himanshu23099/bge_embedding_finetune_v3

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

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 semantic textual similarity, semantic search, paraphrase mining, text classification, clustering, and more.

Model Details

Model Description

• Model Type: Sentence Transformer
• Base model: BAAI/bge-small-en-v1.5
• Maximum Sequence Length: 512 tokens
• Output Dimensionality: 384 tokens
• 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': 384, '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("himanshu23099/bge_embedding_finetune_v3")
# Run inference
sentences = [
    'What is the ritual of Snan or bathing?',
    'Taking bath at the confluence of Ganga, Yamuna and invisible Saraswati during Mahakumbh has special significance. It is believed that by bathing in this holy confluence, all the sins of a person are washed away and he attains salvation.\n\nBathing not only symbolizes personal purification, but it also conveys the message of social harmony and unity, where people from different cultures and communities come together to participate in this sacred ritual.\n\nIt is considered that in special circumstances, the water of rivers also acquires a special life-giving quality, i.e. nectar, which not only leads to spiritual development along with purification of the mind, but also gives physical benefits by getting health.',
    'The art of knitting is a fascinating hobby that allows individuals to create beautiful and functional pieces from yarn. By intertwining strands of wool or cotton, one can produce items ranging from scarves to intricate sweaters. This craft has been passed down through generations, often bringing family members together for cozy evenings filled with creativity and conversation.\n\nKnitting not only provides a sense of accomplishment with every completed project but also promotes focus and relaxation, making it an excellent activity for reducing stress. Furthermore, the choice of colors and patterns can result in vibrant works of art, showcasing the unique style and personality of the knitter. Engaging in this craft often leads to new friendships within community groups that gather to share techniques and ideas, fostering a sense of belonging among enthusiasts.',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 384]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities.shape)
# [3, 3]

Evaluation

Metrics

Information Retrieval

• Dataset: val_evaluator
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.8156
cosine_accuracy@5	0.9929
cosine_accuracy@10	1.0
cosine_precision@1	0.8156
cosine_precision@5	0.1986
cosine_precision@10	0.1
cosine_recall@1	0.8156
cosine_recall@5	0.9929
cosine_recall@10	1.0
cosine_ndcg@5	0.9155
cosine_ndcg@10	0.918
cosine_ndcg@100	0.918
cosine_mrr@5	0.8891
cosine_mrr@10	0.8903
cosine_mrr@100	0.8903
cosine_map@100	0.8903
dot_accuracy@1	0.8156
dot_accuracy@5	0.9929
dot_accuracy@10	1.0
dot_precision@1	0.8156
dot_precision@5	0.1986
dot_precision@10	0.1
dot_recall@1	0.8156
dot_recall@5	0.9929
dot_recall@10	1.0
dot_ndcg@5	0.9155
dot_ndcg@10	0.918
dot_ndcg@100	0.918
dot_mrr@5	0.8891
dot_mrr@10	0.8903
dot_mrr@100	0.8903
dot_map@100	0.8903

Training Details

Training Dataset

Unnamed Dataset

• Size: 563 training samples
• Columns: anchor, positive, and negative
• Approximate statistics based on the first 563 samples:

  	anchor	positive	negative
  type	string	string	string
  details	min: 6 tokens mean: 16.33 tokens max: 30 tokens	min: 3 tokens mean: 93.51 tokens max: 402 tokens	min: 16 tokens mean: 109.62 tokens max: 269 tokens

• Samples:

  anchor	positive	negative
  Are there attached bathrooms in tents?	Attached bathroom facilities in tents vary by vendor and tent type. To know more about the availability of attached bathrooms, please reach out to your chosen Tent City vendor. For more information about these vendors and their services, please click here	The colors of the rainbow blend seamlessly across the canvas of the sky, creating a stunning visual display. Enjoying the beauty of nature can greatly enhance one's mood and inspire creativity. Take a moment to appreciate the vibrant hues and how they interact, as this can lead to a greater understanding of art and light. Exploring different forms of expression allows for personal growth and emotional exploration.
  Are there any discounts for senior citizens or children on buses traveling from the Bus Stand to the Mela?	No, there are no specific discounts available for senior citizens or children on buses traveling from the Bus Stand to the Mela. Standard ticket prices generally apply to all passengers.	The vibrant colors of autumn leaves create a breathtaking scene as they cascade gently to the ground. Local parks become havens for photographers and nature enthusiasts alike, capturing the fleeting beauty of the season. Crisp air invigorates leisurely strolls, while children gather acorns and pinecones, crafting treasures from nature’s bounty.
  Are there any luggage porter services available at Prayagraj Junction for pilgrims heading to the Mela?	Yes, luggage porter services are available at Prayagraj Junction for pilgrims heading to the Mela. These porters, often referred to as coolies	can be hired directly at the station to assist with carrying luggage from the train platform to your onward transport or directly to the Mela area.

