Downloads · 30 days
67
1% of all-time downloads
KarBik/legal-french-matroshka
legal-french-matroshka is a sentence similarity model from KarBik. Use it when you need a score for how close two texts are. It is set up for sentence-transformers.
This is a sentence-transformers model finetuned from intfloat/multilingual-e5-base on the json dataset. It maps sentences & paragraphs to a 768-dimensional dense vector space and can be used for semantic textual simil…
Downloads · 30 days
67
1% of all-time downloads
All-time downloads
6.7K
Public
Parameters
278M
1.1 GB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors1.1 GB · 98%
From the Hugging Face model README
This is a sentence-transformers model finetuned from intfloat/multilingual-e5-base on the json 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.
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()
)
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("KarBik/legal-french-matroshka")
# Run inference
sentences = [
'Les projets de marchés de partenariat conclus pour le compte des acheteurs non autorisés sont instruits par le ministre de tutelle.',
"Qui est responsable de l'instruction des projets de marchés de partenariat conclus pour le compte des acheteurs non autorisés ?",
"Dans quelle situation l'assuré a-t-il besoin d'être assisté ou représenté par un avocat en raison de la défense de la partie adverse ?",
]
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]
<!--
### Direct Usage (Transformers)
<details><summary>Click to see the direct usage in Transformers</summary>
</details>
-->
<!--
### Downstream Usage (Sentence Transformers)
You can finetune this model on your own dataset.
<details><summary>Click to expand</summary>
</details>
-->
<!--
### Out-of-Scope Use
*List how the model may foreseeably be misused and address what users ought not to do with the model.*
-->
dim_768| Metric | Value |
|---|---|
| cosine_accuracy@1 | 0.94 |
| cosine_accuracy@3 | 0.981 |
| cosine_accuracy@5 | 0.987 |
| cosine_accuracy@10 | 0.989 |
| cosine_precision@1 | 0.94 |
| cosine_precision@3 | 0.327 |
| cosine_precision@5 | 0.1974 |
| cosine_precision@10 | 0.0989 |
| cosine_recall@1 | 0.94 |
| cosine_recall@3 | 0.981 |
| cosine_recall@5 | 0.987 |
| cosine_recall@10 | 0.989 |
| cosine_ndcg@10 | 0.9684 |
| cosine_mrr@10 | 0.9614 |
| cosine_map@100 | 0.9617 |
dim_512| Metric | Value |
|---|---|
| cosine_accuracy@1 | 0.942 |
| cosine_accuracy@3 | 0.982 |
| cosine_accuracy@5 | 0.988 |
| cosine_accuracy@10 | 0.989 |
| cosine_precision@1 | 0.942 |
| cosine_precision@3 | 0.3273 |
| cosine_precision@5 | 0.1976 |
| cosine_precision@10 | 0.0989 |
| cosine_recall@1 | 0.942 |
| cosine_recall@3 | 0.982 |
| cosine_recall@5 | 0.988 |
| cosine_recall@10 | 0.989 |
| cosine_ndcg@10 | 0.9696 |
| cosine_mrr@10 | 0.9629 |
| cosine_map@100 | 0.9633 |
dim_256| Metric | Value |
|---|---|
| cosine_accuracy@1 | 0.937 |
| cosine_accuracy@3 | 0.98 |
| cosine_accuracy@5 | 0.985 |
| cosine_accuracy@10 | 0.989 |
| cosine_precision@1 | 0.937 |
| cosine_precision@3 | 0.3267 |
| cosine_precision@5 | 0.197 |
| cosine_precision@10 | 0.0989 |
| cosine_recall@1 | 0.937 |
| cosine_recall@3 | 0.98 |
| cosine_recall@5 | 0.985 |
| cosine_recall@10 | 0.989 |
| cosine_ndcg@10 | 0.9662 |
| cosine_mrr@10 | 0.9585 |
| cosine_map@100 | 0.9588 |
dim_128| Metric | Value |
|---|---|
| cosine_accuracy@1 | 0.93 |
| cosine_accuracy@3 | 0.972 |
| cosine_accuracy@5 | 0.983 |
| cosine_accuracy@10 | 0.988 |
| cosine_precision@1 | 0.93 |
| cosine_precision@3 | 0.324 |
| cosine_precision@5 | 0.1966 |
| cosine_precision@10 | 0.0988 |
| cosine_recall@1 | 0.93 |
| cosine_recall@3 | 0.972 |
| cosine_recall@5 | 0.983 |
| cosine_recall@10 | 0.988 |
| cosine_ndcg@10 | 0.9619 |
| cosine_mrr@10 | 0.9533 |
| cosine_map@100 | 0.9537 |
