kr-manish/bge-base-financial-matryoshka

SentenceTransformer based on BAAI/bge-base-en-v1.5

This is a sentence-transformers model finetuned from BAAI/bge-base-en-v1.5. 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 tokens
• Similarity Function: Cosine Similarity

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("kr-manish/bge-base-financial-matryoshka")
# Run inference
sentences = [
    "No, work-from-home arrangements do not affect an employee's employment terms, compensation, and benefits at Priya Softweb. This clarifies that work-from-home is a flexible work arrangement and does not impact the employee's overall employment status or benefits.",
    'Do work-from-home arrangements affect compensation and benefits at Priya Softweb?',
    'What is the objective of the Work From Home Policy at Priya Softweb?',
]
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

Metric	Value
cosine_accuracy@1	0.6111
cosine_accuracy@3	0.7778
cosine_accuracy@5	0.7778
cosine_accuracy@10	0.8333
cosine_precision@1	0.6111
cosine_precision@3	0.2593
cosine_precision@5	0.1556
cosine_precision@10	0.0833
cosine_recall@1	0.6111
cosine_recall@3	0.7778
cosine_recall@5	0.7778
cosine_recall@10	0.8333
cosine_ndcg@10	0.7192
cosine_mrr@10	0.6829
cosine_map@100	0.6896

Information Retrieval

• Dataset: dim_512
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.5556
cosine_accuracy@3	0.7778
cosine_accuracy@5	0.7778
cosine_accuracy@10	0.8333
cosine_precision@1	0.5556
cosine_precision@3	0.2593
cosine_precision@5	0.1556
cosine_precision@10	0.0833
cosine_recall@1	0.5556
cosine_recall@3	0.7778
cosine_recall@5	0.7778
cosine_recall@10	0.8333
cosine_ndcg@10	0.6973
cosine_mrr@10	0.6537
cosine_map@100	0.6595

Information Retrieval

• Dataset: dim_256
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.4444
cosine_accuracy@3	0.6667
cosine_accuracy@5	0.7778
cosine_accuracy@10	0.8889
cosine_precision@1	0.4444
cosine_precision@3	0.2222
cosine_precision@5	0.1556
cosine_precision@10	0.0889
cosine_recall@1	0.4444
cosine_recall@3	0.6667
cosine_recall@5	0.7778
cosine_recall@10	0.8889
cosine_ndcg@10	0.6562
cosine_mrr@10	0.5836
cosine_map@100	0.5863

Information Retrieval

• Dataset: dim_128
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.4444
cosine_accuracy@3	0.6667
cosine_accuracy@5	0.7222
cosine_accuracy@10	0.7778
cosine_precision@1	0.4444
cosine_precision@3	0.2222
cosine_precision@5	0.1444
cosine_precision@10	0.0778
cosine_recall@1	0.4444
cosine_recall@3	0.6667
cosine_recall@5	0.7222
cosine_recall@10	0.7778
cosine_ndcg@10	0.6174
cosine_mrr@10	0.5653
cosine_map@100	0.5729

Information Retrieval

• Dataset: dim_64
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.3889
cosine_accuracy@3	0.6111
cosine_accuracy@5	0.6667
cosine_accuracy@10	0.7778
cosine_precision@1	0.3889
cosine_precision@3	0.2037
cosine_precision@5	0.1333
cosine_precision@10	0.0778
cosine_recall@1	0.3889
cosine_recall@3	0.6111
cosine_recall@5	0.6667
cosine_recall@10	0.7778
cosine_ndcg@10	0.5655
cosine_mrr@10	0.4992
cosine_map@100	0.5079

Training Details

Training Dataset

Unnamed Dataset

• Size: 160 training samples
• Columns: positive and anchor
• Approximate statistics based on the first 1000 samples:

  	positive	anchor
  type	string	string
  details	min: 18 tokens mean: 93.95 tokens max: 381 tokens	min: 13 tokens mean: 20.32 tokens max: 34 tokens

• Samples:

  positive	anchor
  Priya Softweb's HR Manual provides valuable insights into the company's culture and values. Key takeaways include: * Structure and Transparency: The company emphasizes a structured and transparent approach to its HR processes. This is evident in its clear policies for recruitment, performance appraisals, compensation, leave management, work-from-home arrangements, and incident reporting. * Professionalism and Ethics: Priya Softweb places a high value on professionalism and ethical conduct. Its dress code, guidelines for mobile phone usage, and strict policies against tobacco use within the office all point toward a commitment to maintaining a professional and respectful work environment. * Employee Well-being: The company demonstrates a genuine concern for the well-being of its employees. This is reflected in its comprehensive leave policies, flexible work-from-home arrangements, and efforts to promote a healthy and clean workspace. * Diversity and Inclusion: Priya Softweb is committed to fostering a diverse and inclusive workplace, where employees from all backgrounds feel valued and respected. Its DEI policy outlines the company's commitment to equal opportunities, diverse hiring practices, and inclusive benefits and policies. * Continuous Learning and Development: The company encourages a culture of continuous learning and development, providing opportunities for employees to expand their skillsets and stay current with industry advancements. This is evident in its policies for Ethics & Compliance training and its encouragement of utilizing idle time for self-learning and exploring new technologies. Overall, Priya Softweb's HR Manual reveals a company culture that prioritizes structure, transparency, professionalism, employee well-being, diversity, and a commitment to continuous improvement. The company strives to create a supportive and growth-oriented work environment where employees feel valued and empowered to succeed.	What are the key takeaways from Priya Softweb's HR Manual regarding the company's culture and values?
  Priya Softweb provides allocated basement parking facilities for employees to park their two-wheelers and four-wheelers. However, parking on the ground floor, around the lawn or main premises, is strictly prohibited as this space is reserved for Directors. Employees should use the parking under wings 5 and 6, while other parking spaces are allocated to different wings. Parking two-wheelers in the car parking zone is not permitted, even if space is available. Two-wheelers should be parked in the designated basement space on the main stand, not on the side stand. Employees are encouraged to park in common spaces on a first-come, first-served basis. The company clarifies that it is not responsible for providing parking and that employees park their vehicles at their own risk. This comprehensive parking policy ensures organized parking arrangements and clarifies the company's liability regarding vehicle safety.	What are the parking arrangements at Priya Softweb?
  Investments and declarations must be submitted on or before the 25th of each month through OMS at Priya Softweb.	What is the deadline for submitting investments and declarations at Priya Softweb?

• 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
• 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: False
• 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
• batch_sampler: no_duplicates
• multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	dim_128_cosine_map@100	dim_256_cosine_map@100	dim_512_cosine_map@100	dim_64_cosine_map@100	dim_768_cosine_map@100
1.0	1	0.5729	0.5863	0.6595	0.5079	0.6896
2.0	2	0.5729	0.5863	0.6595	0.5079	0.6896
3.0	3	0.5729	0.5863	0.6595	0.5079	0.6896
3.2	4	0.5729	0.5863	0.6595	0.5079	0.6896

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.10.12
• Sentence Transformers: 3.0.1
• Transformers: 4.41.2
• PyTorch: 2.1.2+cu121
• Accelerate: 0.31.0
• 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}
}