MarekMarik/bge-base-financial-matryoshka

BGE base Financial Matryoshka

This is a sentence-transformers model finetuned from BAAI/bge-base-en-v1.5 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.

Model Details

Model Description

• Model Type: Sentence Transformer
• Base model: BAAI/bge-base-en-v1.5
• Maximum Sequence Length: 512 tokens
• Output Dimensionality: 768 dimensions
• Similarity Function: Cosine Similarity
• Training Dataset:
  • json
• 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': True}) with Transformer model: BertModel 
  (1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': True, 'pooling_mode_mean_tokens': False, '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("MarekMarik/bge-base-financial-matryoshka")
# Run inference
sentences = [
    'Certain vendors have been impacted by volatility in the supply chain financing market.',
    'How have certain vendors been impacted in the supply chain financing market?',
    "What was the total value of the company's cash commitments as of December 31, 2023?",
]
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

• Datasets: dim_768, dim_512, dim_256, dim_128 and dim_64
• Evaluated with InformationRetrievalEvaluator

Metric	dim_768	dim_512	dim_256	dim_128	dim_64
cosine_accuracy@1	0.6871	0.6829	0.68	0.6586	0.6357
cosine_accuracy@3	0.8171	0.8114	0.8071	0.7943	0.7629
cosine_accuracy@5	0.8486	0.8529	0.8486	0.83	0.8143
cosine_accuracy@10	0.9086	0.9086	0.8957	0.8843	0.87
cosine_precision@1	0.6871	0.6829	0.68	0.6586	0.6357
cosine_precision@3	0.2724	0.2705	0.269	0.2648	0.2543
cosine_precision@5	0.1697	0.1706	0.1697	0.166	0.1629
cosine_precision@10	0.0909	0.0909	0.0896	0.0884	0.087
cosine_recall@1	0.6871	0.6829	0.68	0.6586	0.6357
cosine_recall@3	0.8171	0.8114	0.8071	0.7943	0.7629
cosine_recall@5	0.8486	0.8529	0.8486	0.83	0.8143
cosine_recall@10	0.9086	0.9086	0.8957	0.8843	0.87
cosine_ndcg@10	0.796	0.7937	0.7883	0.7728	0.7501
cosine_mrr@10	0.7604	0.7572	0.754	0.737	0.7121
cosine_map@100	0.7641	0.7609	0.7583	0.7419	0.7171

Training Details

Training Dataset

json

• Dataset: json
• Size: 6,300 training samples
• Columns: positive and anchor
• Approximate statistics based on the first 1000 samples:

  	positive	anchor
  type	string	string
  details	min: 8 tokens mean: 45.84 tokens max: 439 tokens	min: 7 tokens mean: 20.62 tokens max: 42 tokens

• Samples:

  positive	anchor
  We adopted SAB 121 during fiscal 2022, with no impact on our consolidated financial statements.	What accounting guidance did the company adopt in fiscal 2022 and what was its impact on the consolidated financial statements?
  Mortgage Solutions revenue decreased 18% in 2023 compared to 2022, due to significantly lower mortgage credit inquiry volumes in 2023 compared to the prior year.	What caused the 18% decline in Mortgage Solutions revenue in 2023 compared to 2022?
  Adoption of SBTi goals would build on our current science-based goals to reduce Scope 1 and 2 carbon emissions by 2.1% per year, to achieve a 40% reduction by the end of fiscal 2030 and a 50% reduction by the end of fiscal 2035.	What is the company's percentage target for reducing Scope 1 and 2 carbon emissions by end of fiscal 2035?

• Loss: MatryoshkaLoss with these parameters:

  {
      "loss": "MultipleNegativesRankingLoss",
      "matryoshka_dims": [
          768,
          512,
          256,
          128,
          64
      ],
      "matryoshka_weights": [
          1,
          1,
          1,
          1,
          1
      ],
      "n_dims_per_step": -1
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: epoch
• per_device_eval_batch_size: 4
• gradient_accumulation_steps: 8
• learning_rate: 2e-05
• num_train_epochs: 4
• lr_scheduler_type: cosine
• warmup_ratio: 0.1
• bf16: True
• tf32: False
• 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: 8
• per_device_eval_batch_size: 4
• 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: 4
• 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: False
• 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: 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: no_duplicates
• multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	Training Loss	dim_768_cosine_ndcg@10	dim_512_cosine_ndcg@10	dim_256_cosine_ndcg@10	dim_128_cosine_ndcg@10	dim_64_cosine_ndcg@10
0.1015	10	0.614	-	-	-	-	-
0.2030	20	0.5098	-	-	-	-	-
0.3046	30	0.426	-	-	-	-	-
0.4061	40	0.3262	-	-	-	-	-
0.5076	50	0.2131	-	-	-	-	-
0.6091	60	0.1892	-	-	-	-	-
0.7107	70	0.3049	-	-	-	-	-
0.8122	80	0.1617	-	-	-	-	-
0.9137	90	0.1214	-	-	-	-	-
1.0	99	-	0.7895	0.7919	0.7800	0.7685	0.7361
1.0102	100	0.147	-	-	-	-	-
1.1117	110	0.0938	-	-	-	-	-
1.2132	120	0.1406	-	-	-	-	-
1.3147	130	0.1058	-	-	-	-	-
1.4162	140	0.1072	-	-	-	-	-
1.5178	150	0.0352	-	-	-	-	-
1.6193	160	0.0568	-	-	-	-	-
1.7208	170	0.1283	-	-	-	-	-
1.8223	180	0.066	-	-	-	-	-
1.9239	190	0.038	-	-	-	-	-
2.0	198	-	0.7945	0.7945	0.7860	0.7736	0.7462
2.0203	200	0.0544	-	-	-	-	-
2.1218	210	0.0333	-	-	-	-	-
2.2234	220	0.042	-	-	-	-	-
2.3249	230	0.0489	-	-	-	-	-
2.4264	240	0.0498	-	-	-	-	-
2.5279	250	0.0119	-	-	-	-	-
2.6294	260	0.0273	-	-	-	-	-
2.7310	270	0.0719	-	-	-	-	-
2.8325	280	0.0366	-	-	-	-	-
2.9340	290	0.0333	-	-	-	-	-
3.0	297	-	0.7927	0.7952	0.7881	0.7743	0.7477
3.0305	300	0.0193	-	-	-	-	-
3.1320	310	0.0254	-	-	-	-	-
3.2335	320	0.0252	-	-	-	-	-
3.3350	330	0.039	-	-	-	-	-
3.4365	340	0.0224	-	-	-	-	-
3.5381	350	0.0091	-	-	-	-	-
3.6396	360	0.0356	-	-	-	-	-
3.7411	370	0.042	-	-	-	-	-
3.8426	380	0.038	-	-	-	-	-
3.9442	390	0.0088	-	-	-	-	-
3.9645	392	-	0.7960	0.7937	0.7883	0.7728	0.7501

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.12.8
• Sentence Transformers: 3.3.1
• Transformers: 4.47.1
• PyTorch: 2.5.1+cu124
• Accelerate: 1.2.1
• Datasets: 3.2.0
• 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",
}

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}
}