Nomic Financial Matryoshka

This is a sentence-transformers model finetuned from nomic-ai/nomic-embed-text-v1 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: nomic-ai/nomic-embed-text-v1
  • Maximum Sequence Length: 8192 tokens
  • Output Dimensionality: 768 tokens
  • Similarity Function: Cosine Similarity
  • Training Dataset:
    • json
  • Language: en
  • License: apache-2.0

Model Sources

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'max_seq_length': 8192, 'do_lower_case': False}) with Transformer model: NomicBertModel 
  (1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, '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("aniket0898/bge-base-financial-matryoshka")
# Run inference
sentences = [
    'How is the information about legal proceedings organized in the financial documents according to the provided context?',
    "The information about legal proceedings is organized under Part II, Item 8 in the section titled 'Financial Statements and Supplementary Data – Note 14'.",
    'We have a match-funding policy that addresses the interest rate risk by aligning the interest rate profile (fixed or floating rate and duration) of our debt portfolio with the interest rate profile of our finance receivable portfolio within a predetermined range on an ongoing basis. In connection with that policy, we use interest rate derivative instruments to modify the debt structure to match assets within the finance receivable portfolio.',
]
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

Metric Value
cosine_accuracy@1 0.7457
cosine_accuracy@3 0.8614
cosine_accuracy@5 0.8957
cosine_accuracy@10 0.93
cosine_precision@1 0.7457
cosine_precision@3 0.2871
cosine_precision@5 0.1791
cosine_precision@10 0.093
cosine_recall@1 0.7457
cosine_recall@3 0.8614
cosine_recall@5 0.8957
cosine_recall@10 0.93
cosine_ndcg@10 0.8399
cosine_mrr@10 0.8108
cosine_map@100 0.8137

Information Retrieval

Metric Value
cosine_accuracy@1 0.7357
cosine_accuracy@3 0.8514
cosine_accuracy@5 0.8914
cosine_accuracy@10 0.93
cosine_precision@1 0.7357
cosine_precision@3 0.2838
cosine_precision@5 0.1783
cosine_precision@10 0.093
cosine_recall@1 0.7357
cosine_recall@3 0.8514
cosine_recall@5 0.8914
cosine_recall@10 0.93
cosine_ndcg@10 0.8353
cosine_mrr@10 0.8047
cosine_map@100 0.8075

Information Retrieval

Metric Value
cosine_accuracy@1 0.7286
cosine_accuracy@3 0.8614
cosine_accuracy@5 0.8857
cosine_accuracy@10 0.9271
cosine_precision@1 0.7286
cosine_precision@3 0.2871
cosine_precision@5 0.1771
cosine_precision@10 0.0927
cosine_recall@1 0.7286
cosine_recall@3 0.8614
cosine_recall@5 0.8857
cosine_recall@10 0.9271
cosine_ndcg@10 0.832
cosine_mrr@10 0.8011
cosine_map@100 0.8041

Information Retrieval

Metric Value
cosine_accuracy@1 0.7129
cosine_accuracy@3 0.8329
cosine_accuracy@5 0.8671
cosine_accuracy@10 0.9143
cosine_precision@1 0.7129
cosine_precision@3 0.2776
cosine_precision@5 0.1734
cosine_precision@10 0.0914
cosine_recall@1 0.7129
cosine_recall@3 0.8329
cosine_recall@5 0.8671
cosine_recall@10 0.9143
cosine_ndcg@10 0.8146
cosine_mrr@10 0.7826
cosine_map@100 0.786

Information Retrieval

Metric Value
cosine_accuracy@1 0.6643
cosine_accuracy@3 0.8043
cosine_accuracy@5 0.8457
cosine_accuracy@10 0.9029
cosine_precision@1 0.6643
cosine_precision@3 0.2681
cosine_precision@5 0.1691
cosine_precision@10 0.0903
cosine_recall@1 0.6643
cosine_recall@3 0.8043
cosine_recall@5 0.8457
cosine_recall@10 0.9029
cosine_ndcg@10 0.7821
cosine_mrr@10 0.7437
cosine_map@100 0.7468

Training Details

Training Dataset

json

  • Dataset: json
  • Size: 6,300 training samples
  • Columns: anchor and positive
  • Approximate statistics based on the first 1000 samples:
    anchor positive
    type string string
    details
    • min: 2 tokens
    • mean: 20.47 tokens
    • max: 40 tokens
    • min: 9 tokens
    • mean: 45.09 tokens
    • max: 272 tokens
  • Samples:
    anchor positive
    What was the stored value of cards and loyalty program balances at the end of fiscal year 2022? Stored value cards and loyalty program at October 2, 2022 showed a balance of approximately $1.503 billion.
    What transformation is planned for Le Jardin located at The Londoner Macao? Le Jardin, located on the southern flank of The Londoner Macao, is to undergo a transformation into a distinctive garden-themed attraction spanning approximately 50,000 square meters.
    What are the key terms of the new Labor Agreement ratified by the UAW in 2023? The key terms and provisions of the Labor Agreement are: General wage increases of 11% upon ratification in 2023, 3% in September each of 2024, 2025 and 2026, and 5% in September 2027; Consolidation of applicable wage classifications for in-progression, temporary and other employees – with employees reaching the top classification rate upon the completion of 156 weeks of active service; The re-establishment of a cost-of-living allowance; Lump sum ratification bonus payments of $5,000 paid to eligible employees in the three months ended December 31, 2023; For members currently employed and enrolled in the Employees’ Pension Plan, an increase of $5.00 to the monthly basic benefit for past and future service provided; A 3.6% increase in company contributions to eligible employees' defined contribution retirement accounts; and Annual contribution of $500 to eligible retirees or surviving spouses.
  • 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_768_cosine_map@100 dim_512_cosine_map@100 dim_256_cosine_map@100 dim_128_cosine_map@100 dim_64_cosine_map@100
0.8122 10 0.7331 - - - - -
0.9746 12 - 0.7871 0.7796 0.7747 0.7546 0.7214
1.6244 20 0.2506 - - - - -
1.9492 24 - 0.8021 0.7990 0.7869 0.7691 0.7371
2.4365 30 0.1029 - - - - -
2.9239 36 - 0.8030 0.8017 0.7926 0.7760 0.7402
3.2487 40 0.054 - - - - -
3.8985 48 - 0.8055 0.799 0.7924 0.7754 0.7383
0.8122 10 0.0397 - - - - -
0.9746 12 - 0.8109 0.7983 0.7974 0.7795 0.7373
1.6244 20 0.0301 - - - - -
1.9492 24 - 0.8115 0.8049 0.8026 0.7839 0.7486
2.4365 30 0.0236 - - - - -
2.9239 36 - 0.8138 0.8082 0.8045 0.7858 0.7470
3.2487 40 0.0131 - - - - -
3.8985 48 - 0.8137 0.8075 0.8041 0.786 0.7468
  • The bold row denotes the saved checkpoint.

Framework Versions

  • Python: 3.8.10
  • Sentence Transformers: 3.2.1
  • Transformers: 4.41.2
  • PyTorch: 2.1.2+cu121
  • Accelerate: 1.0.1
  • 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}
}
Downloads last month
7
Safetensors
Model size
137M params
Tensor type
F32
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for aniket0898/bge-base-financial-matryoshka

Finetuned
(20)
this model

Evaluation results