willy-arison/bge-base-financial-willy3

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("willy-arison/bge-base-financial-willy3")
# Run inference
sentences = [
    'The topic of the first paper is "Fine-tuning gpt-3 for russian text summarization."',
    'What is the topic of the first paper mentioned in the text?',
    "What is the model's response when the date is changed to September 8, 2030?",
]
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 with these parameters:

  {
      "truncate_dim": 768
  }
  

Metric	Value
cosine_accuracy@1	0.4
cosine_accuracy@3	0.5
cosine_accuracy@5	0.5667
cosine_accuracy@10	0.6667
cosine_precision@1	0.4
cosine_precision@3	0.1667
cosine_precision@5	0.1133
cosine_precision@10	0.0667
cosine_recall@1	0.4
cosine_recall@3	0.5
cosine_recall@5	0.5667
cosine_recall@10	0.6667
cosine_ndcg@10	0.5201
cosine_mrr@10	0.4745
cosine_map@100	0.4911

Information Retrieval

• Dataset: dim_512
• Evaluated with InformationRetrievalEvaluator with these parameters:

  {
      "truncate_dim": 512
  }
  

Metric	Value
cosine_accuracy@1	0.3667
cosine_accuracy@3	0.5
cosine_accuracy@5	0.6333
cosine_accuracy@10	0.7
cosine_precision@1	0.3667
cosine_precision@3	0.1667
cosine_precision@5	0.1267
cosine_precision@10	0.07
cosine_recall@1	0.3667
cosine_recall@3	0.5
cosine_recall@5	0.6333
cosine_recall@10	0.7
cosine_ndcg@10	0.5277
cosine_mrr@10	0.4731
cosine_map@100	0.4859

Information Retrieval

• Dataset: dim_256
• Evaluated with InformationRetrievalEvaluator with these parameters:

  {
      "truncate_dim": 256
  }
  

Metric	Value
cosine_accuracy@1	0.4333
cosine_accuracy@3	0.5667
cosine_accuracy@5	0.5667
cosine_accuracy@10	0.6667
cosine_precision@1	0.4333
cosine_precision@3	0.1889
cosine_precision@5	0.1133
cosine_precision@10	0.0667
cosine_recall@1	0.4333
cosine_recall@3	0.5667
cosine_recall@5	0.5667
cosine_recall@10	0.6667
cosine_ndcg@10	0.5385
cosine_mrr@10	0.4984
cosine_map@100	0.5094

Information Retrieval

• Dataset: dim_128
• Evaluated with InformationRetrievalEvaluator with these parameters:

  {
      "truncate_dim": 128
  }
  

Metric	Value
cosine_accuracy@1	0.3
cosine_accuracy@3	0.5
cosine_accuracy@5	0.6
cosine_accuracy@10	0.6
cosine_precision@1	0.3
cosine_precision@3	0.1667
cosine_precision@5	0.12
cosine_precision@10	0.06
cosine_recall@1	0.3
cosine_recall@3	0.5
cosine_recall@5	0.6
cosine_recall@10	0.6
cosine_ndcg@10	0.4445
cosine_mrr@10	0.3939
cosine_map@100	0.4039

Information Retrieval

• Dataset: dim_64
• Evaluated with InformationRetrievalEvaluator with these parameters:

  {
      "truncate_dim": 64
  }
  

Metric	Value
cosine_accuracy@1	0.2667
cosine_accuracy@3	0.4333
cosine_accuracy@5	0.5
cosine_accuracy@10	0.5333
cosine_precision@1	0.2667
cosine_precision@3	0.1444
cosine_precision@5	0.1
cosine_precision@10	0.0533
cosine_recall@1	0.2667
cosine_recall@3	0.4333
cosine_recall@5	0.5
cosine_recall@10	0.5333
cosine_ndcg@10	0.4073
cosine_mrr@10	0.3659
cosine_map@100	0.3844

