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("thang1943/bge-base-financial-matryoshka")
# Run inference
sentences = [
    'New stores | 131 | | 333 | | 464 | | 311 | | 225 | 536',
    'How many new stores did the Dollar Tree segment open in the fiscal year ending January 28, 2023?',
    'How is the discount rate for the Family Dollar goodwill impairment evaluation determined?',
]
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

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	0.6629	0.66	0.6671	0.6343	0.6286
cosine_accuracy@3	0.8129	0.8114	0.8057	0.79	0.7671
cosine_accuracy@5	0.8386	0.84	0.8371	0.8314	0.8143
cosine_accuracy@10	0.8871	0.8871	0.88	0.8729	0.8729
cosine_precision@1	0.6629	0.66	0.6671	0.6343	0.6286
cosine_precision@3	0.271	0.2705	0.2686	0.2633	0.2557
cosine_precision@5	0.1677	0.168	0.1674	0.1663	0.1629
cosine_precision@10	0.0887	0.0887	0.088	0.0873	0.0873
cosine_recall@1	0.6629	0.66	0.6671	0.6343	0.6286
cosine_recall@3	0.8129	0.8114	0.8057	0.79	0.7671
cosine_recall@5	0.8386	0.84	0.8371	0.8314	0.8143
cosine_recall@10	0.8871	0.8871	0.88	0.8729	0.8729
cosine_ndcg@10	0.7771	0.7739	0.7749	0.7568	0.7484
cosine_mrr@10	0.7417	0.7376	0.7411	0.7193	0.7088
cosine_map@100	0.7459	0.7416	0.7452	0.7238	0.7127

Training Details

Training Dataset

json

Dataset: json
Size: 6,300 training samples
Columns: positive and anchor
Approximate statistics based on the first 1000 samples:
positive anchor
type string string
details
min: 6 tokens
mean: 45.49 tokens
max: 371 tokens

min: 7 tokens
mean: 20.33 tokens
max: 41 tokens

	positive	anchor
type	string	string
details	min: 6 tokens mean: 45.49 tokens max: 371 tokens	min: 7 tokens mean: 20.33 tokens max: 41 tokens

Samples:

positive	anchor
`In their 2023 forward-looking statements, Goldman Sachs mentioned that results, financial condition, liquidity, and capital actions may differ, possibly materially, from the anticipated results. Important factors include those described in "Risk Factors" in Part I, Item 1A and "Forward-Looking Statements" in Part I, Item 1.`	`What factors could potentially alter Goldman Sachs' anticipated financial outcomes according to their 2023 forward-looking statements?`
`Visa Direct is part of Visa’s strategy beyond C2B payments and helps facilitate the delivery of funds to eligible cards, deposit accounts and digital wallets across more than 190 countries and territories. Visa Direct supports multiple use cases, such as P2P payments and account-to-account transfers, business and government payouts to individuals or small businesses, merchant settlements and refunds.`	`What is the purpose of Visa Direct?`
The Company's international operations are subject to different, and sometimes more stringent, legal and regulatory requirements, which vary widely by jurisdiction, including anti-corruption laws; economic sanctions laws; various privacy, insurance, tax, tariff and trade laws and regulations; corporate governance, privacy, data protection (including the EU's General Data Protection Regulation which began to apply across the EU during 2018), data mining, data transfer, labor and employment, intellectual property, consumer protection and investment laws and regulations; discriminatory licensing procedures; compulsory cessions of reinsurance; required localization of records and funds; higher premium and income taxes; limitations on dividends and repatriation of capital; and requirements for local participation in an insurer's ownership.	`What types of laws and regulations govern the international operations of a company?`

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: 2
per_device_eval_batch_size: 1
learning_rate: 2e-05
num_train_epochs: 4
lr_scheduler_type: cosine
warmup_ratio: 0.1
bf16: True
tf32: False
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: 2
per_device_eval_batch_size: 1
per_gpu_train_batch_size: None
per_gpu_eval_batch_size: None
gradient_accumulation_steps: 1
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: False
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