rbhatia46/bge-base-financial-nvidia-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("rbhatia46/bge-base-financial-nvidia-matryoshka")
# Run inference
sentences = [
    'The design of the Annual Report, with the consolidated financial statements placed immediately after Part IV, enhances the integration of financial data by maintaining a coherent structure.',
    'How does the structure of the Annual Report on Form 10-K facilitate the integration of the consolidated financial statements?',
    'Where can one find the Glossary of Terms and Acronyms in Item 8?',
]
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	0.6957
cosine_accuracy@3	0.8171
cosine_accuracy@5	0.8629
cosine_accuracy@10	0.9
cosine_precision@1	0.6957
cosine_precision@3	0.2724
cosine_precision@5	0.1726
cosine_precision@10	0.09
cosine_recall@1	0.6957
cosine_recall@3	0.8171
cosine_recall@5	0.8629
cosine_recall@10	0.9
cosine_ndcg@10	0.7971
cosine_mrr@10	0.7642
cosine_map@100	0.7682

Information Retrieval

• Dataset: dim_512
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.6943
cosine_accuracy@3	0.81
cosine_accuracy@5	0.8514
cosine_accuracy@10	0.9
cosine_precision@1	0.6943
cosine_precision@3	0.27
cosine_precision@5	0.1703
cosine_precision@10	0.09
cosine_recall@1	0.6943
cosine_recall@3	0.81
cosine_recall@5	0.8514
cosine_recall@10	0.9
cosine_ndcg@10	0.7951
cosine_mrr@10	0.7618
cosine_map@100	0.7658

Information Retrieval

• Dataset: dim_256
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.7014
cosine_accuracy@3	0.7971
cosine_accuracy@5	0.85
cosine_accuracy@10	0.8886
cosine_precision@1	0.7014
cosine_precision@3	0.2657
cosine_precision@5	0.17
cosine_precision@10	0.0889
cosine_recall@1	0.7014
cosine_recall@3	0.7971
cosine_recall@5	0.85
cosine_recall@10	0.8886
cosine_ndcg@10	0.7933
cosine_mrr@10	0.763
cosine_map@100	0.7678

Information Retrieval

• Dataset: dim_128
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.6957
cosine_accuracy@3	0.8014
cosine_accuracy@5	0.8357
cosine_accuracy@10	0.8843
cosine_precision@1	0.6957
cosine_precision@3	0.2671
cosine_precision@5	0.1671
cosine_precision@10	0.0884
cosine_recall@1	0.6957
cosine_recall@3	0.8014
cosine_recall@5	0.8357
cosine_recall@10	0.8843
cosine_ndcg@10	0.7874
cosine_mrr@10	0.7567
cosine_map@100	0.7614

Information Retrieval

• Dataset: dim_64
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.6571
cosine_accuracy@3	0.7871
cosine_accuracy@5	0.8286
cosine_accuracy@10	0.8757
cosine_precision@1	0.6571
cosine_precision@3	0.2624
cosine_precision@5	0.1657
cosine_precision@10	0.0876
cosine_recall@1	0.6571
cosine_recall@3	0.7871
cosine_recall@5	0.8286
cosine_recall@10	0.8757
cosine_ndcg@10	0.7656
cosine_mrr@10	0.7304
cosine_map@100	0.735

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: 6 tokens mean: 45.53 tokens max: 222 tokens	min: 8 tokens mean: 20.3 tokens max: 45 tokens

• Samples:

  positive	anchor
  Acquisition activity used cash of $765 million in 2023, primarily related to a Beauty acquisition.	How much cash did acquisition activities use in 2023?
  In a financial report, Part IV Item 15 includes Exhibits and Financial Statement Schedules as mentioned.	What content can be expected under Part IV Item 15 in a financial report?
  we had more than 8.3 million fiber consumer wireline broadband customers, adding 1.1 million during the year.	How many fiber consumer wireline broadband customers did the company have at the end of the year?

• 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	1.5751	-	-	-	-	-
0.9746	12	-	-	-	-	-	0.7580
0.8122	10	0.6362	-	-	-	-	-
0.9746	12	-	0.7503	0.7576	0.7653	0.7282	0.7638
1.6244	20	0.4426	-	-	-	-	-
1.9492	24	-	0.7544	0.7662	0.7640	0.7311	0.7676
2.4365	30	0.3217	-	-	-	-	-
2.9239	36	-	0.7608	0.7684	0.7662	0.7341	0.7686
3.2487	40	0.2761	-	-	-	-	-
3.8985	48	-	0.7614	0.7678	0.7658	0.735	0.7682

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.10.6
• Sentence Transformers: 3.0.1
• Transformers: 4.41.2
• PyTorch: 2.1.2+cu121
• Accelerate: 0.31.0
• 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}
}