7500	0.0362	0.0617	0.8909	-
0.9542	7600	0.0442	0.0666	0.8932	-
0.9667	7700	0.0398	0.0648	0.8886	-
0.9793	7800	0.0471	0.0654	0.8921	-
0.9918	7900	0.0716	0.0615	0.8933	-
1.0044	8000	0.0306	0.0735	0.8929	-
1.0169	8100	0.0601	0.0708	0.8927	-
1.0295	8200	0.041	0.0672	0.8939	-
1.0421	8300	0.0311	0.0693	0.8956	-
1.0546	8400	0.0508	0.0700	0.8984	-
1.0672	8500	0.0414	0.0640	0.8933	-
1.0797	8600	0.0451	0.0606	0.8943	-
1.0923	8700	0.0347	0.0668	0.8898	-
1.1048	8800	0.0413	0.0663	0.8965	-
1.1174	8900	0.0369	0.0641	0.8964	-
1.1299	9000	0.0252	0.0543	0.8925	-
1.1425	9100	0.0221	0.0529	0.8879	-
1.1551	9200	0.0306	0.0568	0.8951	-
1.1676	9300	0.0378	0.0616	0.8954	-
1.1802	9400	0.0338	0.0592	0.8913	-
1.1927	9500	0.0207	0.0565	0.8992	-
1.2053	9600	0.0259	0.0600	0.8962	-
1.2178	9700	0.0079	0.0655	0.8950	-
1.2304	9800	0.022	0.0660	0.8959	-
1.2429	9900	0.0296	0.0657	0.8960	-
1.2555	10000	0.0263	0.0667	0.8916	-
1.2680	10100	0.0184	0.0590	0.8951	-
1.2806	10200	0.0254	0.0587	0.8926	-
1.2932	10300	0.0213	0.0627	0.8896	-
1.3057	10400	0.0141	0.0655	0.8905	-
1.3183	10500	0.0077	0.0702	0.8910	-
1.3308	10600	0.0159	0.0700	0.8921	-
1.3434	10700	0.015	0.0674	0.8908	-
1.3559	10800	0.018	0.0698	0.8955	-
1.3685	10900	0.0156	0.0677	0.8908	-
1.3810	11000	0.0219	0.0666	0.8952	-
1.3936	11100	0.015	0.0640	0.8941	-
1.4062	11200	0.0231	0.0634	0.8916	-
1.4187	11300	0.0172	0.0679	0.8940	-
1.4313	11400	0.0228	0.0636	0.8925	-
1.4438	11500	0.0199	0.0655	0.8935	-
1.4564	11600	0.025	0.0687	0.8961	-
1.4689	11700	0.0277	0.0679	0.8922	-
1.4815	11800	0.0227	0.0672	0.8912	-
1.4940	11900	0.0222	0.0679	0.8914	-
1.5066	12000	0.0138	0.0656	0.8929	-
1.5191	12100	0.0107	0.0663	0.8916	-
1.5317	12200	0.0137	0.0580	0.8927	-
1.5443	12300	0.0311	0.0578	0.8948	-
1.5568	12400	0.0198	0.0621	0.8953	-
1.5694	12500	0.0084	0.0638	0.8950	-
1.5819	12600	0.0166	0.0600	0.8959	-
1.5945	12700	0.0251	0.0599	0.8928	-
1.6070	12800	0.0154	0.0624	0.8973	-
1.6196	12900	0.0301	0.0629	0.8937	-
1.6321	13000	0.0198	0.0616	0.8937	-
1.6447	13100	0.0146	0.0601	0.8914	-
1.6573	13200	0.0128	0.0610	0.8945	-
1.6698	13300	0.0092	0.0606	0.8920	-
1.6824	13400	0.0121	0.0595	0.8954	-
1.6949	13500	0.0183	0.0577	0.8918	-
1.7075	13600	0.0245	0.0572	0.8944	-
1.7200	13700	0.0166	0.0592	0.8931	-
1.7326	13800	0.0059	0.0593	0.8929	-
1.7451	13900	0.0087	0.0581	0.8918	-
1.7577	14000	0.0252	0.0595	0.8924	-
1.7702	14100	0.0165	0.0585	0.8976	-
1.7828	14200	0.022	0.0595	0.8976	-
1.7954	14300	0.0143	0.0602	0.8967	-
1.8079	14400	0.0328	0.0608	0.8974	-
1.8205	14500	0.0228	0.0597	0.8983	-
1.8330	14600	0.009	0.0594	0.8979	-
1.8456	14700	0.0188	0.0593	0.8952	-
1.8581	14800	0.0157	0.0583	0.8962	-
1.8707	14900	0.0116	0.0571	0.8969	-
1.8832	15000	0.0183	0.0559	0.8989	-
1.8958	15100	0.0118	0.0554	0.8972	-
1.9083	15200	0.0192	0.0559	0.8970	-
1.9209	15300	0.0109	0.0566	0.8957	-
1.9335	15400	0.0145	0.0566	0.8975	-
1.9460	15500	0.0131	0.0573	0.8965	-
1.9586	15600	0.0104	0.0575	0.8969	-
1.9711	15700	0.0185	0.0581	0.8961	-
1.9837	15800	0.0131	0.0579	0.8943	-
1.9962	15900	0.032	0.0576	0.8943	-
2.0	15930	-	-	-	0.7563

Framework Versions

Python: 3.10.13
Sentence Transformers: 3.2.0
Transformers: 4.41.2
PyTorch: 2.1.2
Accelerate: 0.30.1
Datasets: 2.19.2
Tokenizers: 0.19.1

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",
}

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}
}

BookingCare/multilingual-e5-base-v3.1

SentenceTransformer based on BookingCare/multilingual-e5-base-v2

Model Details

Model Description

Model Sources

Full Model Architecture

Usage

Direct Usage (Sentence Transformers)

Evaluation

Metrics

Information Retrieval

Information Retrieval

Training Details

Training Hyperparameters

Non-Default Hyperparameters

All Hyperparameters

Training Logs

Framework Versions

Citation

BibTeX

Sentence Transformers

MultipleNegativesRankingLoss

BookingCare/
multilingual-e5-base-v3.1