This is a sentence-transformers model finetuned from BAAI/bge-small-en-v1.5. It maps sentences & paragraphs to a 384-dimensional dense vector space and can be used for retrieval.

Model Details

Model Description

Model Type: Sentence Transformer
Base model: BAAI/bge-small-en-v1.5
Maximum Sequence Length: 256 tokens
Output Dimensionality: 384 dimensions
Similarity Function: Cosine Similarity
Supported Modality: Text

Model Sources

Documentation: Sentence Transformers Documentation
Repository: Sentence Transformers on GitHub
Hugging Face: Sentence Transformers on Hugging Face

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'transformer_task': 'feature-extraction', 'modality_config': {'text': {'method': 'forward', 'method_output_name': 'last_hidden_state'}}, 'module_output_name': 'token_embeddings', 'architecture': 'BertModel'})
  (1): Pooling({'embedding_dimension': 384, 'pooling_mode': 'cls', '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.

python

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("vivekkopthsd/fiqa-retriever-bge-small")
# Run inference
queries = [
    'Stock grant, taxes, and the IRS',
]
documents = [
    "I went through this too. There's a safe-harbor provision. If you prepay as estimated tax payments, 110% of your previous year's tax liability, there's no penalty for underpayment of the big liquidity-event tax liability.  https://www.irs.gov/publications/p17/ch04.html That's with the feds. Your state may have different rules. You would be very wise indeed to hire an accountant to prepare your return this year. If I were you I'd ask your company's CFO or finance chief to suggest somebody.   Congratulations, by the way.",
    "The value of the asset doesn't change just because of the exchange rate change. If a thing (valued in USD) costs USD $1 and USD $1 = CAN $1 (so the thing is also valued CAN $1) today and tomorrow CAN $1 worth USD $0.5 - the thing will continue being worth USD $1. If the thing is valued in CAN $, after the exchange rate change, the thing will be worth USD $2, but will still be valued CAN $1. What you're talking about is price quotes, not value. Price quotes will very quickly reach the value, since any deviation will be used by the traders to make profits on arbitrage. And algo-traders will make it happen much quicker than you can even notice the arbitrage existence.",
    '"Things I would specifically draw your attention to: the contract typically allows for an ""option"" to purchase; it does not typically compel purchase, although this is seen the purchase price is negotiated before anything gets signed the option to buy is typically available to the renter for the period of the lease contract (ie., if it\'s a 12 month contract the renter can opt to buy at any time in that 12 months) the amount of rent paid over time that will be applied to the purchase price is negotiated up-front before anything gets signed rent is paid at a slight premium (as Joe notes, if the rent should be $1000 per month, expect to pay $1200 per month) if the renter walks away they walk away empty handed; they do not get back the premium Having said all that - it\'s a contract negotiated between renter and seller and all of this is negotiable.  See also, ehow for a good overview."',
]
query_embeddings = model.encode_query(queries)
document_embeddings = model.encode_document(documents)
print(query_embeddings.shape, document_embeddings.shape)
# [1, 384] [3, 384]

# Get the similarity scores for the embeddings
similarities = model.similarity(query_embeddings, document_embeddings)
print(similarities)
# tensor([[ 0.4181,  0.0066, -0.0466]])

Evaluation

Metrics

Information Retrieval

Dataset: fiqa-test
Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.3858
cosine_accuracy@3	0.5324
cosine_accuracy@5	0.5818
cosine_accuracy@10	0.6667
cosine_precision@1	0.3858
cosine_precision@3	0.2382
cosine_precision@5	0.1728
cosine_precision@10	0.11
cosine_recall@1	0.1952
cosine_recall@3	0.321
cosine_recall@5	0.3716
cosine_recall@10	0.4555
cosine_ndcg@10	0.3923
cosine_mrr@10	0.4731
cosine_map@100	0.3345

Training Details

Training Dataset

Unnamed Dataset

Size: 14,080 training samples
Columns: sentence_0 and sentence_1
Approximate statistics based on the first 100 samples:
sentence_0 sentence_1
type string string
modality text text
details
min: 8 tokens
mean: 15.48 tokens
max: 28 tokens
min: 22 tokens
mean: 185.17 tokens
max: 256 tokens

Samples:

sentence_0	sentence_1
`How come I can't sell short certain stocks? My broker says “no shares are available”`	In finance, short selling (also known as shorting or going short) is the practice of selling assets, usually securities, that have been borrowed from a third party (usually a broker) with the intention of buying identical assets back at a later date to return to the lender. Remember your broker has to borrow it from somewhere, other clients or if they hold those specific stocks themselves. So if it isn't possible for them to lend you those stocks, they wouldn't. High P/E stocks would find more sellers than buyers, and if the broker has to deliver them, it would be a nightmare for him to deliver all those stocks, which he had lent you(others) back to whom he had borrowed from, as well as to people who had gone long(buy) when you went short(sell). And if every body is selling there is going to be a dearth of stocks to be borrowed from as everybody around is selling instead of buying.
`How to choose a company for an IRA?`	"I use TIAA-Cref for my 403(b) and Fidelity for my solo 401(k) and IRAs. I have previously used Vanguard and have also used other discount brokers for my IRA. All of these companies will charge you nothing for an IRA, so there's really no point in comparing cost in that respect. They are all the ""cheapest"" in this respect. Each one will allow you to purchase their mutual funds and those of their partners for free. They will charge you some kind of fee to invest in mutual funds of their competitors (like $35 or something). So the real question is this: which of these institutions offers the best mutual and index funds. While they are not the worst out there, you will find that TIAA-Cref are dominated by both Vanguard and Fidelity. The latter two offer far more and larger funds and their funds will always have lower expense ratios than their TIAA-Cref equivalent. If I could take my money out of TIAA-Cref and put it in Fidelity, I'd do so right now. BTW, you may or may not want t...
`Why are currency forwards needed?`	e.g. a European company has to pay 1 million USD exactly one year from now While that is theoretically possible, that is not a very common case. Mostly likely if they had to make a 1 million USD payment a year from now and they had the cash on hand they would be able to just make the payment today. A more common scenario for currency forwards is for investment hedging. Say that European company wants to buy into a mutual fund of some sort, say FUSEX. That is a USD based mutual fund. You can't buy into it directly with Euros. So if the company wants to buy into the fund they would need to convert their Euros to to USD. But now they have an extra risk parameter. They are not just exposed to the fluctuations of the fund, they are also exposed to the fluctuations of the currency market. Perhaps that fund will make a killing, but the exchange rate will tank and they will lose all their gains. By creating a forward to hedge their currency exposure risk they do not face this risk (flip side...