• Loss: GISTEmbedLoss with these parameters:

  {'guide': SentenceTransformer(
    (0): Transformer({'max_seq_length': 256, 'do_lower_case': False}) with Transformer model: BertModel 
    (1): Pooling({'word_embedding_dimension': 384, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, '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()
  ), 'temperature': 0.01}
  

Evaluation Dataset

Unnamed Dataset

• Size: 141 evaluation samples
• Columns: anchor, positive, and negative
• Approximate statistics based on the first 141 samples:

  	anchor	positive	negative
  type	string	string	string
  details	min: 6 tokens mean: 16.05 tokens max: 30 tokens	min: 8 tokens mean: 88.91 tokens max: 324 tokens	min: 27 tokens mean: 104.84 tokens max: 262 tokens

• Samples:

  anchor	positive	negative
  What family-friendly tours are available?	All tours are designed with families in mind, ensuring a safe, comfortable, and enjoyable experience for all age groups. Whether traveling with children or elderly family members, the tours are structured to accommodate the needs of everyone in the group. Specific tours for senior citizens are also available. To explore them, click here : https://bit.ly/4eWFRoH	The majestic mountains rise against the azure sky, their peaks adorned with glistening snow that sparkles in the sunlight. deep valleys shelter hidden waterfalls, where crystal-clear waters cascade gracefully over rocks, creating a tranquil sound reverberating through the lush landscape. Wildlife thrives here, and one may spot elusive deer grazing in the early morning mist. As dusk settles, the horizon transforms into a canvas of vibrant hues, painting a breathtaking sunset that captivates the soul. Each season unveils unique beauty, inviting adventurers to explore its wonders.
  What are the charges for a private taxi or cab from Prayagraj Airport to the Mela grounds?	Private taxi charges are not fixed	The garden blooms vibrantly with colors and fragrances that attract butterflies and bees. Each petal holds a story from the earth, whispering tales of growth and resilience. Nearby, a small pond reflects the blue sky, while frogs leap joyfully on lily pads, creating ripples that dance across the surface. The sound of rustling leaves accompanies the gentle breeze, making nature's symphony a soothing backdrop for all who pause and appreciate this serene setting. As the sun sets, golden hues envelop the scene, inviting evening creatures to awaken under the twilight.
  What are the options for traveling to the Kumbh Mela if I arrive late at night at Prayagraj Junction?	If you arrive late at night at Prayagraj Junction for the Kumbh Mela, you have majorly 2 options for travel. 1. Taxi/Cabs: You can easily find 24/7 taxi services outside the railway station. Prepaid taxis are the most convenient and safe option. 2. Auto Rickshaws:Auto rickshaws are readily available outside the railway station.	The blooming desert blooms with vibrant colors as dusk approaches. Amidst the sands, ancient stories whisper through the wind, recalling journeys of nomads who tread lightly upon the earth. Some dance beneath the starlit skies, celebrating the beauty of freedom and the vastness of their surroundings. The nocturnal creatures awaken, each sound echoing tales of survival and adventure. Beyond the horizon, a tapestry of dreams unfurls, where every grain of sand holds the promise of a new discovery waiting to be unveiled.

• Loss: GISTEmbedLoss with these parameters:

  {'guide': SentenceTransformer(
    (0): Transformer({'max_seq_length': 256, 'do_lower_case': False}) with Transformer model: BertModel 
    (1): Pooling({'word_embedding_dimension': 384, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, '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()
  ), 'temperature': 0.01}
  

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• per_device_train_batch_size: 16
• gradient_accumulation_steps: 2
• learning_rate: 1e-05
• weight_decay: 0.01
• num_train_epochs: 90
• warmup_ratio: 0.1
• load_best_model_at_end: True

All Hyperparameters

Click to expand

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: steps
• prediction_loss_only: True
• per_device_train_batch_size: 16
• per_device_eval_batch_size: 8
• per_gpu_train_batch_size: None
• per_gpu_eval_batch_size: None
• gradient_accumulation_steps: 2
• eval_accumulation_steps: None
• torch_empty_cache_steps: None
• learning_rate: 1e-05
• weight_decay: 0.01
• adam_beta1: 0.9
• adam_beta2: 0.999
• adam_epsilon: 1e-08
• max_grad_norm: 1.0
• num_train_epochs: 90
• max_steps: -1
• lr_scheduler_type: linear
• 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
• 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
• eval_use_gather_object: False
• batch_sampler: batch_sampler
• multi_dataset_batch_sampler: proportional

