SentenceTransformer

This model was finetuned with Unsloth.

This is a sentence-transformers model trained on the french triplet ds and french custom triplet ds datasets. It maps sentences & paragraphs to a 384-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
Maximum Sequence Length: 256 tokens
Output Dimensionality: 384 dimensions
Similarity Function: Cosine Similarity
Training Datasets:
- french triplet ds
- french custom triplet ds

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': 256, 'do_lower_case': False, 'architecture': 'BertModel'})
  (1): Pooling({'word_embedding_dimension': 384, '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.

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("thomasavare/all-MiniLM-L6-v2-med-v0")
# Run inference
sentences = [
    'Plan de soins post-opératoires après une chirurgie de kyste osseux anévrismaux',
    'Les exercices de renforcement et de musculation de la hanche ont été commencés tôt, et à la cinquième semaine, on a commencé à marcher avec des béquilles, et quatre semaines plus tard, on a abandonné les béquilles et on a encouragé le patient à marcher de façon autonome.',
    "Le patient a été conseillé de continuer à faire un suivi régulier auprès de son fournisseur de soins de santé primaires et de ses dentistes pour gérer les problèmes postopératoires et s'assurer qu'il n'y a pas de récidive de la maladie.",
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 384]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities)
# tensor([[1.0000, 0.3373, 0.4268],
#         [0.3373, 1.0000, 0.4083],
#         [0.4268, 0.4083, 1.0000]])

Training Details

Training Datasets

french triplet ds

Dataset: french triplet ds
Size: 232,684 training samples
Columns: anchor, positive, and negative

Approximate statistics based on the first 1000 samples:

	anchor	positive	negative
type	string	string	string
details	min: 6 tokens mean: 18.1 tokens max: 42 tokens	min: 9 tokens mean: 55.16 tokens max: 239 tokens	min: 10 tokens mean: 54.65 tokens max: 228 tokens

Samples:

anchor	positive	negative
`masse kystique sur l' utérus`	`Le patient s'est présenté aux urgences avec une douleur abdominale sévère dans le quadrant inférieur gauche, qui a été diagnostiquée comme une masse kystique sur la paroi antérieure gauche de l'utérus.`	`Symptômes: masse pelvienne à croissance rapide et taux sériques accrus de marqueurs tumoraux`
`Plan de soins post-démarrage pour les patients atteints d' une BPAN`	`La patiente sera suivie en consultation externe avec une surveillance étroite de sa nutrition et de ses habitudes de comportement pour s'assurer qu'elle ne revienne pas à ses comportements antérieurs.`	`Le patient a été libéré le 30e jour d'hospitalisation avec de l'aspirine seule.`
`Comment l'état du patient a- t- il réagi au traitement?`	`Le patient a répondu positivement au traitement prescrit par la diéthylcarbamazine et aucun suivi n' est nécessaire.`	`Les symptômes du patient se sont résolus après avoir reçu un traitement et subi des échocardiogrammes de suivi.`

Loss: MultipleNegativesRankingLoss with these parameters:

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

french custom triplet ds

Dataset: french custom triplet ds
Size: 251,939 training samples
Columns: anchor, positive, and negative

Approximate statistics based on the first 1000 samples:

	anchor	positive	negative
type	string	string	string
details	min: 3 tokens mean: 16.31 tokens max: 74 tokens	min: 3 tokens mean: 16.44 tokens max: 72 tokens	min: 3 tokens mean: 15.43 tokens max: 61 tokens

Samples:

anchor	positive	negative
`Cholera`	`Maladie infectieuse`	`Mittelschmerz`
`Choléra`	`Cholera`	`Astringents et détergents locaux`
`Maladie infectieuse`	`Choléra`	`collision avec tout objet, fixe ou mobile ou en mouvement`

Loss: MultipleNegativesRankingLoss with these parameters:

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

Training Hyperparameters

Non-Default Hyperparameters

per_device_train_batch_size: 512
learning_rate: 2e-05
warmup_steps: 0.05
bf16: True
project: icd10-embeddings
trackio_space_id: thomasavare/icd10-embeddings
warmup_ratio: 0.05
prompts: {'anchor': 'Instruct : Represent the disease in a standardized clinical concept\nQuery :', 'positive': 'Instruct : Represent the disease in a standardized clinical concept\nQuery :', 'negative': 'Instruct : Represent the disease in a standardized clinical concept\nQuery :'}
batch_sampler: no_duplicates

