rezarahim/bge-finetuned

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

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

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("rezarahim/bge-finetuned")
# Run inference
sentences = [
    "What are the potential risks to the company's operating results?",
    " The company's operating results may be impacted by challenges in integrating acquisition target systems, difficulties with system integration with key suppliers and customers, training and change management needs, loss of or inability to sell to a significant number of customers, and changes in purchasing patterns or decisions by channel partners.",
    ' The change resulted in an increase in operating income of $135 million and net income of $114 million after tax, or $0.05 per both basic and diluted share.',
]
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: bge-base-en
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.4607
cosine_accuracy@3	0.6011
cosine_accuracy@5	0.6685
cosine_accuracy@10	0.7416
cosine_precision@1	0.4607
cosine_precision@3	0.2004
cosine_precision@5	0.1337
cosine_precision@10	0.0742
cosine_recall@1	0.4607
cosine_recall@3	0.6011
cosine_recall@5	0.6685
cosine_recall@10	0.7416
cosine_ndcg@10	0.595
cosine_mrr@10	0.5488
cosine_map@100	0.5588
dot_accuracy@1	0.4607
dot_accuracy@3	0.6011
dot_accuracy@5	0.6685
dot_accuracy@10	0.7416
dot_precision@1	0.4607
dot_precision@3	0.2004
dot_precision@5	0.1337
dot_precision@10	0.0742
dot_recall@1	0.4607
dot_recall@3	0.6011
dot_recall@5	0.6685
dot_recall@10	0.7416
dot_ndcg@10	0.595
dot_mrr@10	0.5488
dot_map@100	0.5588

Information Retrieval

• Dataset: bge-base-en
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.7303
cosine_accuracy@3	0.8315
cosine_accuracy@5	0.8876
cosine_accuracy@10	0.927
cosine_precision@1	0.7303
cosine_precision@3	0.2772
cosine_precision@5	0.1775
cosine_precision@10	0.0927
cosine_recall@1	0.7303
cosine_recall@3	0.8315
cosine_recall@5	0.8876
cosine_recall@10	0.927
cosine_ndcg@10	0.8246
cosine_mrr@10	0.7922
cosine_map@100	0.7975
dot_accuracy@1	0.7303
dot_accuracy@3	0.8315
dot_accuracy@5	0.8876
dot_accuracy@10	0.927
dot_precision@1	0.7303
dot_precision@3	0.2772
dot_precision@5	0.1775
dot_precision@10	0.0927
dot_recall@1	0.7303
dot_recall@3	0.8315
dot_recall@5	0.8876
dot_recall@10	0.927
dot_ndcg@10	0.8246
dot_mrr@10	0.7922
dot_map@100	0.7975

Training Details

Training Dataset

train

• Dataset: train
• Size: 178 training samples
• Columns: anchor and positive
• Approximate statistics based on the first 178 samples:

  	anchor	positive
  type	string	string
  details	min: 10 tokens mean: 22.24 tokens max: 43 tokens	min: 3 tokens mean: 37.76 tokens max: 118 tokens

• Samples:

  anchor	positive
  What is the publication date of the NVIDIA Corporation Annual Report 2024?	February 21st, 2024
  What is the filing date of the 10-K report for NVIDIA Corporation in 2004?	The filing dates of the 10-K reports for NVIDIA Corporation in 2004 are May 20th, March 29th, and April 25th.
  What is the purpose of the section of the filing that requires the registrant to indicate whether it has submitted electronically every Interactive Data File required to be submitted during the preceding 12 months?	The purpose of this section is to comply with Rule 405 of Regulation S-T, which requires the registrant to submit electronic files for certain financial information.

• Loss: MultipleNegativesRankingLoss with these parameters:

  {
      "scale": 20.0,
      "similarity_fct": "cos_sim"
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: epoch
• per_device_train_batch_size: 4
• per_device_eval_batch_size: 16
• gradient_accumulation_steps: 16
• learning_rate: 2e-05
• num_train_epochs: 25
• 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: 4
• 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: 25
• 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
• eval_on_start: False
• use_liger_kernel: False
• eval_use_gather_object: False
• batch_sampler: no_duplicates
• multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	Training Loss	bge-base-en_cosine_map@100
0	0	-	0.5588
0.7111	2	-	0.5609
1.7778	5	-	0.5889
2.8444	8	-	0.6191
3.5556	10	0.749	-
3.9111	11	-	0.6704
4.9778	14	-	0.7009
5.6889	16	-	0.7158
6.7556	19	-	0.7454
7.1111	20	0.363	-
7.8222	22	-	0.7633
8.8889	25	-	0.7685
9.9556	28	-	0.7816
10.6667	30	0.2181	0.7857
11.7333	33	-	0.7866
12.8	36	-	0.7939
13.8667	39	-	0.7953
14.2222	40	0.1552	-
14.9333	42	-	0.7962
16.0	45	-	0.7975
16.7111	47	-	0.7975
17.7778	50	0.1315	0.7975

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.10.12
• Sentence Transformers: 3.1.1
• Transformers: 4.45.2
• PyTorch: 2.5.1+cu121
• 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",
}

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}
}