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 dimensions
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.

python

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("CatkinChen/BAAI_bge-base-en-v1.5_retrieval_finetuned_2025-03-29_23-29-50")
# Run inference
sentences = [
    'Who is the leader of the Order of the Phoenix?',
    "Book: 7, Chapter: 2\nPassage: The only things left out of place were today's Daily Prophet, still lying on the bed, and on top of it, the piece of broken mirror. Harry moved across the room, slid the mirror fragment off today's Prophet, and unfolded the newspaper. He had merely glanced at the headline when he had taken the rolled-up paper from the delivery owl early that morning and thrown it aside, after noting that it said nothing about Voldemort. Harry was sure that the Ministry was leaning on the Prophet to suppress news about Voldemort. It was only now, therefore, that he saw what he had missed. Across the bottom half of the front page a smaller headline was set over a picture of Dumbledore striding along looking harried:\nDUMBLEDORE - THE TRUTH AT LAST? Coming next week, the shocking story of the flawed genius considered by many to be the greatest wizard of his generation. Stripping away the popular image of serene, silver-bearded wisdom, Rita Skeeter reveals the disturbed childhood, the lawless youth, the lifelong feuds, and the guilty secrets that Dumbledore carried to his grave. WHY was the man tipped to be Minister of Magic content to remain a mere headmaster? WHAT was the real purpose of the secret organization known as the Order of the Phoenix? HOW did Dumbledore really meet his end? The answers to these and many more questions are explored in the explosive new biography, The Life and Lies of Albus Dumbledore, by Rita Skeeter, exclusively interviewed by Betty Braithwaite, page 13, inside.",
    'Book: 6, Chapter: 8\nPassage: "Really?" said Harry, taking care not to catch Ron\'s eye; the last time he had met Hagrid\'s half-brother, a vicious giant with a talent for ripping up trees by the roots, his vocabulary had comprised five words, two of which he was unable to pronounce properly. "Oh yeah, he\'s really come on," said Hagrid proudly. "Yeh\'ll be amazed. I\'m thinkin\' o\' trainin\' him up as me assistant." Ron snorted loudly, but managed to pass it off as a violent sneeze.',
]
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

Triplet

Dataset: test-eval
Evaluated with TripletEvaluator

Metric	Value
cosine_accuracy	0.8571

Training Details

Training Dataset

Unnamed Dataset

Size: 309 training samples
Columns: sentence_0, sentence_1, and sentence_2
Approximate statistics based on the first 309 samples:
sentence_0 sentence_1 sentence_2
type string string string
details
min: 9 tokens
mean: 17.74 tokens
max: 53 tokens
min: 46 tokens
mean: 212.92 tokens
max: 512 tokens
min: 35 tokens
mean: 180.01 tokens
max: 473 tokens

Samples:

sentence_0	sentence_1	sentence_2
`How does receiving the Hogwarts acceptance letter change Harry's life and reveal his true heritage?`	Book: 1, Chapter: 8[object Object]Passage: "Another Weasley, eh?" said Hagrid, glancing at Ron's freckles. "I spent half me life chasin' yer twin brothers away from the forest." The rock cakes were shapeless lumps with raisins that almost broke their teeth, but Harry and Ron pretended to be enjoying them as they told Hagrid all about their first lessons. Fang rested his head on Harry's knee and drooled all over his robes. Harry and Ron were delighted to hear Hagrid call Filch "that old git''[object Object]"An' as fer that cat, Mrs. Norris, I'd like ter introduce her to Fang sometime. D'yeh know, every time I go up ter the school, she follows me everywhere? Can't get rid of her - Filch puts her up to it."	`Book: 1, Chapter: 17[object Object]Passage: "But Snape always seemed to hate me so much." "Oh, he does," said Quirrell casually, "heavens, yes. He was at Hogwarts with your father, didn't you know? They loathed each other. But he never wanted you dead." "But I heard you a few days ago, sobbing - I thought Snape was threatening you. For the first time, a spasm of fear flitted across Quirrell's face.`
`Who is the supposed criminal believed to be after Harry?`	Book: 3, Chapter: 5[object Object]Passage: "He looks like one good hex would finish him off, doesn't he? Anyway ..." He turned to Harry. "What were you going to tell us?" Harry explained all about Mr. and Mrs. Weasley's argument and the warning Mr. Weasley had just given him. When he'd finished, Ron looked thunderstruck, and Hermione had her hands over her mouth. She finally lowered them to say, "Sirius Black escaped to come after you? Oh, Harry ... you'll have to be really, really careful. Don't go looking for trouble, Harry -"[object Object]"I don't go looking for trouble," said Harry, nettled. "Trouble usually finds me." "How thick would Harry have to be, to go looking for a nutter who wants to kill him?"	`Book: 2, Chapter: 6[object Object]Passage: He was about to follow Ron and Hermione inside when Lockhart's hand shot out. "Harry! I've been wanting a word - you don't mind if he's a couple of minutes late, do you, Professor Sprout?" Judging by Professor Sprout's scowl, she did mind, but Lockhart said, "That's the ticket," and closed the greenhouse door in her face. "Harry," said Lockhart, his large white teeth gleaming in the sunlight as he shook his head.`
`What is the name of the potion that allows the drinker to assume the form of someone else?`	`Book: 2, Chapter: 9[object Object]Passage: "What we'd need to do is to get inside the Slytherin common room and ask Malfoy a few questions without him realizing it's us." "But that's impossible," Harry said as Ron laughed. "No, it's not," said Hermione. "All we'd need would be some Polyjuice Potion." "What's that?" said Ron and Harry together. "Snape mentioned it in class a few weeks ago -"[object Object]"D'you think we've got nothing better to do in Potions than listen to Snape?" muttered Ron.`	Book: 5, Chapter: 4[object Object]Passage: And she hurried off downstairs again. Harry crossed the dingy landing, turned the bedroom doorknob, which was shaped like a serpent's head, and opened the door. He caught a brief glimpse of a gloomy high-ceilinged, twin-bedded room, then there was a loud twittering noise, followed by an even louder shriek, and his vision was completely obscured by a large quantity of very bushy hair - Hermione had thrown herself onto him in a hug that nearly knocked him flat, while Ron's tiny owl, Pigwidgeon, zoomed excitedly round and round their heads. "HARRY! Ron, he's here, Harry's here! We didn't hear you arrive! Oh, how are you? Are you all right? Have you been furious with us? I bet you have, I know our letters were useless - but we couldn't tell you anything, Dumbledore made us swear we wouldn't, oh, we've got so much to tell you, and you've got to tell us - the dementors! When we heard - and that Ministry hearing - it's just outrageous, I've looked it all up, they ...

Loss: TripletLoss with these parameters:
json
1{ 2 "distance_metric": "TripletDistanceMetric.COSINE", 3 "triplet_margin": 0.5 4}

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: steps
per_device_train_batch_size: 4
per_device_eval_batch_size: 4
num_train_epochs: 1
multi_dataset_batch_sampler: round_robin

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
eval_strategy: steps
prediction_loss_only: True
per_device_train_batch_size: 4
per_device_eval_batch_size: 4
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: 5e-05
weight_decay: 0.0
adam_beta1: 0.9
adam_beta2: 0.999
adam_epsilon: 1e-08
max_grad_norm: 1
num_train_epochs: 1
max_steps: -1
lr_scheduler_type: linear
lr_scheduler_kwargs: {}
warmup_ratio: 0.0
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: False
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}
tp_size: 0
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
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: batch_sampler
multi_dataset_batch_sampler: round_robin

Training Logs

Epoch	Step	test-eval_cosine_accuracy
0.1282	10	0.8143
0.2564	20	0.8286
0.3846	30	0.8357
0.5128	40	0.8500
0.6410	50	0.8429
0.7692	60	0.8500
0.8974	70	0.8571
1.0	78	0.8571
-1	-1	0.8571

Framework Versions

Python: 3.12.2
Sentence Transformers: 3.4.1
Transformers: 4.50.0
PyTorch: 2.4.1
Accelerate: 1.4.0
Datasets: 2.19.2
Tokenizers: 0.21.0

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",
}

TripletLoss

bibtex

@misc{hermans2017defense,
    title={In Defense of the Triplet Loss for Person Re-Identification},
    author={Alexander Hermans and Lucas Beyer and Bastian Leibe},
    year={2017},
    eprint={1703.07737},
    archivePrefix={arXiv},
    primaryClass={cs.CV}
}

	sentence_0	sentence_1	sentence_2
type	string	string	string
details	min: 9 tokens mean: 17.74 tokens max: 53 tokens	min: 46 tokens mean: 212.92 tokens max: 512 tokens	min: 35 tokens mean: 180.01 tokens max: 473 tokens