sylvain471/bl_ademe_large

test qwen2 Matryoshka

This is a sentence-transformers model finetuned from actualdata/bilingual-embedding-large. It maps sentences & paragraphs to a 1024-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: actualdata/bilingual-embedding-large
• Maximum Sequence Length: 512 tokens
• Output Dimensionality: 1024 tokens
• Similarity Function: Cosine Similarity
• Language: en
• License: apache-2.0

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': 512, 'do_lower_case': False}) with Transformer model: BilingualModel 
  (1): Pooling({'word_embedding_dimension': 1024, '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("sylvain471/bl_ademe_large")
# Run inference
sentences = [
    " Le PCS intègre l'énergie libérée par la condensation de l'eau après la combustion, tandis que le PCI ne l'intègre pas.",
    " Qu'est-ce qui distingue le Pouvoir Calorifique Supérieur (PCS) du Pouvoir Calorifique Inférieur (PCI) ?",
    " La proportion d'énergie utilisée dans l'eau chaude sanitaire pour les résidences principales (métropole uniquement) est-elle supérieure à 1 % ?",
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 1024]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities.shape)
# [3, 3]

Evaluation

Metrics

Information Retrieval

• Dataset: dim_1024
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.3168
cosine_accuracy@3	0.4254
cosine_accuracy@5	0.477
cosine_accuracy@10	0.5562
cosine_precision@1	0.3168
cosine_precision@3	0.1418
cosine_precision@5	0.0954
cosine_precision@10	0.0556
cosine_recall@1	0.3168
cosine_recall@3	0.4254
cosine_recall@5	0.477
cosine_recall@10	0.5562
cosine_ndcg@10	0.4276
cosine_mrr@10	0.3876
cosine_map@100	0.3994

Information Retrieval

• Dataset: dim_896
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.3223
cosine_accuracy@3	0.4236
cosine_accuracy@5	0.4733
cosine_accuracy@10	0.5488
cosine_precision@1	0.3223
cosine_precision@3	0.1412
cosine_precision@5	0.0947
cosine_precision@10	0.0549
cosine_recall@1	0.3223
cosine_recall@3	0.4236
cosine_recall@5	0.4733
cosine_recall@10	0.5488
cosine_ndcg@10	0.4272
cosine_mrr@10	0.3894
cosine_map@100	0.4018

Information Retrieval

• Dataset: dim_512
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.3315
cosine_accuracy@3	0.4236
cosine_accuracy@5	0.4751
cosine_accuracy@10	0.5488
cosine_precision@1	0.3315
cosine_precision@3	0.1412
cosine_precision@5	0.095
cosine_precision@10	0.0549
cosine_recall@1	0.3315
cosine_recall@3	0.4236
cosine_recall@5	0.4751
cosine_recall@10	0.5488
cosine_ndcg@10	0.4309
cosine_mrr@10	0.3943
cosine_map@100	0.4065

Information Retrieval

• Dataset: dim_256
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.3076
cosine_accuracy@3	0.4125
cosine_accuracy@5	0.4678
cosine_accuracy@10	0.5396
cosine_precision@1	0.3076
cosine_precision@3	0.1375
cosine_precision@5	0.0936
cosine_precision@10	0.054
cosine_recall@1	0.3076
cosine_recall@3	0.4125
cosine_recall@5	0.4678
cosine_recall@10	0.5396
cosine_ndcg@10	0.4156
cosine_mrr@10	0.3769
cosine_map@100	0.3896

Information Retrieval

• Dataset: dim_128
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.2965
cosine_accuracy@3	0.4052
cosine_accuracy@5	0.4475
cosine_accuracy@10	0.5396
cosine_precision@1	0.2965
cosine_precision@3	0.1351
cosine_precision@5	0.0895
cosine_precision@10	0.054
cosine_recall@1	0.2965
cosine_recall@3	0.4052
cosine_recall@5	0.4475
cosine_recall@10	0.5396
cosine_ndcg@10	0.4079
cosine_mrr@10	0.3672
cosine_map@100	0.3789

Training Details

Training Dataset

Unnamed Dataset

• Size: 4,885 training samples
• Columns: positive and anchor
• Approximate statistics based on the first 1000 samples:

  	positive	anchor
  type	string	string
  details	min: 3 tokens mean: 32.82 tokens max: 185 tokens	min: 2 tokens mean: 26.77 tokens max: 71 tokens

