SentenceTransformer based on intfloat/multilingual-e5-base

This is a sentence-transformers model finetuned from intfloat/multilingual-e5-base on the grag-go-idf-only-pos dataset. 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 Sources

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.

from sentence_transformers import SentenceTransformer

# Download from the ๐Ÿค— Hub
model = SentenceTransformer("debug")
# Run inference
sentences = [
    'The weather is lovely today.',
    "It's so sunny outside!",
    'He drove to the stadium.',
]
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

Semantic Similarity

Metric Value
pearson_cosine nan
spearman_cosine nan

Binary Classification

Metric Value
cosine_accuracy 0.8
cosine_accuracy_threshold 0.8185
cosine_f1 0.8889
cosine_f1_threshold 0.8185
cosine_precision 1.0
cosine_recall 0.8
cosine_ap 1.0
cosine_mcc 0.0

Training Details

Training Dataset

grag-go-idf-only-pos

  • Dataset: grag-go-idf-only-pos at 9743952
  • Size: 5,302 training samples
  • Columns: sentence1, sentence2, and label
  • Approximate statistics based on the first 1000 samples:
    sentence1 sentence2 label
    type string string int
    details
    • min: 142 tokens
    • mean: 260.2 tokens
    • max: 340 tokens
    • min: 32 tokens
    • mean: 37.2 tokens
    • max: 44 tokens
    • 1: 100.00%
  • Samples:
    sentence1 sentence2 label
    Procรฉdures et dรฉmarches: Le dรฉpรดt des demandes de subvention se fait en ligne sur la plateforme rรฉgionale mesdemarches.iledefrance.fr : Session de dรฉpรดt unique pour les nouvelles demandes : du 30 septembre au 4 novembre 2024 (11 heures) pour des festivals qui se dรฉroulent entre le 1er mars 2025 et le 28 fรฉvrier 2026 (vote ร  la CP de mars 2025). Pour les demandes de renouvellement, un mail est envoyรฉ aux structures concernรฉes par le service du Spectacle vivant en amont de chaque session de dรฉpรดt.
    Bรฉnรฉficiaires: Professionnel - Culture, Association - Fondation, Association - Rรฉgie par la loi de 1901, Association - ONG, Collectivitรฉ ou institution - Communes de 10 000 ร  20 000 hab, Collectivitรฉ ou institution - Autre (GIP, copropriรฉtรฉ, EPA...), Collectivitรฉ ou institution - Communes de 2000 ร  10 000 hab, Collectivitรฉ ou institution - Communes de < 2000 hab, Collectivitรฉ ou institution - Communes de > 20 000 hab, Collectivitรฉ ou institution - Dรฉpartement, Collectivitรฉ ou institution - EPC...
    Collectivitรฉ ou institution - EPCI --- PEUT_Bร‰Nร‰FICIER ---> demandes de subvention 1
    Type de project: Dans le cadre de sa stratรฉgie ยซโ€ฏImpact 2028โ€ฏยป, la Rรฉgion sโ€™engage dans la dรฉfense de la souverainetรฉ industrielle en renforรงant son soutien ร  une industrie circulaire et dรฉcarbonรฉe, porteuse dโ€™innovations et crรฉatrice dโ€™emplois. PM'up Jeunes pousses industrielles soutient les projets dโ€™implantation dโ€™une premiรจre usine tournรฉe vers la dรฉcarbonation, lโ€™efficacitรฉ รฉnergรฉtique et la circularitรฉ des processus de production. Ces projets peuvent prendre l'une de ces formes : Une premiรจre unitรฉ de production industrielle, aprรจs une phase de prototypage,Une ligne pilote de production industrielle, en interne ou chez un tiers situรฉ en รŽle-de-France, ร  condition que sa production soit destinรฉe ร  de premiรจres commercialisations,La transformation dโ€™une unitรฉ de production pilote ร  une unitรฉ de production industrielle Rรฉgion รŽle-de-France --- soutient ---> industrie dรฉcarbonรฉe 1
    Type de project: Lโ€™excรจs de prรฉcipitations tout au long de lโ€™annรฉe a conduit ร  une chute spectaculaire des rendements des cรฉrรฉales dโ€™รฉtรฉ et des protรฉagineux (blรฉ, orge, pois, fรฉverole, etc.) que produisent 90% des agriculteurs dโ€™รŽle-de-France, historique grenier ร  blรฉ du pays. Tributaires naturels du fleurissement des cultures, les apiculteurs professionnels de la rรฉgion ont รฉgalement souffert de ces dรฉrรจglements climatiques.La Rรฉgion accompagne les exploitations concernรฉes en leur apportant une aide exceptionnelle. excรจs de prรฉcipitations --- DIMINUE ---> rendements des protรฉagineux 1
  • Loss: MultipleNegativesRankingLoss with these parameters:
    {
        "scale": 20.0,
        "similarity_fct": "cos_sim"
    }
    

