metadata

language:
  - en
license: apache-2.0
tags:
  - sentence-transformers
  - sentence-similarity
  - feature-extraction
  - generated_from_trainer
  - dataset_size:6300
  - loss:MatryoshkaLoss
  - loss:MultipleNegativesRankingLoss
base_model: BAAI/bge-base-en-v1.5
widget:
  - source_sentence: >-
      We enter into forward currency contracts in order to hedge a portion of
      the foreign currency exposure associated with the translation of our net
      investment in our Canadian subsidiary.
    sentences:
      - >-
        How much did Delta Air Lines spend on debt and finance lease obligations
        in 2023?
      - >-
        What mechanisms does the company use to hedge foreign currency exposure
        for its Canadian subsidiary?
      - >-
        How did operating overhead expenses change for NIKE from fiscal 2022 to
        fiscal 2023?
  - source_sentence: >-
      We calculate return on invested hat capital (ROIC) by dividing adjusted
      ROIC operating profit for the prior four quarters by the average invested
      capital.
    sentences:
      - >-
        What was the fair value of U.S. government and agency securities as of
        June 30, 2022?
      - How is the Return on Invested Capital (ROIC) calculated?
      - >-
        What business outcomes is HPE focused on accelerating with its
        technological solutions?
  - source_sentence: >-
      Expenses from our comparable owned and leased hotels increased $137
      million, on a currency neutral basis, as a result of increased occupancy
      and cost inflation both driving higher labor costs, utilities and other
      operating expenses, as well as an increase in rent expense.
    sentences:
      - >-
        How did the expenses from comparable owned and leased hotels change and
        what were the contributing factors?
      - >-
        What do environmental laws require from suppliers in terms of
        operations?
      - >-
        What energy management technologies does the Enphase bidirectional EV
        charger integrate with?
  - source_sentence: >-
      The Advancing Agility & Automation Initiative at The Hershey Company is
      projected to result in total pre-tax costs of $200,000 to $250,000 from
      inception through 2026. This includes costs for program office execution
      and third-party costs supporting the design and implementation of the new
      organizational structure, as well as implementation and technology
      capability costs and employee severance and related separation benefits.
    sentences:
      - What was the total amortization expense for The Hershey Company in 2021?
      - >-
        How much did net cash used in financing activities decrease in fiscal
        2023 compared to the previous fiscal year?
      - >-
        What is the total projected pre-tax cost of The Hershey Company's
        Advancing Agility & Automation Initiative through 2026?
  - source_sentence: >-
      Structural costs typically do not have a directly proportionate
      relationship to production volume and include costs such as manufacturing,
      engineering, and administrative expenses. These costs can be adjusted over
      time in response to external factors.
    sentences:
      - >-
        How does Ford Motor Company handle its structural costs in relation to
        production volume changes?
      - >-
        What were the total future minimum lease payments under all
        non-cancelable operating leases for the company as of December 31, 2023?
      - >-
        What guidelines does the FASB provide for the measurement of fair value
        when quoted prices are not available?
pipeline_tag: sentence-similarity
library_name: sentence-transformers
metrics:
  - cosine_accuracy@1
  - cosine_accuracy@3
  - cosine_accuracy@5
  - cosine_accuracy@10
  - cosine_precision@1
  - cosine_precision@3
  - cosine_precision@5
  - cosine_precision@10
  - cosine_recall@1
  - cosine_recall@3
  - cosine_recall@5
  - cosine_recall@10
  - cosine_ndcg@10
  - cosine_mrr@10
  - cosine_map@100
model-index:
  - name: BGE base Financial Matryoshka
    results:
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: dim 768
          type: dim_768
        metrics:
          - type: cosine_accuracy@1
            value: 0.72
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.8257142857142857
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.8585714285714285
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.8942857142857142
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.72
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.2752380952380953
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.1717142857142857
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.08942857142857143
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.72
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.8257142857142857
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.8585714285714285
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.8942857142857142
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.8077694527772951
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.7800079365079364
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.7837848752496734
            name: Cosine Map@100
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: dim 512
          type: dim_512
        metrics:
          - type: cosine_accuracy@1
            value: 0.7157142857142857
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.8242857142857143
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.8642857142857143
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.8914285714285715
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.7157142857142857
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.2747619047619047
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.17285714285714285
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.08914285714285713
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.7157142857142857
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.8242857142857143
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.8642857142857143
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.8914285714285715
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.805259563189015
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.7773735827664396
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.7813006780341183
            name: Cosine Map@100
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: dim 256
          type: dim_256
        metrics:
          - type: cosine_accuracy@1
            value: 0.7028571428571428
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.8171428571428572
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.8542857142857143
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.8814285714285715
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.7028571428571428
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.2723809523809524
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.17085714285714285
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.08814285714285712
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.7028571428571428
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.8171428571428572
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.8542857142857143
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.8814285714285715
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.7945503213768784
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.7664075963718817
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.7709929668571353
            name: Cosine Map@100
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: dim 128
          type: dim_128
        metrics:
          - type: cosine_accuracy@1
            value: 0.6785714285714286
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.8028571428571428
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.8542857142857143
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.8814285714285715
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.6785714285714286
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.26761904761904765
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.17085714285714285
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.08814285714285712
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.6785714285714286
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.8028571428571428
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.8542857142857143
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.8814285714285715
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.7829387132685872
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.7509529478458048
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.7549309056916426
            name: Cosine Map@100
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: dim 64
          type: dim_64
        metrics:
          - type: cosine_accuracy@1
            value: 0.6485714285714286
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.77
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.8142857142857143
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.8657142857142858
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.6485714285714286
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.2566666666666667
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.16285714285714287
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.08657142857142856
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.6485714285714286
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.77
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.8142857142857143
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.8657142857142858
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.755512484642688
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.7203905895691608
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.7247515061294347
            name: Cosine Map@100