• Samples:

  positive	anchor
  Lorsque le traitement spécifique par catégorie de déchets produits par la Personne Morale est inconnu, le taux moyen local ou sectoriel de traitement en fin de vie (incinération, mise en décharge, recyclage, compostage, etc.) est utilisé. Le transport est également un paramètre à intégrer au calcul.	Quels sont les paramètres clés par type de traitement à prendre en compte pour réaliser un bilan d'émissions de gaz à effet de serre ?
  Une analyse de cycle de vie fournit un moyen efficace et systémique pour évaluer les impacts environnementaux d’un produit, d’un service, d’une entreprise ou d’un procédé.	Qu'est-ce qu'une évaluation de cycle de vie (ACV) ?
  1 469,2 t CO2e.	Quel est le total des émissions annuelles de l'entreprise GAMMA ?

• Loss: MatryoshkaLoss with these parameters:

  {
      "loss": "MultipleNegativesRankingLoss",
      "matryoshka_dims": [
          1024,
          896,
          512,
          256,
          128
      ],
      "matryoshka_weights": [
          1,
          1,
          1,
          1,
          1
      ],
      "n_dims_per_step": -1
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: epoch
• per_device_train_batch_size: 16
• gradient_accumulation_steps: 8
• learning_rate: 2e-05
• num_train_epochs: 20
• lr_scheduler_type: cosine
• warmup_ratio: 0.1
• bf16: True
• tf32: True
• load_best_model_at_end: True
• optim: adamw_torch_fused
• batch_sampler: no_duplicates

