This is a sentence-transformers model finetuned from intfloat/multilingual-e5-large-instruct on the clef-me5-mined-pairs-train-pairs dataset. It maps sentences & paragraphs to a 1024-dimensional dense vector space and can be used for retrieval.

Model Details

Model Description

Model Type: Sentence Transformer
Base model: intfloat/multilingual-e5-large-instruct
Maximum Sequence Length: 256 tokens
Output Dimensionality: 1024 dimensions
Similarity Function: Cosine Similarity
Supported Modality: Text
Training Dataset:
- clef-me5-mined-pairs-train-pairs

Model Sources

Documentation: Sentence Transformers Documentation
Repository: Sentence Transformers on GitHub
Hugging Face: Sentence Transformers on Hugging Face

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'transformer_task': 'feature-extraction', 'modality_config': {'text': {'method': 'forward', 'method_output_name': 'last_hidden_state'}}, 'module_output_name': 'token_embeddings', 'architecture': 'XLMRobertaModel'})
  (1): Pooling({'embedding_dimension': 1024, 'pooling_mode': 'mean', '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("MinhPhuc0804/me5-256-kiem-tra-di-t1-v2.3-epoch-10")
# Run inference
sentences = [
    'query: @user That’s not what some of the data suggests. 25% of those who had a light infection the first time required an ER visit the 2nd time. And we know there is a mounting load with reinfections.',
    "passage: diagnoses occur closer to the index date for infection or reinfection in the Omicron BA epoch.\n\ntitle: SARS-CoV-2 Reinfection is Preceded by Unique Biomarkers and Related to Initial Infection Timing and Severity: an N3C RECOVER EHR-Based Cohort Study\nWe report lower albumin levels leading up to reinfection and a statistically significant association of severity between first infection and reinfection (chi-squared value: 9446.2, p-value: 0) with a medium effect size (Cramer's V: 0.18, DoF = 4).",
    'passage: randomization to hospital discharge.\n\ntitle: Effect of a Single High Dose of Vitamin D<sub>3</sub> on Hospital Length of Stay in Patients With Moderate to Severe COVID-19\nPrespecified secondary outcomes included mortality during hospitalization; the number of patients admitted to the intensive care unit; the number of patients who required mechanical ventilation and the duration of mechanical ventilation; and serum levels of 25-hydroxyvitamin D, total calcium, creatinine, and C-reactive protein.Of 240 randomized patients, 237 were included in the primary analysis (mean [SD] age, 56.2 [14.4] years; 104 [43.9%] women; mean [SD] baseline 25-hydroxyvitamin D level, 20.9 [9.2] ng/mL). Median (interquartile range) length of stay was not significantly different between the vitamin D3 (7.0 [4.0-10.0] days) and placebo groups (7.0 [5.0-13.0] days) (log-rank P = .59; unadjusted hazard ratio for hospital discharge, 1.07 [95% CI, 0.82-1.39]; P = .62).',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 1024]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities)
# tensor([[1.0000, 0.4792, 0.1923],
#         [0.4792, 1.0000, 0.0620],
#         [0.1923, 0.0620, 1.0000]])

Evaluation

Metrics

Information Retrieval

Dataset: CT26-dev-split
Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.6926
cosine_accuracy@3	0.8588
cosine_accuracy@5	0.9024
cosine_accuracy@10	0.9346
cosine_precision@1	0.6926
cosine_precision@3	0.2863
cosine_precision@5	0.1805
cosine_precision@10	0.0935
cosine_recall@1	0.6926
cosine_recall@3	0.8588
cosine_recall@5	0.9024
cosine_recall@10	0.9346
cosine_ndcg@10	0.8204
cosine_mrr@10	0.7831
cosine_map@100	0.7853

Training Details

Training Dataset

clef-me5-mined-pairs-train-pairs

Dataset: clef-me5-mined-pairs-train-pairs
Size: 18,281 training samples
Columns: anchor and positive
Approximate statistics based on the first 1000 samples:
anchor positive
type string string
details
min: 26 tokens
mean: 59.43 tokens
max: 104 tokens
min: 26 tokens
mean: 190.97 tokens
max: 256 tokens

