RK-1235/bge-base-FIR-matryoshka-BASELINE-10epochs-FT

BGE base Financial Matryoshka

This is a sentence-transformers model finetuned from BAAI/bge-base-en 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
• 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("RK-1235/bge-base-FIR-matryoshka-BASELINE-10epochs-FT")
# Run inference
sentences = [
    'Item 8. Financial Statements and Supplementary Data. The Consolidated Financial Statements, together with the Notes thereto and the report thereon dated February 16, 2024, of PricewaterhouseCoopers LLP, the Firm’s independent registered public accounting firm (PCAOB ID 238).',
    'What type of data does Item 8 in a financial document contain?',
    "How did the assumptions and estimates used for assessing the fair value of reporting units potentially impact the company's financial statements?",
]
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 with these parameters:

  {
      "truncate_dim": 768
  }
  

Metric	Value
cosine_accuracy@1	0.2041
cosine_accuracy@3	0.3908
cosine_accuracy@5	0.4557
cosine_accuracy@10	0.5427
cosine_precision@1	0.2041
cosine_precision@3	0.1303
cosine_precision@5	0.0911
cosine_precision@10	0.0543
cosine_recall@1	0.2041
cosine_recall@3	0.3908
cosine_recall@5	0.4557
cosine_recall@10	0.5427
cosine_ndcg@10	0.3713
cosine_mrr@10	0.3167
cosine_map@100	0.3257

Information Retrieval

• Dataset: dim_512
• Evaluated with InformationRetrievalEvaluator with these parameters:

  {
      "truncate_dim": 512
  }
  

Metric	Value
cosine_accuracy@1	0.1788
cosine_accuracy@3	0.3845
cosine_accuracy@5	0.4494
cosine_accuracy@10	0.5222
cosine_precision@1	0.1788
cosine_precision@3	0.1282
cosine_precision@5	0.0899
cosine_precision@10	0.0522
cosine_recall@1	0.1788
cosine_recall@3	0.3845
cosine_recall@5	0.4494
cosine_recall@10	0.5222
cosine_ndcg@10	0.3521
cosine_mrr@10	0.2975
cosine_map@100	0.3072

Information Retrieval

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

  {
      "truncate_dim": 256
  }
  

Metric	Value
cosine_accuracy@1	0.1756
cosine_accuracy@3	0.3386
cosine_accuracy@5	0.3924
cosine_accuracy@10	0.4968
cosine_precision@1	0.1756
cosine_precision@3	0.1129
cosine_precision@5	0.0785
cosine_precision@10	0.0497
cosine_recall@1	0.1756
cosine_recall@3	0.3386
cosine_recall@5	0.3924
cosine_recall@10	0.4968
cosine_ndcg@10	0.3278
cosine_mrr@10	0.2749
cosine_map@100	0.284

Information Retrieval

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

  {
      "truncate_dim": 128
  }
  

Metric	Value
cosine_accuracy@1	0.1345
cosine_accuracy@3	0.2769
cosine_accuracy@5	0.3434
cosine_accuracy@10	0.4019
cosine_precision@1	0.1345
cosine_precision@3	0.0923
cosine_precision@5	0.0687
cosine_precision@10	0.0402
cosine_recall@1	0.1345
cosine_recall@3	0.2769
cosine_recall@5	0.3434
cosine_recall@10	0.4019
cosine_ndcg@10	0.2643
cosine_mrr@10	0.2206
cosine_map@100	0.2315

Information Retrieval

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

  {
      "truncate_dim": 64
  }
  

Metric	Value
cosine_accuracy@1	0.0854
cosine_accuracy@3	0.1946
cosine_accuracy@5	0.2484
cosine_accuracy@10	0.3165
cosine_precision@1	0.0854
cosine_precision@3	0.0649
cosine_precision@5	0.0497
cosine_precision@10	0.0316
cosine_recall@1	0.0854
cosine_recall@3	0.1946
cosine_recall@5	0.2484
cosine_recall@10	0.3165
cosine_ndcg@10	0.1936
cosine_mrr@10	0.1553
cosine_map@100	0.1641

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: 6 tokens mean: 46.06 tokens max: 371 tokens	min: 8 tokens mean: 20.8 tokens max: 51 tokens

• Samples:

  positive	anchor
  As of December 31, 2023, a 5 percent change in the contingent consideration liabilities would result in a change in income before income taxes of $5.2 million.	How would a 5% change in the contingent consideration liabilities impact income before taxes as of December 31, 2023?
  NIKE, Inc.'s principal business activity involves the design, development, and worldwide marketing and selling of athletic footwear, apparel, equipment, accessories, and services.	What is the principal business activity of NIKE, Inc.?
  During 2023, changes in foreign currencies relative to the U.S. dollar negatively impacted net sales by approximately $3,484, 156 basis points, compared to 2022, attributable to our Canadian and Other International operations.	What was the overall impact of foreign currencies on net sales in 2023?

• 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: 10
• 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
• 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: 10
• 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: 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_768_cosine_ndcg@10	dim_512_cosine_ndcg@10	dim_256_cosine_ndcg@10	dim_128_cosine_ndcg@10	dim_64_cosine_ndcg@10
0.8122	10	89.0763	-	-	-	-	-
1.0	13	-	0.4022	0.3835	0.3505	0.2911	0.1835
1.5685	20	36.7538	-	-	-	-	-
2.0	26	-	0.3725	0.3591	0.3218	0.2753	0.1978
2.3249	30	17.7869	-	-	-	-	-
3.0	39	-	0.3680	0.3558	0.3284	0.2638	0.2000
3.0812	40	10.5904	-	-	-	-	-
3.8934	50	7.9568	-	-	-	-	-
4.0	52	-	0.3634	0.3487	0.3245	0.2589	0.1999
4.6497	60	5.5002	-	-	-	-	-
5.0	65	-	0.3648	0.3551	0.3211	0.2595	0.1968
5.4061	70	5.3314	-	-	-	-	-
6.0	78	-	0.3693	0.3548	0.3257	0.2621	0.1977
6.1624	80	4.6165	-	-	-	-	-
6.9746	90	4.7811	-	-	-	-	-
7.0	91	-	0.3698	0.3532	0.3293	0.2637	0.1954
7.7310	100	3.978	-	-	-	-	-
8.0	104	-	0.3713	0.3523	0.3273	0.2637	0.1952
8.4873	110	4.1624	-	-	-	-	-
9.0	117	-	0.3707	0.3517	0.3264	0.2639	0.1949
9.2437	120	3.4956	-	-	-	-	-
10.0	130	3.9661	0.3713	0.3521	0.3278	0.2643	0.1936

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.10.12
• Sentence Transformers: 4.1.0
• Transformers: 4.52.2
• PyTorch: 2.6.0+cu124
• Accelerate: 1.7.0
• 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}
}