Training Logs

Click to expand

Epoch	Step	Training Loss	Validation Loss	val_evaluator_cosine_map@100
0.5556	10	0.9623	0.5803	0.7676
1.1111	20	0.8653	0.5278	0.7684
1.6667	30	0.9346	0.4556	0.7692
2.2222	40	0.8058	0.3928	0.7687
2.7778	50	0.6639	0.3282	0.7723
3.3333	60	0.4974	0.2657	0.7784
3.8889	70	0.4447	0.2130	0.7877
4.4444	80	0.4309	0.1753	0.7922
5.0	90	0.2755	0.1320	0.7951
5.5556	100	0.3105	0.0826	0.8029
6.1111	110	0.1539	0.0479	0.8106
6.6667	120	0.22	0.0312	0.8141
7.2222	130	0.235	0.0173	0.8245
7.7778	140	0.1517	0.0119	0.8257
8.3333	150	0.1328	0.0095	0.8311
8.8889	160	0.1175	0.0055	0.8319
9.4444	170	0.1178	0.0037	0.8308
10.0	180	0.0598	0.0034	0.8338
10.5556	190	0.0958	0.0030	0.8324
11.1111	200	0.0681	0.0019	0.8331
11.6667	210	0.069	0.0013	0.8406
12.2222	220	0.0327	0.0009	0.8522
12.7778	230	0.0833	0.0006	0.8589
13.3333	240	0.0806	0.0005	0.8596
13.8889	250	0.0714	0.0004	0.8658
14.4444	260	0.0813	0.0004	0.8659
15.0	270	0.0512	0.0003	0.8676
15.5556	280	0.043	0.0003	0.8677
16.1111	290	0.0526	0.0003	0.8677
16.6667	300	0.0291	0.0002	0.8651
17.2222	310	0.0487	0.0002	0.8662
17.7778	320	0.054	0.0002	0.8621
18.3333	330	0.067	0.0002	0.8652
18.8889	340	0.0415	0.0002	0.8652
19.4444	350	0.0484	0.0002	0.8652
20.0	360	0.0304	0.0002	0.8690
20.5556	370	0.025	0.0002	0.8697
21.1111	380	0.0549	0.0002	0.8697
21.6667	390	0.0375	0.0002	0.8736
22.2222	400	0.0293	0.0002	0.8749
22.7778	410	0.0558	0.0002	0.8728
23.3333	420	0.0458	0.0002	0.8730
23.8889	430	0.0235	0.0002	0.8730
24.4444	440	0.0515	0.0002	0.8730
25.0	450	0.0337	0.0002	0.8734
25.5556	460	0.0376	0.0002	0.8734
26.1111	470	0.0189	0.0003	0.8734
26.6667	480	0.032	0.0002	0.8734
27.2222	490	0.025	0.0002	0.8695
27.7778	500	0.0258	0.0002	0.8704
28.3333	510	0.0351	0.0002	0.8681
28.8889	520	0.0285	0.0002	0.8679
29.4444	530	0.0263	0.0002	0.8679
30.0	540	0.0901	0.0002	0.8679
30.5556	550	0.0323	0.0001	0.8686
31.1111	560	0.0406	0.0001	0.8728
31.6667	570	0.0302	0.0001	0.8712
32.2222	580	0.0195	0.0001	0.8718
32.7778	590	0.0665	0.0001	0.8718
33.3333	600	0.0153	0.0001	0.8728
33.8889	610	0.0378	0.0001	0.8728
34.4444	620	0.0369	0.0001	0.8763
35.0	630	0.0238	0.0001	0.8706
35.5556	640	0.0275	0.0001	0.8720
36.1111	650	0.0469	0.0001	0.8708
36.6667	660	0.0438	0.0001	0.8788
37.2222	670	0.0333	0.0001	0.8800
37.7778	680	0.0186	0.0001	0.8765
38.3333	690	0.0308	0.0001	0.8765
38.8889	700	0.0713	0.0001	0.8767
39.4444	710	0.0188	0.0001	0.8767
40.0	720	0.0205	0.0001	0.8767
40.5556	730	0.0261	0.0001	0.8767
41.1111	740	0.0193	0.0001	0.8755
41.6667	750	0.0367	0.0000	0.8755
42.2222	760	0.0515	0.0000	0.8755
42.7778	770	0.0649	0.0000	0.8844
43.3333	780	0.0333	0.0000	0.8879
43.8889	790	0.0498	0.0000	0.8868
44.4444	800	0.0324	0.0000	0.8832
45.0	810	0.0321	0.0000	0.8832
45.5556	820	0.0354	0.0000	0.8832
46.1111	830	0.04	0.0000	0.8868
46.6667	840	0.0176	0.0000	0.8868
47.2222	850	0.0297	0.0000	0.8868
47.7778	860	0.0469	0.0000	0.8868
48.3333	870	0.025	0.0000	0.8868
48.8889	880	0.0425	0.0000	0.8868
49.4444	890	0.0475	0.0000	0.8868