Samples:

anchor	positive
`query: Peer-reviewed scientific studies project universal mask use would spare nearly 130,000 U.S. lives by February. But Utah carries on with a baffling, lax patchwork of feeble mask rules. If elected, I’ll shield your family. #PetersonProtects #utpol`	passage: 469,578–578,347) lives could be lost to COVID-19 across the United States by 28 February 2021.[object Object][object Object]title: Modeling COVID-19 scenarios for the United States[object Object]We find that achieving universal mask use (95% mask use in public) could be sufficient to ameliorate the worst effects of epidemic resurgences in many states. Universal mask use could save an additional 129,574 (85,284–170,867) lives from September 22, 2020 through the end of February 2021, or an additional 95,814 (60,731–133,077) lives assuming a lesser adoption of mask wearing (85%), when compared to the reference scenario.
`query: "Seroprevalence studies may fail to detect people who have had mild covid-19. Consideration should be given to [...] calibration of assay thresholds, the breadth of the antibody response, and the role of mucosal antibodies"`	passage: was used to analyse the data.[object Object][object Object]title: Are we underestimating seroprevalence of SARS-CoV-2?[object Object][object Object] The study indicates a fairly high knowledge level of STI/HIV (89.4%) with more males (87%) and younger participants (88%) possessing good knowledge of STI/HIV. Majority of the participants are sexually active (63.3%) and of this are more males (61.3%) and younger participants (60%). Findings show that age (β = 0.025; t = 0.04; p > 0.05) of the three predictor variables was not a predictor of attitude towards STI/HIV. However, knowledge of STI/HIV (β = 0.459; t = 5.032; p < 0.05) and sexual behaviour (β = 0.341; t = 4.278; p < 0.05) were strong predictors of attitude towards STI/HIV. [object Object] This study shows the need for strong advocacy, enlightenment and community mobilisation for improved awareness of STI/HIV.
`query: Rebuilding of October #Arctic sea ice volume spanning over the past 100 years... [Side-by-side look between PIOMAS-20C and PIOMAS data sets now refreshed through October 2021. Model details available at`	`passage: title: Arctic Sea Ice Volume Variability over 1901–2010: A Model-Based Reconstruction`
abstract: Abstract PIOMAS-20C, an Arctic sea ice reconstruction for 1901–2010, is produced by forcing the Pan-Arctic Ice Ocean Modeling and Assimilation System (PIOMAS) with ERA-20C atmospheric data. ERA-20C performance over Arctic sea ice is assessed by comparisons with measurements and data from other reanalyses. ERA-20C performs similarly with respect to the annual cycle of downwelling radiation, air temperature, and wind speed compared to reanalyses with more extensive data assimilation such as ERA-Interim and MERRA. PIOMAS-20C sea ice thickness and volume are then compared with in situ and aircraft remote sensing observations for the period of ~1950–2010. Error statistics are similar to those for PIOMAS. We compare the magnitude and patterns of sea ice variability between the first half of the twentieth century (1901–40) and the more recent period (1980–2010), both marked by sea ice decl...

Loss: MultipleNegativesRankingLoss with these parameters:
json
1{ 2 "scale": 20.0, 3 "similarity_fct": "cos_sim", 4 "gather_across_devices": false, 5 "directions": [ 6 "query_to_doc" 7 ], 8 "partition_mode": "joint", 9 "hardness_mode": null, 10 "hardness_strength": 0.0 11}

Evaluation Dataset

clef-me5-mined-pairs-train-pairs

Dataset: clef-me5-mined-pairs-train-pairs
Size: 963 evaluation samples
Columns: anchor and positive
Approximate statistics based on the first 963 samples:
anchor positive
type string string
details
min: 24 tokens
mean: 59.09 tokens
max: 138 tokens
min: 30 tokens
mean: 189.75 tokens
max: 256 tokens

