MistyDragon/bge-base-financial-matryoshka

BGE base Financial Matryoshka

This is a sentence-transformers model finetuned from BAAI/bge-small-en-v1.5 on the json dataset. It maps sentences & paragraphs to a 384-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-small-en-v1.5
• Maximum Sequence Length: 512 tokens
• Output Dimensionality: 384 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': 384, '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("MistyDragon/bge-base-financial-matryoshka")
# Run inference
sentences = [
    'On June 7, 2023, the Company and Epic filed petitions with the Circuit Court requesting further review of the decision. On June 30, 2023, the Circuit Court denied both petitions.',
    'When did the Circuit Court deny the petitions filed by Apple and Epic for further review of the decision?',
    'What was the adjusted debt to EBITDAR ratio for the fiscal year ended August 26, 2023?',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 384]

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

Evaluation

Metrics

Information Retrieval

• Dataset: dim_256
• Evaluated with InformationRetrievalEvaluator with these parameters:

  {
      "truncate_dim": 256
  }
  

Metric	Value
cosine_accuracy@1	0.6886
cosine_accuracy@3	0.8143
cosine_accuracy@5	0.8557
cosine_accuracy@10	0.8971
cosine_precision@1	0.6886
cosine_precision@3	0.2714
cosine_precision@5	0.1711
cosine_precision@10	0.0897
cosine_recall@1	0.6886
cosine_recall@3	0.8143
cosine_recall@5	0.8557
cosine_recall@10	0.8971
cosine_ndcg@10	0.7952
cosine_mrr@10	0.7624
cosine_map@100	0.7662

Information Retrieval

• Dataset: dim_128
• Evaluated with InformationRetrievalEvaluator with these parameters:

  {
      "truncate_dim": 128
  }
  

Metric	Value
cosine_accuracy@1	0.6886
cosine_accuracy@3	0.7971
cosine_accuracy@5	0.8443
cosine_accuracy@10	0.8914
cosine_precision@1	0.6886
cosine_precision@3	0.2657
cosine_precision@5	0.1689
cosine_precision@10	0.0891
cosine_recall@1	0.6886
cosine_recall@3	0.7971
cosine_recall@5	0.8443
cosine_recall@10	0.8914
cosine_ndcg@10	0.7889
cosine_mrr@10	0.7563
cosine_map@100	0.76

Information Retrieval

• Dataset: dim_64
• Evaluated with InformationRetrievalEvaluator with these parameters:

  {
      "truncate_dim": 64
  }
  

Metric	Value
cosine_accuracy@1	0.6571
cosine_accuracy@3	0.7786
cosine_accuracy@5	0.8229
cosine_accuracy@10	0.8629
cosine_precision@1	0.6571
cosine_precision@3	0.2595
cosine_precision@5	0.1646
cosine_precision@10	0.0863
cosine_recall@1	0.6571
cosine_recall@3	0.7786
cosine_recall@5	0.8229
cosine_recall@10	0.8629
cosine_ndcg@10	0.7609
cosine_mrr@10	0.7282
cosine_map@100	0.733

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: 4 tokens mean: 45.35 tokens max: 512 tokens	min: 7 tokens mean: 20.38 tokens max: 42 tokens

• Samples:

  positive	anchor
  In a competitive home improvement market, the company focuses on factors such as customer experience, price, quality, product availability, and delivery options to attract and retain customers.	How does the company compete with other businesses in the home improvement market?
  Historically, the majority of revenue from North American and European regions, which experience higher sales of solar products in the second, third, and fourth quarters, have been affected by seasonal customer demand trends.	What seasonal sales trend does the solar sector typically exhibit?
  We expect to have adequate supplies or sources of availability of raw materials necessary to meet our needs; however, there always are risks and uncertainties with respect to the supply of raw materials that could impact availability in sufficient quantities and at cost effective prices to meet our needs.	What are some of the risk factors associated with raw material supplies for a vehicle manufacturer?

• Loss: MatryoshkaLoss with these parameters:

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

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: epoch
• per_device_eval_batch_size: 16
• 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: 16
• 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
• 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_256_cosine_ndcg@10	dim_128_cosine_ndcg@10	dim_64_cosine_ndcg@10
0.1015	10	4.6757	-	-	-
0.2030	20	3.7954	-	-	-
0.3046	30	2.8699	-	-	-
0.4061	40	1.9741	-	-	-
0.5076	50	1.3134	-	-	-
0.6091	60	1.3616	-	-	-
0.7107	70	1.4887	-	-	-
0.8122	80	1.6192	-	-	-
0.9137	90	1.1806	-	-	-
1.0	99	-	0.7859	0.7743	0.7488
1.0102	100	1.2233	-	-	-
1.1117	110	0.8571	-	-	-
1.2132	120	0.9667	-	-	-
1.3147	130	0.9762	-	-	-
1.4162	140	0.8002	-	-	-
1.5178	150	0.7964	-	-	-
1.6193	160	0.9702	-	-	-
1.7208	170	0.653	-	-	-
1.8223	180	0.8967	-	-	-
1.9239	190	0.3674	-	-	-
2.0	198	-	0.7922	0.7882	0.7593
2.0203	200	0.5451	-	-	-
2.1218	210	0.8927	-	-	-
2.2234	220	0.603	-	-	-
2.3249	230	0.5169	-	-	-
2.4264	240	0.7078	-	-	-
2.5279	250	0.67	-	-	-
2.6294	260	0.8605	-	-	-
2.7310	270	0.9597	-	-	-
2.8325	280	0.503	-	-	-
2.9340	290	0.5416	-	-	-
3.0	297	-	0.7936	0.7898	0.7625
3.0305	300	0.5506	-	-	-
3.1320	310	0.7123	-	-	-
3.2335	320	0.5213	-	-	-
3.3350	330	0.8629	-	-	-
3.4365	340	0.683	-	-	-
3.5381	350	0.4859	-	-	-
3.6396	360	0.5015	-	-	-
3.7411	370	0.5414	-	-	-
3.8426	380	0.6515	-	-	-
3.9442	390	0.6247	-	-	-
4.0	396	-	0.7952	0.7889	0.7609

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.11.13
• Sentence Transformers: 4.1.0
• Transformers: 4.52.4
• PyTorch: 2.6.0+cu124
• Accelerate: 1.8.1
• Datasets: 3.6.0
• Tokenizers: 0.21.1

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}
}