All Hyperparameters

Click to expand

per_device_train_batch_size: 512
num_train_epochs: 3
max_steps: -1
learning_rate: 2e-05
lr_scheduler_type: linear
lr_scheduler_kwargs: None
warmup_steps: 0.05
optim: adamw_torch_fused
optim_args: None
weight_decay: 0.0
adam_beta1: 0.9
adam_beta2: 0.999
adam_epsilon: 1e-08
optim_target_modules: None
gradient_accumulation_steps: 1
average_tokens_across_devices: True
max_grad_norm: 1.0
label_smoothing_factor: 0.0
bf16: True
fp16: False
bf16_full_eval: False
fp16_full_eval: False
tf32: None
gradient_checkpointing: False
gradient_checkpointing_kwargs: None
torch_compile: False
torch_compile_backend: None
torch_compile_mode: None
use_liger_kernel: False
liger_kernel_config: None
use_cache: False
neftune_noise_alpha: None
torch_empty_cache_steps: None
auto_find_batch_size: False
log_on_each_node: True
logging_nan_inf_filter: True
include_num_input_tokens_seen: no
log_level: passive
log_level_replica: warning
disable_tqdm: False
project: icd10-embeddings
trackio_space_id: thomasavare/icd10-embeddings
eval_strategy: no
per_device_eval_batch_size: 8
prediction_loss_only: True
eval_on_start: False
eval_do_concat_batches: True
eval_use_gather_object: False
eval_accumulation_steps: None
include_for_metrics: []
batch_eval_metrics: False
save_only_model: False
save_on_each_node: False
enable_jit_checkpoint: False
push_to_hub: False
hub_private_repo: None
hub_model_id: None
hub_strategy: every_save
hub_always_push: False
hub_revision: None
load_best_model_at_end: False
ignore_data_skip: False
restore_callback_states_from_checkpoint: False
full_determinism: False
seed: 42
data_seed: None
use_cpu: 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
dataloader_drop_last: False
dataloader_num_workers: 0
dataloader_pin_memory: True
dataloader_persistent_workers: False
dataloader_prefetch_factor: None
remove_unused_columns: True
label_names: None
train_sampling_strategy: random
length_column_name: length
ddp_find_unused_parameters: None
ddp_bucket_cap_mb: None
ddp_broadcast_buffers: False
ddp_backend: None
ddp_timeout: 1800
fsdp: []
fsdp_config: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}
deepspeed: None
debug: []
skip_memory_metrics: True
do_predict: False
resume_from_checkpoint: None
warmup_ratio: 0.05
local_rank: -1
prompts: {'anchor': 'Instruct : Represent the disease in a standardized clinical concept\nQuery :', 'positive': 'Instruct : Represent the disease in a standardized clinical concept\nQuery :', 'negative': 'Instruct : Represent the disease in a standardized clinical concept\nQuery :'}
batch_sampler: no_duplicates
multi_dataset_batch_sampler: proportional
router_mapping: {}
learning_rate_mapping: {}

Training Logs

Epoch	Step	Training Loss
0.1055	100	5.7088
0.2110	200	4.9171
0.3165	300	4.8061
0.4219	400	4.7313
0.5274	500	4.7004
0.6329	600	4.6518
0.7384	700	4.6307
0.8439	800	4.6992
0.9494	900	4.5778
1.0549	1000	4.4946
1.1603	1100	4.6055
1.2658	1200	4.5647
1.3713	1300	4.5245
1.4768	1400	4.5631
1.5823	1500	4.5186
1.6878	1600	4.5509
1.7932	1700	4.5756
1.8987	1800	4.6112
2.0042	1900	4.4410
2.1097	2000	4.6082
2.2152	2100	4.5329
2.3207	2200	4.5414
2.4262	2300	4.5330
2.5316	2400	4.5384
2.6371	2500	4.5075
2.7426	2600	4.5156
2.8481	2700	4.5750
2.9536	2800	4.6071

Framework Versions

Python: 3.12.12
Sentence Transformers: 5.2.3
Transformers: 5.2.0
PyTorch: 2.10.0+cu128
Accelerate: 1.13.0
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}
}

Downloads last month: 4

Safetensors

Model size

22.7M params

Tensor type

BF16

Papers for thomasavare/all-MiniLM-L6-v2-med-v0-merged

Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks

Paper • 1908.10084 • Published Aug 27, 2019 • 12

Efficient Natural Language Response Suggestion for Smart Reply

Paper • 1705.00652 • Published May 1, 2017