dim_64| Metric | Value |
|---|---|
| cosine_accuracy@1 | 0.901 |
| cosine_accuracy@3 | 0.966 |
| cosine_accuracy@5 | 0.977 |
| cosine_accuracy@10 | 0.989 |
| cosine_precision@1 | 0.901 |
| cosine_precision@3 | 0.322 |
| cosine_precision@5 | 0.1954 |
| cosine_precision@10 | 0.0989 |
| cosine_recall@1 | 0.901 |
| cosine_recall@3 | 0.966 |
| cosine_recall@5 | 0.977 |
| cosine_recall@10 | 0.989 |
| cosine_ndcg@10 | 0.9478 |
| cosine_mrr@10 | 0.9342 |
| cosine_map@100 | 0.9346 |
| positive | anchor | |
|---|---|---|
| type | string | string |
| details | <ul><li>min: 11 tokens</li><li>mean: 141.81 tokens</li><li>max: 512 tokens</li></ul> | <ul><li>min: 11 tokens</li><li>mean: 57.29 tokens</li><li>max: 262 tokens</li></ul> |
| positive | anchor |
|---|---|
| <code>Sauf dispositions contraires des conventions internationales, l'émission de titres comportant la mention visée à l'article 51 entraîne l'obligation pour l'organisme émetteur d'opérer, sur les produits de ces titres et pendant toute la durée de ceux-ci, la retenue à la source édictée par le 1 de l'article 119 bis du code général des impôts. Le montant de cette retenue doit être versé au comptable désigné par l'administration, dans les conditions et suivant les modalités fixées par le 1 de l'article 1672 et l'article 1673 dudit code.</code> | <code>Quelle est l'obligation de l'organisme émetteur concernant les produits de titres émis avec la mention visée à l'article 51, et comment doit-il opérer la retenue à la source pendant la durée de ces titres ?</code> |
| <code>Lorsque l'allocation est attribuée en application du troisième alinéa de l'article L. 232-12 et du cinquième alinéa de l'article L. 232-14 , le montant forfaitaire attribué est, respectivement, égal, à domicile, à 50 % du montant du plafond mentionné à l'article L. 232-3-1 correspondant au degré de perte d'autonomie le plus important, et, en établissement, à 50 % du tarif afférent à la dépendance de l'établissement considéré applicable aux résidents classés dans les groupes iso-ressources 1 et 2. Cette avance s'impute sur les montants de l'allocation personnalisée d'autonomie versée ultérieurement.</code> | <code>Quel est le montant forfaitaire attribué lorsqu'une allocation est octroyée en application du troisième alinéa de l'article L. 232-12 et du cinquième alinéa de l'article L. 232-14, selon que l'allocation est perçue à domicile ou en établissement ?</code> |
| <code>La taxe devient exigible au moment où le poids lourd : 1° Entre sur le réseau, si la condition mentionnée au 1° de l'article L. 421-202 est remplie ; 2° Franchit un point de la section de tarification déterminé par l'autorité compétente, si cette même condition n'est pas remplie.</code> | <code>Quel est le moment où la taxe devient exigible pour un poids lourd en fonction de son entrée dans le réseau ou de son franchissement d'un point de tarification déterminé ?</code> |
{
"loss": "MultipleNegativesRankingLoss",
"matryoshka_dims": [
768,
512,
256,
128,
64
],
"matryoshka_weights": [
1,
1,
1,
1,
1
],
"n_dims_per_step": -1
}
eval_strategy: epochper_device_train_batch_size: 32per_device_eval_batch_size: 16gradient_accumulation_steps: 16learning_rate: 2e-05num_train_epochs: 4lr_scheduler_type: cosinewarmup_ratio: 0.1bf16: Truetf32: Trueload_best_model_at_end: Trueoptim: adamw_torch_fusedbatch_sampler: no_duplicatesoverwrite_output_dir: Falsedo_predict: Falseeval_strategy: epochprediction_loss_only: Trueper_device_train_batch_size: 32per_device_eval_batch_size: 16per_gpu_train_batch_size: Noneper_gpu_eval_batch_size: Nonegradient_accumulation_steps: 16eval_accumulation_steps: Nonelearning_rate: 2e-05weight_decay: 0.0adam_beta1: 0.9adam_beta2: 0.999adam_epsilon: 1e-08max_grad_norm: 1.0num_train_epochs: 4max_steps: -1lr_scheduler_type: cosinelr_scheduler_kwargs: {}warmup_ratio: 0.1warmup_steps: 0log_level: passivelog_level_replica: warninglog_on_each_node: Truelogging_nan_inf_filter: Truesave_safetensors: Truesave_on_each_node: Falsesave_only_model: Falserestore_callback_states_from_checkpoint: Falseno_cuda: Falseuse_cpu: Falseuse_mps_device: Falseseed: 42data_seed: Nonejit_mode_eval: Falseuse_ipex: Falsebf16: Truefp16: Falsefp16_opt_level: O1half_precision_backend: autobf16_full_eval: Falsefp16_full_eval: Falsetf32: Truelocal_rank: 0ddp_backend: Nonetpu_num_cores: Nonetpu_metrics_debug: Falsedebug: []dataloader_drop_last: Falsedataloader_num_workers: 0dataloader_prefetch_factor: Nonepast_index: -1disable_tqdm: Falseremove_unused_columns: Truelabel_names: Noneload_best_model_at_end: Trueignore_data_skip: Falsefsdp: []fsdp_min_num_params: 0fsdp_config: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}fsdp_transformer_layer_cls_to_wrap: Noneaccelerator_config: {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}deepspeed: Nonelabel_smoothing_factor: 0.0optim: adamw_torch_fusedoptim_args: Noneadafactor: Falsegroup_by_length: Falselength_column_name: lengthddp_find_unused_parameters: Noneddp_bucket_cap_mb: Noneddp_broadcast_buffers: Falsedataloader_pin_memory: Truedataloader_persistent_workers: Falseskip_memory_metrics: Trueuse_legacy_prediction_loop: Falsepush_to_hub: Falseresume_from_checkpoint: Nonehub_model_id: Nonehub_strategy: every_savehub_private_repo: Falsehub_always_push: Falsegradient_checkpointing: Falsegradient_checkpointing_kwargs: Noneinclude_inputs_for_metrics: Falseeval_do_concat_batches: Truefp16_backend: autopush_to_hub_model_id: Nonepush_to_hub_organization: Nonemp_parameters:auto_find_batch_size: Falsefull_determinism: Falsetorchdynamo: Noneray_scope: lastddp_timeout: 1800torch_compile: Falsetorch_compile_backend: Nonetorch_compile_mode: Nonedispatch_batches: Nonesplit_batches: Noneinclude_tokens_per_second: Falseinclude_num_input_tokens_seen: Falseneftune_noise_alpha: Noneoptim_target_modules: Nonebatch_eval_metrics: Falsebatch_sampler: no_duplicatesmulti_dataset_batch_sampler: proportional| Epoch | Step | Training Loss | dim_128_cosine_map@100 | dim_256_cosine_map@100 | dim_512_cosine_map@100 | dim_64_cosine_map@100 | dim_768_cosine_map@100 |
|---|---|---|---|---|---|---|---|
| 0 | 0 | - | 0.8447 | 0.9084 | 0.9190 | 0.6362 | 0.9236 |
| 0.5674 | 10 | 5.322 | - | - | - | - | - |
| 0.9645 | 17 | - | 0.9353 | 0.9413 | 0.9488 | 0.9197 | 0.9453 |
| 1.1348 | 20 | 0.3395 | - | - | - | - | - |
| 1.7021 | 30 | 0.0929 | - | - | - | - | - |
| 1.9858 | 35 | - | 0.9517 | 0.9571 | 0.9631 | 0.9357 | 0.9625 |
| 2.2695 | 40 | 0.0408 | - | - | - | - | - |
| 2.8369 | 50 | 0.0264 | - | - | - | - | - |
| 2.9504 | 52 | - | 0.9513 | 0.9579 | 0.9634 | 0.9357 | 0.9620 |
| 3.4043 | 60 | 0.0209 | - | - | - | - | - |
| 3.8582 | 68 | - | 0.9537 | 0.9588 | 0.9633 | 0.9346 | 0.9617 |
@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",
}
@misc{kusupati2024matryoshka,
title={Matryoshka Representation Learning},
author={Aditya Kusupati and Gantavya Bhatt and Aniket Rege and Matthew Wallingford and Aditya Sinha and Vivek Ramanujan and William Howard-Snyder and Kaifeng Chen and Sham Kakade and Prateek Jain and Ali Farhadi},
year={2024},
eprint={2205.13147},
archivePrefix={arXiv},
primaryClass={cs.LG}
}
@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}
}
<!--
## Glossary
*Clearly define terms in order to be accessible across audiences.*
-->
<!--
## Model Card Authors
*Lists the people who create the model card, providing recognition and accountability for the detailed work that goes into its construction.*
-->
<!--
## Model Card Contact
*Provides a way for people who have updates to the Model Card, suggestions, or questions, to contact the Model Card authors.*
-->