Samples:

anchor	positive
`query: I reckon we’ll be hearing more about plitidepsin, which (in vitro, at least) is 27.5 times stronger than remdesivir #COVID19`	passage: title: Plitidepsin has potent preclinical efficacy against SARS-CoV-2 by targeting the host protein eEF1A[object Object]abstract: Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) viral proteins interact with the eukaryotic translation machinery, and inhibitors of translation have potent antiviral effects. We found that the drug plitidepsin (aplidin), which has limited clinical approval, possesses antiviral activity (90% inhibitory concentration = 0.88 nM) that is more potent than remdesivir against SARS-CoV-2 in vitro by a factor of 27.5, with limited toxicity in cell culture. Through the use of a drug-resistant mutant, we show that the antiviral activity of plitidepsin against SARS-CoV-2 is mediated through inhibition of the known target eEF1A (eukaryotic translation elongation factor 1A).
`query: 2020 research on #LongCovid. Pre‑vaccine era. “Young, low risk patients with ongoing symptoms of #covid19 had signs of damage to multiple organs four months after initially being infected.”`	`passage: title: Long covid: Damage to multiple organs presents in young, low risk patients[object Object]abstract: Young, low risk patients with ongoing symptoms of covid-19 had signs of damage to multiple organs four months after initially being infected, a preprint study has suggested.`
`query: L'inflammation indépendante provoquée par les macrophages encourage-t-elle les lésions alvéolaires dans la COVID-19 ?`	`passage: title: Does autonomous macrophage-driven inflammation promote alveolar damage in COVID-19?[object Object]abstract: [object Object][object Object]`

Loss: MultipleNegativesRankingLoss with these parameters:
json
1{ 2 "scale": 20.0, 3 "similarity_fct": "cos_sim", 4 "gather_across_devices": false, 5 "directions": [ 6 "query_to_doc" 7 ], 8 "partition_mode": "joint", 9 "hardness_mode": null, 10 "hardness_strength": 0.0 11}

Training Hyperparameters

Non-Default Hyperparameters

per_device_train_batch_size: 64
per_device_eval_batch_size: 64
learning_rate: 1.6e-05
num_train_epochs: 10
warmup_ratio: 0.1
fp16: True
dataloader_num_workers: 8
load_best_model_at_end: 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
torch_empty_cache_steps: None
learning_rate: 1.6e-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: 10
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: 8
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}
parallelism_config: 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
hub_revision: None
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
liger_kernel_config: None
eval_use_gather_object: False
average_tokens_across_devices: False
prompts: None
batch_sampler: batch_sampler
multi_dataset_batch_sampler: proportional
router_mapping: {}
learning_rate_mapping: {}

Training Logs

Epoch	Step	Training Loss	Validation Loss	CT26-dev-split_cosine_ndcg@10
1.0	286	-	0.3706	0.8055
1.7483	500	0.6016	-	-
2.0	572	-	0.3239	0.8161
3.0	858	-	0.3250	0.8084
3.4965	1000	0.1923	-	-
4.0	1144	-	0.3174	0.8120
5.0	1430	-	0.3163	0.8129
5.2448	1500	0.1003	-	-
6.0	1716	-	0.3005	0.8148
6.9930	2000	0.0679	-	-
7.0	2002	-	0.2999	0.8226
8.0	2288	-	0.3027	0.8235
8.7413	2500	0.053	-	-
9.0	2574	-	0.3017	0.8186
10.0	2860	-	0.299	0.8204

The bold row denotes the saved checkpoint.

Training Time

Training: 21.4 minutes

Framework Versions

Python: 3.12.6
Sentence Transformers: 5.4.1
Transformers: 4.56.0
PyTorch: 2.8.0+cu129
Accelerate: 1.10.1
Datasets: 4.8.5
Tokenizers: 0.22.0

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{oord2019representationlearningcontrastivepredictive,
      title={Representation Learning with Contrastive Predictive Coding},
      author={Aaron van den Oord and Yazhe Li and Oriol Vinyals},
      year={2019},
      eprint={1807.03748},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/1807.03748},
}

	anchor	positive
type	string	string
details	min: 26 tokens mean: 59.43 tokens max: 104 tokens	min: 26 tokens mean: 190.97 tokens max: 256 tokens