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("aired/bge-base-financial-matryoshka")
# Run inference
sentences = [
    'Structural costs typically do not have a directly proportionate relationship to production volume and include costs such as manufacturing, engineering, and administrative expenses. These costs can be adjusted over time in response to external factors.',
    'How does Ford Motor Company handle its structural costs in relation to production volume changes?',
    'What were the total future minimum lease payments under all non-cancelable operating leases for the company 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	0.72	0.7157	0.7029	0.6786	0.6486
cosine_accuracy@3	0.8257	0.8243	0.8171	0.8029	0.77
cosine_accuracy@5	0.8586	0.8643	0.8543	0.8543	0.8143
cosine_accuracy@10	0.8943	0.8914	0.8814	0.8814	0.8657
cosine_precision@1	0.72	0.7157	0.7029	0.6786	0.6486
cosine_precision@3	0.2752	0.2748	0.2724	0.2676	0.2567
cosine_precision@5	0.1717	0.1729	0.1709	0.1709	0.1629
cosine_precision@10	0.0894	0.0891	0.0881	0.0881	0.0866
cosine_recall@1	0.72	0.7157	0.7029	0.6786	0.6486
cosine_recall@3	0.8257	0.8243	0.8171	0.8029	0.77
cosine_recall@5	0.8586	0.8643	0.8543	0.8543	0.8143
cosine_recall@10	0.8943	0.8914	0.8814	0.8814	0.8657
cosine_ndcg@10	0.8078	0.8053	0.7946	0.7829	0.7555
cosine_mrr@10	0.78	0.7774	0.7664	0.751	0.7204
cosine_map@100	0.7838	0.7813	0.771	0.7549	0.7248

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: 9 tokens
mean: 45.81 tokens
max: 512 tokens

min: 7 tokens
mean: 20.45 tokens
max: 42 tokens

	positive	anchor
type	string	string
details	min: 9 tokens mean: 45.81 tokens max: 512 tokens	min: 7 tokens mean: 20.45 tokens max: 42 tokens

Samples:

positive	anchor
`GEICO markets its policies mainly by direct response methods where most customers apply for coverage directly to the company via the Internet or over the telephone.`	`What are the primary marketing methods used by GEICO?`
`In addition, most group health plans and issuers of group or individual health insurance coverage are required to disclose personalized pricing information to their participants, beneficiaries, and enrollees through an online consumer tool, by phone, or in paper form, upon request. Cost estimates must be provided in real-time based on cost-sharing information that is accurate at the time of the request.`	`What are the requirements for health insurers and group health plans in providing cost estimates to consumers?`
`Fair values of indefinite-lived intangible assets are determined based on the income approach.`	`What method is used to determine the fair value of indefinite-lived intangible assets?`

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
fp16: 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: False
fp16: True
fp16_opt_level: O1
half_precision_backend: auto
bf16_full_eval: False
fp16_full_eval: False
tf32: None
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
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	1.6045	-	-	-	-	-
0.9746	12	-	0.7895	0.7895	0.7764	0.7680	0.7277
1.6244	20	0.6975	-	-	-	-	-
1.9492	24	-	0.8044	0.8026	0.7924	0.7819	0.7515
2.4365	30	0.4732	-	-	-	-	-
2.9239	36	-	0.8064	0.8060	0.7944	0.7825	0.7549
3.2487	40	0.4182	-	-	-	-	-
3.8985	48	-	0.8078	0.8053	0.7946	0.7829	0.7555

The bold row denotes the saved checkpoint.

Framework Versions

Python: 3.10.12
Sentence Transformers: 3.3.1
Transformers: 4.41.2
PyTorch: 2.1.2+cu121
Accelerate: 1.1.1
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}
}