metadata

language:
  - en
license: apache-2.0
tags:
  - sentence-transformers
  - sentence-similarity
  - feature-extraction
  - generated_from_trainer
  - dataset_size:5600
  - loss:MatryoshkaLoss
  - loss:MultipleNegativesRankingLoss
base_model: BAAI/bge-base-en-v1.5
widget:
  - source_sentence: >-
      The Federal Energy Regulatory Commission (“FERC”) has also taken steps to
      enable the participation of energy storage in wholesale energy markets.
    sentences:
      - >-
        What segment-specific regulations apply to CVS Health Corporation's
        Pharmacy & Consumer Wellness segment?
      - >-
        What types of contracts does the company have for its health insurance
        plans, and how does premium revenue recognition function under these
        contracts?
      - >-
        What federal agency has taken steps to facilitate energy storage
        participation in wholesale energy markets?
  - source_sentence: >-
      Investments in subsidiaries and partnerships which we do not control but
      have significant influence are accounted for under the equity method.
    sentences:
      - >-
        How does the company aim to protect the health and well-being of the
        communities it operates in?
      - >-
        What are the key factors affecting the evaluation of the Economic Value
        of Equity (EVE) at the Charles Schwab Corporation?
      - >-
        What accounting method does the company use to account for investments
        in subsidiaries and partnerships where it does not control but has
        significant influence?
  - source_sentence: >-
      Item 8 of IBM's 2023 Annual Report includes financial statements and
      supplementary data spanning pages 44 through 121.
    sentences:
      - >-
        What entities are included among the Guarantors that guarantee each
        other’s debt securities as described in Comcast’s 2023 Annual Report?
      - >-
        What uncertainties exist regarding projections of future cash needs and
        cash flows?
      - >-
        How many pages in IBM's 2023 Annual Report to Stockholders are dedicated
        to financial statements and supplementary data?
  - source_sentence: >-
      Our compensation philosophy creates the framework for our rewards
      strategy, which focuses on five key elements: pay-for-performance,
      external market-based research, internal equity, fiscal responsibility,
      and legal compliance.
    sentences:
      - >-
        What financial instruments does the company invest in that are sensitive
        to interest rates?
      - What elements are included in the company's compensation programs?
      - >-
        What is the expected maximum potential loss from hurricane events for
        Chubb as of the end of 2023?
  - source_sentence: >-
      Outside of the U.S., many countries have established vehicle safety
      standards and regulations and are likely to adopt additional, more
      stringent requirements in the future.
    sentences:
      - >-
        What percentage of the company's sales categories in fiscal 2023 were
        failure and maintenance related?
      - >-
        What competitive factors influence Chubb International's international
        operations?
      - >-
        What changes are occurring with vehicle safety regulations outside of
        the U.S.?
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: 0.6885714285714286
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.8278571428571428
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.8728571428571429
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.9164285714285715
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.6885714285714286
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.275952380952381
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.17457142857142854
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.09164285714285714
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.6885714285714286
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.8278571428571428
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.8728571428571429
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.9164285714285715
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.8042449175537354
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.768181405895692
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.7712863400405022
            name: Cosine Map@100
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: dim 512
          type: dim_512
        metrics:
          - type: cosine_accuracy@1
            value: 0.6864285714285714
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.8292857142857143
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.8728571428571429
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.9135714285714286
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.6864285714285714
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.2764285714285714
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.17457142857142854
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.09135714285714285
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.6864285714285714
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.8292857142857143
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.8728571428571429
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.9135714285714286
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.8024352620004916
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.7665753968253971
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.7697268174707245
            name: Cosine Map@100
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: dim 256
          type: dim_256
        metrics:
          - type: cosine_accuracy@1
            value: 0.68
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.825
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.8635714285714285
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.9042857142857142
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.68
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.275
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.1727142857142857
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.09042857142857141
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.68
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.825
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.8635714285714285
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.9042857142857142
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.7955058944909328
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.7603066893424041
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.7637281364444245
            name: Cosine Map@100
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: dim 128
          type: dim_128
        metrics:
          - type: cosine_accuracy@1
            value: 0.6621428571428571
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.7964285714285714
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.8457142857142858
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.8907142857142857
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.6621428571428571
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.2654761904761905
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.16914285714285712
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.08907142857142857
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.6621428571428571
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.7964285714285714
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.8457142857142858
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.8907142857142857
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.7772894744328753
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.7408999433106581
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.7449491476160666
            name: Cosine Map@100
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: dim 64
          type: dim_64
        metrics:
          - type: cosine_accuracy@1
            value: 0.6285714285714286
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.7635714285714286
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.8057142857142857
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.8642857142857143
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.6285714285714286
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.2545238095238095
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.16114285714285712
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.08642857142857142
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.6285714285714286
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.7635714285714286
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.8057142857142857
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.8642857142857143
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.7447153698860624
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.7067037981859416
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.7112341263725279
            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("YxBxRyXJx/bge-base-financial-matryoshka")
# Run inference
sentences = [
    'Outside of the U.S., many countries have established vehicle safety standards and regulations and are likely to adopt additional, more stringent requirements in the future.',
    'What changes are occurring with vehicle safety regulations outside of the U.S.?',
    "What competitive factors influence Chubb International's international operations?",
]
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	0.6886	0.6864	0.68	0.6621	0.6286
cosine_accuracy@3	0.8279	0.8293	0.825	0.7964	0.7636
cosine_accuracy@5	0.8729	0.8729	0.8636	0.8457	0.8057
cosine_accuracy@10	0.9164	0.9136	0.9043	0.8907	0.8643
cosine_precision@1	0.6886	0.6864	0.68	0.6621	0.6286
cosine_precision@3	0.276	0.2764	0.275	0.2655	0.2545
cosine_precision@5	0.1746	0.1746	0.1727	0.1691	0.1611
cosine_precision@10	0.0916	0.0914	0.0904	0.0891	0.0864
cosine_recall@1	0.6886	0.6864	0.68	0.6621	0.6286
cosine_recall@3	0.8279	0.8293	0.825	0.7964	0.7636
cosine_recall@5	0.8729	0.8729	0.8636	0.8457	0.8057
cosine_recall@10	0.9164	0.9136	0.9043	0.8907	0.8643
cosine_ndcg@10	0.8042	0.8024	0.7955	0.7773	0.7447
cosine_mrr@10	0.7682	0.7666	0.7603	0.7409	0.7067
cosine_map@100	0.7713	0.7697	0.7637	0.7449	0.7112