All Hyperparameters

Click to expand

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: epoch
• prediction_loss_only: True
• per_device_train_batch_size: 16
• per_device_eval_batch_size: 8
• per_gpu_train_batch_size: None
• per_gpu_eval_batch_size: None
• gradient_accumulation_steps: 8
• eval_accumulation_steps: None
• torch_empty_cache_steps: None
• learning_rate: 2e-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: 20
• max_steps: -1
• lr_scheduler_type: cosine
• 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: True
• fp16: False
• fp16_opt_level: O1
• half_precision_backend: auto
• bf16_full_eval: False
• fp16_full_eval: False
• tf32: True
• 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_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: False
• hub_always_push: False
• gradient_checkpointing: False
• gradient_checkpointing_kwargs: None
• include_inputs_for_metrics: False
• 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
• eval_use_gather_object: False
• batch_sampler: no_duplicates
• multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	Training Loss	dim_1024_cosine_map@100	dim_128_cosine_map@100	dim_256_cosine_map@100	dim_512_cosine_map@100	dim_896_cosine_map@100
0.2614	10	5.4141	-	-	-	-	-
0.5229	20	4.2823	-	-	-	-	-
0.7843	30	3.0162	-	-	-	-	-
0.9935	38	-	0.3636	0.3170	0.3407	0.3566	0.3668
1.0458	40	2.5846	-	-	-	-	-
1.3072	50	2.2069	-	-	-	-	-
1.5686	60	1.7585	-	-	-	-	-
1.8301	70	1.3099	-	-	-	-	-
1.9869	76	-	0.3979	0.3353	0.3726	0.3895	0.3983
2.0915	80	1.1449	-	-	-	-	-
2.3529	90	1.0137	-	-	-	-	-
2.6144	100	0.6402	-	-	-	-	-
2.8758	110	0.4931	-	-	-	-	-
2.9804	114	-	0.4026	0.3568	0.3808	0.3882	0.3992
3.1373	120	0.4662	-	-	-	-	-
3.3987	130	0.3782	-	-	-	-	-
3.6601	140	0.2696	-	-	-	-	-
3.9216	150	0.2478	-	-	-	-	-
4.0	153	-	0.3805	0.3460	0.3613	0.3680	0.3850
4.1830	160	0.2655	-	-	-	-	-
4.4444	170	0.1952	-	-	-	-	-
4.7059	180	0.1494	-	-	-	-	-
4.9673	190	0.1482	-	-	-	-	-
4.9935	191	-	0.3806	0.3619	0.3702	0.3799	0.3814
5.2288	200	0.161	-	-	-	-	-
5.4902	210	0.1282	-	-	-	-	-
5.7516	220	0.0888	-	-	-	-	-
5.9869	229	-	0.3936	0.3685	0.3758	0.3870	0.3916
6.0131	230	0.1042	-	-	-	-	-
6.2745	240	0.126	-	-	-	-	-
6.5359	250	0.103	-	-	-	-	-
6.7974	260	0.0467	-	-	-	-	-
6.9804	267	-	0.4022	0.3689	0.3897	0.3950	0.4022
7.0588	270	0.0581	-	-	-	-	-
7.3203	280	0.0728	-	-	-	-	-
7.5817	290	0.064	-	-	-	-	-
7.8431	300	0.0271	-	-	-	-	-
8.0	306	-	0.4010	0.3756	0.3872	0.3988	0.4021
8.1046	310	0.0452	-	-	-	-	-
8.3660	320	0.0613	-	-	-	-	-
8.6275	330	0.0294	-	-	-	-	-
8.8889	340	0.0396	-	-	-	-	-
8.9935	344	-	0.3914	0.3722	0.3801	0.3916	0.3939
9.1503	350	0.024	-	-	-	-	-
9.4118	360	0.0253	-	-	-	-	-
9.6732	370	0.017	-	-	-	-	-
9.9346	380	0.0163	-	-	-	-	-
9.9869	382	-	0.3901	0.3660	0.3796	0.3892	0.3904
10.1961	390	0.0191	-	-	-	-	-
10.4575	400	0.017	-	-	-	-	-
10.7190	410	0.0108	-	-	-	-	-
10.9804	420	0.0118	0.3994	0.3789	0.3896	0.4065	0.4018
11.2418	430	0.0111	-	-	-	-	-
11.5033	440	0.011	-	-	-	-	-
11.7647	450	0.0052	-	-	-	-	-
12.0	459	-	0.4030	0.3772	0.3986	0.4034	0.3999
12.0261	460	0.0144	-	-	-	-	-
12.2876	470	0.0068	-	-	-	-	-
12.5490	480	0.0061	-	-	-	-	-
12.8105	490	0.0039	-	-	-	-	-
12.9935	497	-	0.4022	0.3733	0.3869	0.3995	0.3983
13.0719	500	0.0074	-	-	-	-	-
13.3333	510	0.005	-	-	-	-	-
13.5948	520	0.0045	-	-	-	-	-
13.8562	530	0.0035	-	-	-	-	-
13.9869	535	-	0.4027	0.3779	0.3891	0.4015	0.3999
14.1176	540	0.0047	-	-	-	-	-
14.3791	550	0.0043	-	-	-	-	-
14.6405	560	0.0038	-	-	-	-	-
14.9020	570	0.0034	-	-	-	-	-
14.9804	573	-	0.3954	0.3734	0.3875	0.3982	0.3962
15.1634	580	0.0037	-	-	-	-	-
15.4248	590	0.0039	-	-	-	-	-
15.6863	600	0.0034	-	-	-	-	-
15.9477	610	0.0033	-	-	-	-	-
16.0	612	-	0.3966	0.3720	0.3852	0.3948	0.3936
16.2092	620	0.0038	-	-	-	-	-
16.4706	630	0.0034	-	-	-	-	-
16.7320	640	0.0029	-	-	-	-	-
16.9935	650	0.0033	0.3968	0.3723	0.3844	0.3977	0.3966
17.2549	660	0.0034	-	-	-	-	-
17.5163	670	0.0033	-	-	-	-	-
17.7778	680	0.0028	-	-	-	-	-
17.9869	688	-	0.3965	0.3695	0.3861	0.3960	0.3969
18.0392	690	0.0033	-	-	-	-	-
18.3007	700	0.0033	-	-	-	-	-
18.5621	710	0.0036	-	-	-	-	-
18.8235	720	0.0026	-	-	-	-	-
18.9804	726	-	0.3962	0.3701	0.3819	0.3951	0.3964
19.0850	730	0.003	-	-	-	-	-
19.3464	740	0.0036	-	-	-	-	-
19.6078	750	0.0033	-	-	-	-	-
19.8693	760	0.0031	0.3994	0.3789	0.3896	0.4065	0.4018

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.10.12
• Sentence Transformers: 3.0.1
• Transformers: 4.44.2
• PyTorch: 2.4.1+cu121
• Accelerate: 0.34.2
• Datasets: 2.21.0
• Tokenizers: 0.19.1

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",
}

MatryoshkaLoss

@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}
}

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}
}