Evaluation Dataset

grag-go-idf-only-pos

  • Dataset: grag-go-idf-only-pos at 9743952
  • Size: 1,325 evaluation samples
  • Columns: sentence1, sentence2, and label
  • Approximate statistics based on the first 1000 samples:
    sentence1 sentence2 label
    type string string int
    details
    • min: 31 tokens
    • mean: 86.2 tokens
    • max: 160 tokens
    • min: 25 tokens
    • mean: 28.6 tokens
    • max: 33 tokens
    • 1: 100.00%
  • Samples:
    sentence1 sentence2 label
    Date de dรฉbut: non prรฉcisรฉe
    Date de fin (clรดture): non prรฉcisรฉe
    Date de dรฉbut de la future campagne: non prรฉcisรฉe
    Date de fin --- EST ---> non prรฉcisรฉe 1
    Prรฉcision sure les bรฉnรฉficiaires: Communes,ร‰tablissements publics de coopรฉration intercommunale (avec ou sans fiscalitรฉ propre),ร‰tablissements publics territoriaux franciliens,Dรฉpartements,Amรฉnageurs publics et privรฉs (lorsque ces derniers interviennent ร  la demande ou pour le compte d'une collectivitรฉ prรฉcitรฉe). Amรฉnageurs privรฉs --- INTERVIENT_POUR ---> Dรฉpartements 1
    Type de project: Le programme propose des rencontres le samedi aprรจs-midi dans une universitรฉ ou une grande รฉcole rรฉputรฉe, entre les professionnels bรฉnรฉvoles et les lycรฉens et collรฉgiens sous la forme d'atelier thรฉmatiques. Ces moments de rencontre touchent ร  une grande multitude de domaines dโ€™activitรฉs. L'objectif est de donner lโ€™opportunitรฉ aux jeunes les plus enclavรฉs dโ€™รฉchanger avec des intervenants professionnels aux parcours atypiques et inspirants. Les intervenants suscitent les ambitions et รฉlargissent les perspectives des รฉlรจves. rencontres --- impliquent ---> professionnels bรฉnรฉvoles 1
  • Loss: MultipleNegativesRankingLoss with these parameters:
    {
        "scale": 20.0,
        "similarity_fct": "cos_sim"
    }
    

Training Hyperparameters

Non-Default Hyperparameters

  • eval_strategy: epoch
  • per_device_train_batch_size: 2
  • per_device_eval_batch_size: 2
  • num_train_epochs: 1
  • use_cpu: True
  • dataloader_pin_memory: False

All Hyperparameters

Click to expand
  • overwrite_output_dir: False
  • do_predict: False
  • eval_strategy: epoch
  • prediction_loss_only: True
  • per_device_train_batch_size: 2
  • per_device_eval_batch_size: 2
  • 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: 5e-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: 1
  • max_steps: -1
  • lr_scheduler_type: linear
  • lr_scheduler_kwargs: {}
  • warmup_ratio: 0.0
  • 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: True
  • use_mps_device: False
  • seed: 42
  • data_seed: None
  • jit_mode_eval: False
  • use_ipex: False
  • bf16: False
  • 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: 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: False
  • 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
  • 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
  • eval_on_start: False
  • use_liger_kernel: False
  • 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 EmbeddingSimEval_spearman_cosine BinaryClassifEval_cosine_ap
0.6667 2 0.6092 - - -
1.0 3 - 0.2053 nan 1.0

Framework Versions

  • Python: 3.11.11
  • Sentence Transformers: 3.4.1
  • Transformers: 4.48.3
  • PyTorch: 2.6.0+cpu
  • Accelerate: 1.4.0
  • Datasets: 3.3.1
  • Tokenizers: 0.21.0

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
6
Safetensors
Model size
278M params
Tensor type
F32
ยท
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for romain125/debug

Finetuned
(48)
this model

Evaluation results