Training Details

Training Dataset

json

• Dataset: json
• Size: 264 training samples
• Columns: positive and anchor
• Approximate statistics based on the first 264 samples:

  	positive	anchor
  type	string	string
  details	min: 3 tokens mean: 15.37 tokens max: 61 tokens	min: 9 tokens mean: 18.22 tokens max: 37 tokens

• Samples:

  positive	anchor
  Hospital systems may wish to update an LLM with their current medical guidelines.	Give an example of a specific domain or industry that might want to update a language model with their own knowledge.
  The Gemini models struggle to learn a significant proportion of the data even after 20 or 30 epochs.	How do the Gemini models perform in learning the training data compared to the OpenAI models?
  Anthropic, Google, OpenAI	Which companies have contributed to the rapid iteration and evolution of Large Language Models?

• 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: 50
• lr_scheduler_type: cosine
• warmup_ratio: 0.1
• bf16: 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: 50
• 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: 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: 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
• 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
1.0	1	-	0.3407	0.3622	0.3385	0.3376	0.3703
2.0	2	-	0.3739	0.3652	0.3634	0.3429	0.3613
3.0	3	-	0.3742	0.3666	0.3584	0.3495	0.3504
4.0	4	-	0.3838	0.3666	0.3580	0.3480	0.3488
1.0	1	-	0.3742	0.3666	0.3584	0.3495	0.3504
2.0	2	-	0.3939	0.3767	0.3611	0.3626	0.3302
3.0	3	-	0.3844	0.3896	0.3873	0.3589	0.3388
4.0	4	-	0.3885	0.4011	0.3989	0.3581	0.3515
5.0	5	-	0.3790	0.4024	0.3984	0.3658	0.3563
6.0	6	-	0.3815	0.4047	0.3998	0.3904	0.3573
7.0	7	-	0.3868	0.4164	0.4038	0.4024	0.3512
8.0	8	-	0.3971	0.4224	0.4047	0.3933	0.3518
9.0	9	-	0.3971	0.4224	0.4062	0.3933	0.3536
10.0	10	69.6696	0.3971	0.4353	0.4062	0.4056	0.3579
1.0	1	-	0.3971	0.4353	0.4062	0.4056	0.3579
2.0	2	-	0.3885	0.4454	0.4181	0.3947	0.3558
3.0	3	-	0.4141	0.4482	0.4424	0.4020	0.3584
4.0	4	-	0.4215	0.4502	0.4443	0.3854	0.3571
5.0	5	-	0.4402	0.4596	0.4366	0.3874	0.3671
6.0	6	-	0.4402	0.4500	0.4372	0.3901	0.3645
7.0	7	-	0.4551	0.4557	0.4390	0.4045	0.3489
8.0	8	-	0.4658	0.4637	0.4390	0.3865	0.3487
9.0	9	-	0.4658	0.4741	0.4586	0.3827	0.3587
10.0	10	41.2785	0.4590	0.4741	0.4510	0.3941	0.3600
11.0	11	-	0.4590	0.4741	0.4515	0.3841	0.3604
12.0	12	-	0.4605	0.4802	0.4556	0.3864	0.3945
13.0	13	-	0.4590	0.4792	0.4433	0.3864	0.3768
14.0	14	-	0.4595	0.4844	0.4447	0.3964	0.3812
15.0	15	-	0.4605	0.4945	0.4447	0.3974	0.3869
16.0	16	-	0.4611	0.4951	0.4465	0.3974	0.3861
17.0	17	-	0.4605	0.4951	0.4465	0.3974	0.3869
