Yohhei/bge-base-financial-matryoshka

BGE base Financial Matryoshka

This is a sentence-transformers model finetuned from BAAI/bge-base-en-v1.5. 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 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': 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("Yohhei/bge-base-financial-matryoshka")
# Run inference
sentences = [
    'The consolidated financial statements are incorporated by reference in the Annual Report on Form 10-K, indicating they are treated as part of the document for legal and reporting purposes.',
    'What does it mean for financial statements to be incorporated by reference?',
    'What is contained within the pages 163-309 of the financial section?',
]
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

• Dataset: dim_768
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.7014
cosine_accuracy@3	0.8271
cosine_accuracy@5	0.8714
cosine_accuracy@10	0.9029
cosine_precision@1	0.7014
cosine_precision@3	0.2757
cosine_precision@5	0.1743
cosine_precision@10	0.0903
cosine_recall@1	0.7014
cosine_recall@3	0.8271
cosine_recall@5	0.8714
cosine_recall@10	0.9029
cosine_ndcg@10	0.8043
cosine_mrr@10	0.7725
cosine_map@100	0.7766

Information Retrieval

• Dataset: dim_512
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.7
cosine_accuracy@3	0.8329
cosine_accuracy@5	0.8686
cosine_accuracy@10	0.9043
cosine_precision@1	0.7
cosine_precision@3	0.2776
cosine_precision@5	0.1737
cosine_precision@10	0.0904
cosine_recall@1	0.7
cosine_recall@3	0.8329
cosine_recall@5	0.8686
cosine_recall@10	0.9043
cosine_ndcg@10	0.8041
cosine_mrr@10	0.7718
cosine_map@100	0.7757

Information Retrieval

• Dataset: dim_256
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.7
cosine_accuracy@3	0.8214
cosine_accuracy@5	0.8557
cosine_accuracy@10	0.89
cosine_precision@1	0.7
cosine_precision@3	0.2738
cosine_precision@5	0.1711
cosine_precision@10	0.089
cosine_recall@1	0.7
cosine_recall@3	0.8214
cosine_recall@5	0.8557
cosine_recall@10	0.89
cosine_ndcg@10	0.7977
cosine_mrr@10	0.7678
cosine_map@100	0.7727

Information Retrieval

• Dataset: dim_128
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.6786
cosine_accuracy@3	0.8257
cosine_accuracy@5	0.8529
cosine_accuracy@10	0.8857
cosine_precision@1	0.6786
cosine_precision@3	0.2752
cosine_precision@5	0.1706
cosine_precision@10	0.0886
cosine_recall@1	0.6786
cosine_recall@3	0.8257
cosine_recall@5	0.8529
cosine_recall@10	0.8857
cosine_ndcg@10	0.7864
cosine_mrr@10	0.7541
cosine_map@100	0.7586

Information Retrieval

• Dataset: dim_64
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.6643
cosine_accuracy@3	0.7829
cosine_accuracy@5	0.8157
cosine_accuracy@10	0.8643
cosine_precision@1	0.6643
cosine_precision@3	0.261
cosine_precision@5	0.1631
cosine_precision@10	0.0864
cosine_recall@1	0.6643
cosine_recall@3	0.7829
cosine_recall@5	0.8157
cosine_recall@10	0.8643
cosine_ndcg@10	0.7635
cosine_mrr@10	0.7314
cosine_map@100	0.7361

Training Details

Training Dataset

Unnamed Dataset

• 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: 45.16 tokens max: 512 tokens	min: 7 tokens mean: 20.44 tokens max: 45 tokens

• Samples:

  positive	anchor
  Highlights during fiscal year 2023 include the following: We generated $18,085 million of cash from operations.	What was the amount of cash generated from operations by the company in fiscal year 2023?
  U.S. government and agency securities	$
  For assets under development, assets are grouped and assessed for impairment by estimating the undiscounted cash flows, which include remaining construction costs, over the asset's remaining useful life. If cash flows do not exceed the carrying amount, impairment based on fair value versus carrying value is considered.	How is the impairment of assets assessed for projects still under development?

• 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_train_batch_size: 32
• per_device_eval_batch_size: 16
• gradient_accumulation_steps: 16
• learning_rate: 2e-05
• num_train_epochs: 4
• 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: 32
• per_device_eval_batch_size: 16
• per_gpu_train_batch_size: None
• per_gpu_eval_batch_size: None
• gradient_accumulation_steps: 16
• eval_accumulation_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: 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
• batch_sampler: no_duplicates
• multi_dataset_batch_sampler: proportional

Training Logs

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
0.8122	10	1.5313	-	-	-	-	-
0.9746	12	-	0.7416	0.7521	0.7554	0.7079	0.7609
1.6244	20	0.6553	-	-	-	-	-
1.9492	24	-	0.7549	0.7693	0.7732	0.7318	0.7716
2.4365	30	0.445	-	-	-	-	-
2.9239	36	-	0.7565	0.7738	0.7746	0.7367	0.7763
3.2487	40	0.3917	-	-	-	-	-
3.8985	48	-	0.7586	0.7727	0.7757	0.7361	0.7766

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.8.10
• Sentence Transformers: 3.0.1
• Transformers: 4.41.2
• PyTorch: 2.1.2+cu121
• Accelerate: 0.32.0
• Datasets: 2.19.1
• 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}
}