schawla2/e5-unsupervised-financial-matryoshka

E5 unsupervised Financial Matryoshka

This is a sentence-transformers model finetuned from intfloat/e5-large-unsupervised on the json dataset. It maps sentences & paragraphs to a 1024-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: intfloat/e5-large-unsupervised
• Maximum Sequence Length: 512 tokens
• Output Dimensionality: 1024 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': False}) with Transformer model: BertModel 
  (1): Pooling({'word_embedding_dimension': 1024, '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("schawla2/e5-unsupervised-financial-matryoshka")
# Run inference
sentences = [
    'What was the total cost and expenses reported by Berkshire Hathaway for the year ended December 31, 2023?',
    'Total costs and expenses | | 321,144 | | | 266,484 | | | 243,752',
    'Qulipta (atogepant) is a calcitonin gene-related peptide receptor antagonist indicated for the preventive treatment of episodic and chronic migraine in adults. Qulipta is commercialized in the United States and Canada and is approved in the European Union under the brand name Aquipta.',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 1024]

# 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	0.7271	0.7271	0.72	0.7086	0.6757
cosine_accuracy@3	0.85	0.8457	0.8457	0.8286	0.8186
cosine_accuracy@5	0.8786	0.88	0.8714	0.8729	0.8457
cosine_accuracy@10	0.9114	0.9129	0.9057	0.8986	0.8843
cosine_precision@1	0.7271	0.7271	0.72	0.7086	0.6757
cosine_precision@3	0.2833	0.2819	0.2819	0.2762	0.2729
cosine_precision@5	0.1757	0.176	0.1743	0.1746	0.1691
cosine_precision@10	0.0911	0.0913	0.0906	0.0899	0.0884
cosine_recall@1	0.7271	0.7271	0.72	0.7086	0.6757
cosine_recall@3	0.85	0.8457	0.8457	0.8286	0.8186
cosine_recall@5	0.8786	0.88	0.8714	0.8729	0.8457
cosine_recall@10	0.9114	0.9129	0.9057	0.8986	0.8843
cosine_ndcg@10	0.8225	0.8224	0.816	0.8074	0.7862
cosine_mrr@10	0.7937	0.7931	0.7869	0.7777	0.7543
cosine_map@100	0.7974	0.7967	0.7907	0.7816	0.7588

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: 8 tokens mean: 20.8 tokens max: 51 tokens	min: 7 tokens mean: 45.24 tokens max: 326 tokens

• Samples:

  anchor	positive
  How many full-time employees did Microsoft report as of June 30, 2023?	As of June 30, 2023, we employed approximately 221,000 people on a full-time basis, 120,000 in the U.S. and 101,000 internationally.
  What was the total amount CSC paid for Series G preferred stock repurchases in 2023?	In 2023, CSC repurchased 42,036 depositary shares representing interests in Series G preferred stock for a total amount of $42 million.
  What does Note 13 in the Annual Report on Form 10-K discuss?	For a discussion of legal and other proceedings in which we are involved, see Note 13 - Commitments and Contingencies in the Notes to Consolidated Financial Statements.

• 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_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: 8
• 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: 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: 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
• 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
0.2030	10	9.3166	-	-	-	-	-
0.4061	20	3.7163	-	-	-	-	-
0.6091	30	2.8216	-	-	-	-	-
0.8122	40	1.9313	-	-	-	-	-
1.0	50	1.5613	0.8230	0.8237	0.8153	0.8036	0.7771
1.2030	60	1.0926	-	-	-	-	-
1.4061	70	0.3367	-	-	-	-	-
1.6091	80	0.3958	-	-	-	-	-
1.8122	90	0.6527	-	-	-	-	-
2.0	100	0.4483	0.8202	0.8209	0.8118	0.8033	0.7792
2.2030	110	0.1823	-	-	-	-	-
2.4061	120	0.0494	-	-	-	-	-
2.6091	130	0.1204	-	-	-	-	-
2.8122	140	0.2021	-	-	-	-	-
3.0	150	0.2088	0.8211	0.8213	0.8148	0.8064	0.7825
3.2030	160	0.062	-	-	-	-	-
3.4061	170	0.022	-	-	-	-	-
3.6091	180	0.0654	-	-	-	-	-
3.8122	190	0.1481	-	-	-	-	-
3.934	196	-	0.8225	0.8224	0.816	0.8074	0.7862

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.10.16
• Sentence Transformers: 3.3.1
• Transformers: 4.48.1
• PyTorch: 2.5.1+cu124
• Accelerate: 1.3.0
• Datasets: 3.3.2
• Tokenizers: 0.21.0

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}
}