18.0	18	-	0.4605	0.4955	0.4588	0.4097	0.3905
19.0	19	-	0.4605	0.4951	0.4479	0.3974	0.3869
20.0	20	24.2435	0.4605	0.4951	0.4479	0.3964	0.3869
1.0	1	-	0.4605	0.4955	0.4588	0.4097	0.3905
2.0	2	-	0.4613	0.4951	0.4472	0.3969	0.3779
3.0	3	-	0.4736	0.4955	0.4697	0.3979	0.3844
4.0	4	-	0.4736	0.4958	0.4705	0.3985	0.3844
5.0	5	-	0.4739	0.4993	0.4724	0.3965	0.3873
6.0	6	-	0.4828	0.4916	0.4756	0.4146	0.3773
7.0	7	-	0.4832	0.5012	0.5002	0.4023	0.3817
8.0	8	-	0.4928	0.5012	0.5057	0.4061	0.3802
9.0	9	-	0.4947	0.5005	0.5192	0.4184	0.4012
10.0	10	14.397	0.4951	0.5105	0.5174	0.4151	0.3935
11.0	11	-	0.4968	0.5114	0.5218	0.4151	0.3935
12.0	12	-	0.4973	0.5225	0.5151	0.4328	0.3983
13.0	13	-	0.4979	0.5214	0.5147	0.4318	0.3926
14.0	14	-	0.5004	0.5229	0.5147	0.4151	0.4023
15.0	15	-	0.4989	0.5258	0.5167	0.4318	0.4066
16.0	16	-	0.5033	0.5269	0.5167	0.4336	0.4091
17.0	17	-	0.5048	0.5282	0.5167	0.4336	0.3995
18.0	18	-	0.5048	0.5308	0.5175	0.4336	0.4091
19.0	19	-	0.5033	0.5207	0.5175	0.4325	0.4091
20.0	20	8.6526	0.5033	0.5207	0.5160	0.4325	0.4203
21.0	21	-	0.5156	0.5207	0.5214	0.4336	0.4202
22.0	22	-	0.5156	0.5218	0.5229	0.4306	0.4202
23.0	23	-	0.5171	0.5222	0.5175	0.4354	0.4100
24.0	24	-	0.5156	0.5207	0.5175	0.4354	0.4096
25.0	25	-	0.5252	0.5222	0.5204	0.4311	0.4082
26.0	26	-	0.5267	0.5222	0.5160	0.4355	0.4316
27.0	27	-	0.5267	0.5218	0.5160	0.4362	0.4316
28.0	28	-	0.5171	0.5218	0.5175	0.4372	0.4329
29.0	29	-	0.5171	0.5177	0.5175	0.4372	0.4358
30.0	30	5.2469	0.5171	0.5207	0.5242	0.4420	0.4252
31.0	31	-	0.5062	0.5191	0.5365	0.4474	0.4262
32.0	32	-	0.5062	0.5238	0.5365	0.4474	0.4343
33.0	33	-	0.5048	0.5200	0.5279	0.4430	0.4292
34.0	34	-	0.5048	0.5200	0.5269	0.4430	0.4466
35.0	35	-	0.5062	0.5214	0.5269	0.4430	0.4292
36.0	36	-	0.5085	0.5223	0.5269	0.4430	0.4389
37.0	37	-	0.5085	0.5223	0.5262	0.4430	0.4196
38.0	38	-	0.5201	0.5281	0.5385	0.4474	0.4239
39.0	39	-	0.5201	0.5238	0.5385	0.4474	0.4239
40.0	40	4.274	0.5201	0.5267	0.5256	0.4445	0.4196
41.0	41	-	0.5201	0.5223	0.5262	0.4474	0.4239
42.0	42	-	0.5078	0.5277	0.5379	0.4445	0.4073
43.0	43	-	0.5215	0.5281	0.5379	0.4445	0.4196
44.0	44	-	0.5215	0.5291	0.5256	0.4489	0.4196
45.0	45	-	0.5201	0.5277	0.5256	0.4445	0.4196
46.0	46	-	0.5078	0.5277	0.5262	0.4445	0.4196
47.0	47	-	0.5201	0.5277	0.5379	0.4445	0.4073
48.0	48	-	0.5215	0.5291	0.5262	0.4445	0.4073
49.0	49	-	0.5215	0.5291	0.5385	0.4489	0.4116
50.0	50	3.6753	0.5201	0.5277	0.5385	0.4445	0.4073

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.11.13
• Sentence Transformers: 4.1.0
• Transformers: 4.52.4
• PyTorch: 2.6.0+cu124
• Accelerate: 1.8.1
• Datasets: 3.6.0
• Tokenizers: 0.21.1

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}
}