50.0	900	0.0529	0.0000	0.8868
50.5556	910	0.0312	0.0000	0.8868
51.1111	920	0.0385	0.0000	0.8832
51.6667	930	0.0316	0.0000	0.8832
52.2222	940	0.0361	0.0000	0.8832
52.7778	950	0.053	0.0000	0.8832
53.3333	960	0.0226	0.0000	0.8868
53.8889	970	0.0781	0.0000	0.8868
54.4444	980	0.03	0.0000	0.8868
55.0	990	0.0349	0.0000	0.8832
55.5556	1000	0.0539	0.0000	0.8832
56.1111	1010	0.0351	0.0000	0.8832
56.6667	1020	0.0506	0.0000	0.8832
57.2222	1030	0.0204	0.0000	0.8832
57.7778	1040	0.0254	0.0000	0.8844
58.3333	1050	0.0274	0.0000	0.8844
58.8889	1060	0.001	0.0000	0.8844
59.4444	1070	0.049	0.0000	0.8844
60.0	1080	0.028	0.0000	0.8844
60.5556	1090	0.0477	0.0000	0.8844
61.1111	1100	0.0304	0.0000	0.8844
61.6667	1110	0.0188	0.0000	0.8844
62.2222	1120	0.0247	0.0000	0.8879
62.7778	1130	0.0428	0.0000	0.8879
63.3333	1140	0.0218	0.0000	0.8879
63.8889	1150	0.0476	0.0000	0.8868
64.4444	1160	0.021	0.0000	0.8868
65.0	1170	0.0435	0.0000	0.8856
65.5556	1180	0.0311	0.0000	0.8856
66.1111	1190	0.0275	0.0000	0.8856
66.6667	1200	0.0405	0.0000	0.8891
67.2222	1210	0.0009	0.0000	0.8891
67.7778	1220	0.0506	0.0000	0.8891
68.3333	1230	0.0538	0.0000	0.8891
68.8889	1240	0.0251	0.0000	0.8891
69.4444	1250	0.0168	0.0000	0.8891
70.0	1260	0.0527	0.0000	0.8903
70.5556	1270	0.0491	0.0000	0.8903
71.1111	1280	0.0092	0.0000	0.8903
71.6667	1290	0.0257	0.0000	0.8903
72.2222	1300	0.0455	0.0	0.8903
72.7778	1310	0.0271	0.0000	0.8903
73.3333	1320	0.04	0.0000	0.8903
73.8889	1330	0.0171	0.0000	0.8903
74.4444	1340	0.0157	0.0000	0.8903
75.0	1350	0.0323	0.0000	0.8903
75.5556	1360	0.0372	0.0000	0.8903
76.1111	1370	0.0109	0.0000	0.8903
76.6667	1380	0.0358	0.0000	0.8903
77.2222	1390	0.0279	0.0000	0.8903
77.7778	1400	0.0173	0.0000	0.8903
78.3333	1410	0.0409	0.0000	0.8903
78.8889	1420	0.0139	0.0000	0.8903
79.4444	1430	0.0123	0.0000	0.8903
80.0	1440	0.0232	0.0000	0.8903
80.5556	1450	0.0145	0.0000	0.8903
81.1111	1460	0.0261	0.0000	0.8903
81.6667	1470	0.0137	0.0000	0.8903
82.2222	1480	0.0146	0.0000	0.8903
82.7778	1490	0.0096	0.0000	0.8903
83.3333	1500	0.0245	0.0000	0.8903
83.8889	1510	0.0312	0.0000	0.8903
84.4444	1520	0.0174	0.0000	0.8903
85.0	1530	0.0437	0.0000	0.8903
85.5556	1540	0.0301	0.0000	0.8903
86.1111	1550	0.0119	0.0000	0.8903
86.6667	1560	0.0554	0.0000	0.8903
87.2222	1570	0.021	0.0000	0.8903
87.7778	1580	0.029	0.0000	0.8903
88.3333	1590	0.0132	0.0000	0.8903
88.8889	1600	0.0339	0.0000	0.8903
89.4444	1610	0.0412	0.0000	0.8903
90.0	1620	0.0847	0.0000	0.8903

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.10.12
• Sentence Transformers: 3.2.1
• Transformers: 4.44.2
• PyTorch: 2.5.0+cu121
• Accelerate: 0.34.2
• Datasets: 3.1.0
• Tokenizers: 0.19.1

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",
}

GISTEmbedLoss

@misc{solatorio2024gistembed,
    title={GISTEmbed: Guided In-sample Selection of Training Negatives for Text Embedding Fine-tuning},
    author={Aivin V. Solatorio},
    year={2024},
    eprint={2402.16829},
    archivePrefix={arXiv},
    primaryClass={cs.LG}
}