Training Details

Training Dataset

json

Dataset: json
Size: 5,600 training samples
Columns: positive and anchor
Approximate statistics based on the first 1000 samples:
positive anchor
type string string
details
min: 4 tokens
mean: 44.34 tokens
max: 512 tokens

min: 9 tokens
mean: 20.46 tokens
max: 46 tokens

	positive	anchor
type	string	string
details	min: 4 tokens mean: 44.34 tokens max: 512 tokens	min: 9 tokens mean: 20.46 tokens max: 46 tokens

Samples:

positive	anchor
`Z-net is AutoZone's proprietary electronic catalog and enables AutoZoners to efficiently look up parts that customers need, providing complete job solutions and information based on vehicle specifics. It also tracks inventory availability across different locations.`	`What is the purpose of Z-net in AutoZone stores?`
`In 2023, the allowance for loan and lease losses was $13.3 billion on total loans and leases of $1,050.2 billion, which excludes loans accounted for under the fair value option.`	`What was the total amount of loans and leases at Bank of America by the end of 2023, excluding those accounted for under the fair value option?`
We significantly improved features in Service Manager™, which installers can use from their mobile devices to get service instantly. We continue to provide 24/7 support for installers and Enphase system owners globally across our phone, online chat, and email communications channel. We continue to train our customer service agents with a goal of reducing average customer wait times to under one minute, and we continue to expand our network of field service technicians in the United States, Europe and Australia to provide direct homeowner assistance.	`What measures has Enphase Energy, Inc. taken to improve customer service 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: 2
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: 2
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
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
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
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
0.9143	10	1.4537	0.7992	0.7952	0.7900	0.7703	0.7350
1.8286	20	0.6857	0.8042	0.8024	0.7955	0.7773	0.7447

The bold row denotes the saved checkpoint.

Framework Versions

Python: 3.10.12
Sentence Transformers: 3.3.0
Transformers: 4.46.2
PyTorch: 2.5.1+cu124
Accelerate: 1.1.1
Datasets: 3.1.0
Tokenizers: 0.20.3

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}
}