This is a sentence-transformers model finetuned from sentence-transformers/paraphrase-multilingual-mpnet-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: sentence-transformers/paraphrase-multilingual-mpnet-base-v2
Maximum Sequence Length: 128 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': 128, '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})
)

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("Fatoumataa/embedding_billingue_francais_bambara1")
# Run inference
sentences = [
    'expliquez le rôle du contrat social selon john locke.',
    "pour locke, le contrat social est un accord par lequel les individus sortent de l'état de nature pour protéger leurs droits naturels et établir un gouvernement limité. il fonde la société civile.",
    'protection des droits naturels,le contrat social sert à rendre le peuple entièrement soumis à un souverain absolu. (locke défend un gouvernement limité, pas une soumission absolue comme hobbes).',
]
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: bambara_eval
Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.875
cosine_accuracy@3	0.965
cosine_accuracy@5	0.98
cosine_accuracy@10	0.985
cosine_precision@3	0.3217
cosine_precision@5	0.196
cosine_precision@10	0.0985
cosine_recall@3	0.965
cosine_recall@5	0.98
cosine_recall@10	0.985
cosine_ndcg@10	0.937
cosine_mrr@10	0.9208
cosine_map@100	0.9218

Information Retrieval

Dataset: bambara_eval
Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.755
cosine_accuracy@3	0.95
cosine_accuracy@5	0.965
cosine_accuracy@10	0.98
cosine_precision@3	0.3167
cosine_precision@5	0.193
cosine_precision@10	0.098
cosine_recall@3	0.95
cosine_recall@5	0.965
cosine_recall@10	0.98
cosine_ndcg@10	0.8822
cosine_mrr@10	0.8492
cosine_map@100	0.8503

Training Details

Training Dataset

Unnamed Dataset

Size: 3,225 training samples
Columns: anchor, positive, negative, and label

Approximate statistics based on the first 1000 samples:

	anchor	positive	negative	label
type	string	string	string	float
details	min: 6 tokens mean: 20.85 tokens max: 77 tokens	min: 7 tokens mean: 59.01 tokens max: 128 tokens	min: 11 tokens mean: 51.72 tokens max: 128 tokens	min: 1.0 mean: 1.0 max: 1.0

Samples:

anchor	positive	negative	label
`quel instrument de musique traditionnel malien est une sorte de harpe-luth à 21 cordes, emblématique des griots ?`	`kora ye mali laadala fɔlifɛn ye min bɛ fɔ ni juru 21 ye, griotw bɛ baara kɛ ni a ye kosɛbɛ walasa ka fara u ka maanaw ni dɔnkiliw kan.`	`le balafon est la harpe-luth à 21 cordes. (confusion avec un xylophone africain).`	`1.0`
`fɛn jumɛnw tun bɛ mali mansamara ka jago siratigɛ la?`	`les principaux produits échangés dans l'mansaya du mali étaient l'or, le sel, et les esclaves, complétés par des produits locaux comme le kola, le tissu et les céréales.`	`victor hugo tun ye faransi kanuya sɛbɛnnikɛla ŋana ye. 'les misérables' ye a ka baara tɔgɔba dɔ ye, gafe min bɛ sigida ka tɔɔrɔ jira.`	`1.0`
`calcule le double de 9.`	`18. sabula 9 siɲɛ fila ye 18 ye.`	`24. erreur fréquente dans le calcul de double.`	`1.0`

Loss: MultipleNegativesRankingLoss with these parameters:
json
1{ 2 "scale": 20.0, 3 "similarity_fct": "cos_sim" 4}

Evaluation Dataset

Unnamed Dataset

Size: 817 evaluation samples
Columns: anchor, positive, negative, and label

Approximate statistics based on the first 817 samples:

	anchor	positive	negative	label
type	string	string	string	float
details	min: 5 tokens mean: 20.91 tokens max: 55 tokens	min: 8 tokens mean: 58.13 tokens max: 128 tokens	min: 11 tokens mean: 51.09 tokens max: 128 tokens	min: 1.0 mean: 1.0 max: 1.0

Samples:

anchor	positive	negative	label
`kɔrɔfɔli bɛ dɛmɛ jumɛn don seko ni dɔnko faamuyali la ?`	`la critique d'art offre des perspectives d'analyse, éclaire le contexte, et propose des interprétations, aidant ainsi le public à approfondir sa compréhension et son appréciation des œuvres.`	`tifoyidi bɛ sɔrɔ kosɛbɛ ji walima dumuni dunni fɛ min nɔgɔlen don banakisɛ bɛ mɔgɔ min na, o ka banakɔtaa fɛ. banakunbɛnni ye ji minta, bolo saniya ani dumuni tobili ye.`	`1.0`
`polifoni ye mun ye fɔli la?`	`la polyphonie est une technique de composition où plusieurs mélodies indépendantes sont jouées simultanément. elle crée une richesse harmonique.`	`mali la, waati jiginni bɛ kɛ sababu ye ka ja waati janw ni sanjiw kɛ minnu tɛ kɛ tuma bɛɛ. o bɛ suman tigɛ dɔgɔya ani ka sɛnɛ bila farati la. a ka gɛlɛn sɛnɛkɛlaw ma.`	`1.0`
`quelles sont les habitudes à adopter pour améliorer la qualité de son sommeil ?`	`walasa ka sunɔgɔ ɲuman sɔrɔ, aw ka kan ka sunɔgɔcogo dɔ sigi sen kan tuma bɛɛ, ka aw yɛrɛ tanga ɛkranw ma sani aw ka sunɔgɔ, ka dibi ni lafiya sigida dilan, ani ka farikoloɲɛnajɛ kɛ tile fɛ. aw bɛ aw yɛrɛ tanga kafeyin ma wula fɛ, o fana nafa ka bon.`	`manger une grande quantité de nourriture juste avant de dormir. (cela peut perturber la digestion et le sommeil, c'est une mauvaise habitude, erreur conceptuelle).`	`1.0`

Loss: MultipleNegativesRankingLoss with these parameters:
json
1{ 2 "scale": 20.0, 3 "similarity_fct": "cos_sim" 4}

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: steps
per_device_train_batch_size: 16
per_device_eval_batch_size: 32
learning_rate: 1e-05
weight_decay: 0.01
num_train_epochs: 8
warmup_ratio: 0.1
fp16: True
load_best_model_at_end: True

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
eval_strategy: steps
prediction_loss_only: True
per_device_train_batch_size: 16
per_device_eval_batch_size: 32
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: 1e-05
weight_decay: 0.01
adam_beta1: 0.9
adam_beta2: 0.999
adam_epsilon: 1e-08
max_grad_norm: 1.0
num_train_epochs: 8
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: 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
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

Training Logs

Epoch	Step	Training Loss	Validation Loss	bambara_eval_cosine_ndcg@10
0.2475	50	1.0849	-	-
0.4950	100	0.5428	0.5113	0.8522
0.7426	150	0.2423	-	-
0.9901	200	0.2538	0.2678	0.8872
1.2376	250	0.1232	-	-
1.4851	300	0.103	0.2104	0.9165
1.7327	350	0.0991	-	-
1.9802	400	0.1284	0.1878	0.9181
2.2277	450	0.0427	-	-
2.4752	500	0.0311	0.1282	0.9205
2.7228	550	0.0514	-	-
2.9703	600	0.0778	0.1101	0.9370
3.2178	650	0.0233	-	-
3.4653	700	0.0298	0.1158	0.9408
3.7129	750	0.0298	-	-
3.9604	800	0.0163	0.1134	0.9366
4.2079	850	0.0238	-	-
4.4554	900	0.0155	0.1225	0.9387
4.7030	950	0.0219	-	-
4.9505	1000	0.0095	0.1049	0.9376
-1	-1	-	-	0.9370
0.2475	50	3.2372	-	-
0.4950	100	2.0213	2.1683	0.4056
0.7426	150	1.3029	-	-
0.9901	200	0.844	0.9643	0.5978
1.2376	250	0.5074	-	-
1.4851	300	0.3533	0.5925	0.7219
1.7327	350	0.3124	-	-
1.9802	400	0.243	0.4317	0.8044
2.2277	450	0.143	-	-
2.4752	500	0.1499	0.3547	0.8294
2.7228	550	0.1366	-	-
2.9703	600	0.1302	0.2870	0.8478
3.2178	650	0.0999	-	-
3.4653	700	0.0805	0.2736	0.8403
3.7129	750	0.0606	-	-
3.9604	800	0.073	0.2396	0.8666
4.2079	850	0.0515	-	-
4.4554	900	0.0451	0.2285	0.8709
4.7030	950	0.0591	-	-
4.9505	1000	0.05	0.2258	0.8707
5.1980	1050	0.0435	-	-
5.4455	1100	0.059	0.2192	0.8722
5.6931	1150	0.0496	-	-
5.9406	1200	0.0408	0.2132	0.8724
6.1881	1250	0.0491	-	-
6.4356	1300	0.038	0.2060	0.8783
6.6832	1350	0.032	-	-
6.9307	1400	0.0385	0.1998	0.8822
7.1782	1450	0.0363	-	-
7.4257	1500	0.0349	0.1982	0.8750
7.6733	1550	0.038	-	-
7.9208	1600	0.0523	0.1944	0.8753
-1	-1	-	-	0.8822

The bold row denotes the saved checkpoint.

Framework Versions

Python: 3.11.13
Sentence Transformers: 4.1.0
Transformers: 4.53.3
PyTorch: 2.6.0+cu124
Accelerate: 1.9.0
Datasets: 4.0.0
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",
}

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}
}