Loss: MultipleNegativesRankingLoss with these parameters:
json
1{ 2 "scale": 20.0, 3 "similarity_fct": "cos_sim", 4 "gather_across_devices": false, 5 "directions": [ 6 "query_to_doc" 7 ], 8 "partition_mode": "joint", 9 "hardness_mode": null, 10 "hardness_strength": 0.0 11}

Training Hyperparameters

Non-Default Hyperparameters

per_device_train_batch_size: 256
fp16: True
per_device_eval_batch_size: 256
multi_dataset_batch_sampler: round_robin

All Hyperparameters

Click to expand

per_device_train_batch_size: 256
num_train_epochs: 3
max_steps: -1
learning_rate: 5e-05
lr_scheduler_type: linear
lr_scheduler_kwargs: None
warmup_steps: 0
optim: adamw_torch_fused
optim_args: None
weight_decay: 0.0
adam_beta1: 0.9
adam_beta2: 0.999
adam_epsilon: 1e-08
optim_target_modules: None
gradient_accumulation_steps: 1
average_tokens_across_devices: True
max_grad_norm: 1
label_smoothing_factor: 0.0
bf16: False
fp16: True
bf16_full_eval: False
fp16_full_eval: False
tf32: None
gradient_checkpointing: False
gradient_checkpointing_kwargs: None
torch_compile: False
torch_compile_backend: None
torch_compile_mode: None
use_liger_kernel: False
liger_kernel_config: None
use_cache: False
neftune_noise_alpha: None
torch_empty_cache_steps: None
auto_find_batch_size: False
log_on_each_node: True
logging_nan_inf_filter: True
include_num_input_tokens_seen: no
log_level: passive
log_level_replica: warning
disable_tqdm: False
project: huggingface
trackio_space_id: None
trackio_bucket_id: None
trackio_static_space_id: None
per_device_eval_batch_size: 256
prediction_loss_only: True
eval_on_start: False
eval_do_concat_batches: True
eval_use_gather_object: False
eval_accumulation_steps: None
include_for_metrics: []
batch_eval_metrics: False
save_only_model: False
save_on_each_node: False
enable_jit_checkpoint: False
push_to_hub: False
hub_private_repo: None
hub_model_id: None
hub_strategy: every_save
hub_always_push: False
hub_revision: None
load_best_model_at_end: False
ignore_data_skip: False
restore_callback_states_from_checkpoint: False
full_determinism: False
seed: 42
data_seed: None
use_cpu: False
accelerator_config: {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}
parallelism_config: None
dataloader_drop_last: False
dataloader_num_workers: 0
dataloader_pin_memory: True
dataloader_persistent_workers: False
dataloader_prefetch_factor: None
remove_unused_columns: True
label_names: None
train_sampling_strategy: random
length_column_name: length
ddp_find_unused_parameters: None
ddp_bucket_cap_mb: None
ddp_broadcast_buffers: False
ddp_static_graph: None
ddp_backend: None
ddp_timeout: 1800
fsdp: None
fsdp_config: None
deepspeed: None
debug: []
skip_memory_metrics: True
do_predict: False
resume_from_checkpoint: None
warmup_ratio: None
local_rank: -1
prompts: None
batch_sampler: batch_sampler
multi_dataset_batch_sampler: round_robin
router_mapping: {}
learning_rate_mapping: {}

Training Logs

Epoch	Step	fiqa-test_cosine_ndcg@10
-1	-1	0.3923

Training Time

Training: 1.7 minutes

Framework Versions

Python: 3.12.3
Sentence Transformers: 5.6.1
Transformers: 5.10.1
PyTorch: 2.12.0+cu130
Accelerate: 1.13.0
Datasets: 4.8.5
Tokenizers: 0.22.2

Citation

BibTeX

Sentence Transformers

bibtex

@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

bibtex

@misc{oord2019representationlearningcontrastivepredictive,
      title={Representation Learning with Contrastive Predictive Coding},
      author={Aaron van den Oord and Yazhe Li and Oriol Vinyals},
      year={2019},
      eprint={1807.03748},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/1807.03748},
}

	sentence_0	sentence_1
type	string	string
modality	text	text
details	min: 8 tokens mean: 15.48 tokens max: 28 tokens	min: 22 tokens mean: 185.17 tokens max: 256 tokens