metadata

base_model: BAAI/bge-small-en
datasets:
  - sentence-transformers/hotpotqa
language:
  - en
library_name: sentence-transformers
license: apache-2.0
metrics:
  - cosine_accuracy
  - dot_accuracy
  - manhattan_accuracy
  - euclidean_accuracy
  - max_accuracy
pipeline_tag: sentence-similarity
tags:
  - sentence-transformers
  - sentence-similarity
  - feature-extraction
  - generated_from_trainer
  - dataset_size:76064
  - loss:MatryoshkaLoss
  - loss:TripletLoss
widget:
  - source_sentence: The person who released "Sun Arise" was born in what year?
    sentences:
      - >-
        Peter Frampton Peter Kenneth Frampton (born 22 April 1950) is an English
        rock musician, singer, songwriter, producer, and guitarist. He was
        previously associated with the bands Humble Pie and The Herd. At the end
        of his 'group' career was Frampton's international breakthrough album
        his live release, "Frampton Comes Alive!" The album sold in the United
        States more than 8 million copies and spawned several single hits. Since
        then he has released several major albums. He has also worked with David
        Bowie and both Matt Cameron and Mike McCready from Pearl Jam, among
        others.
      - >-
        Sun Arise "Sun Arise" is the fourth single released by the Australian
        singer-songwriter Rolf Harris. Released in January 1961 in Australia and
        October 1962 in the UK, it was Harris' third charting hit in Australia
        (following "The Big Black Hat" in 1960) and second in the UK (following
        "Tie Me Kangaroo Down, Sport" also 1960). Unlike his early chart hits,
        "Sun Arise" was not a comedy record, but came within the genre of world
        music with its didgeridoo-inspired sound.
      - >-
        Circa Survive Circa Survive is an American rock band from the
        Philadelphia suburb of Doylestown, formed in 2004. The band, led by
        Anthony Green, consists of former members from Saosin, This Day Forward,
        and Taken.
  - source_sentence: >-
      What year was Chuang Chia-jung's partner in the 2010 MPS Group
      Championships – Doubles born?
    sentences:
      - >-
        Ko Olina Station and Center Ko Olina Station and Ko Olina Center make up
        a lifestyle center in the resort town of Ko Olina, a neighborhood in
        Kapolei, Hawaii. The shopping mall opened in 2009 and consists of two
        centers located across a street from each other. Ko Olina Station
        debuted in 2009, while the more recent Ko Olina Center finished
        construction in 2010. The centers contain a total of approximately 31
        retail tenants, with the majority of them being native Hawaiian
        businesses, such as ABC Stores and Peter Merriman's MonkeyPod Kitchen.
      - >-
        2010 MPS Group Championships – Doubles Chuang Chia-jung and Sania Mirza
        were the defenders of championship title, but Mirza chose not to
        compete.
      - >-
        Lu Chia-hung Lu Chia-hung (; born 4 March 1997) is a Taiwanese male
        badminton player.
  - source_sentence: >-
      What son of Zeus in Greek mythology was said to have fatheres an Argonaut
      seer?
    sentences:
      - >-
        All Net Resort and Arena All Net Resort and Arena is a planned
        entertainment complex in Las Vegas. A project of businessman and former
        basketball player Jackie Robinson, the complex would encompass a resort
        hotel, a retail and restaurant streetscape, and a multi-purpose indoor
        arena with a retractable roof. Its location is set on the Strip at the
        former site of a Wet 'n Wild waterpark, next to the SLS Las Vegas in
        Winchester, Nevada. Designed by the Cuningham Group, it was planned to
        open in 2017, but is delayed until 2018 or 2019.
      - >-
        Piras (mythology) In Greek mythology, Piras (Ancient Greek: Πείραντα)
        was a king of Argos, otherwise also known as Piren, Peiren, Peiras,
        Peirasus and Piranthus.
      - >-
        Idmon In Greek mythology, Idmon was an Argonaut seer. Allegedly a son of
        Apollo, he had Abas (or Ampycus) as his mortal father. His mother was
        Asteria, daughter of Coronus, or Cyrene, or else Antianeira, daughter of
        Pheres. By Laothoe he had a son Thestor. Idmon foresaw his own death in
        the Argonaut expedition, but joined anyway. During the outbound voyage
        of "Argo", a boar killed him in the land of the Mariandyni, in Bithynia.
  - source_sentence: >-
      In what year was the drama film in which Dorothy Duffy played Rose /
      Patricia released?
    sentences:
      - >-
        Keith Davis (safety) Keith Lamont Davis (born December 30, 1978) is a
        former American football safety in the National Football League for the
        Dallas Cowboys. He played college football at Sam Houston State
        University.
      - >-
        Dorothy Duffy Dorothy Duffy (born in Douglas Bridge, Northern Ireland)
        is an Irish actress. She is best known for her performance as Rose /
        Patricia in "The Magdalene Sisters".
      - >-
        The Franchise Affair (film) The Franchise Affair is a 1951 British
        thriller film directed by Lawrence Huntington and starring Michael
        Denison, Dulcie Gray, Anthony Nicholls and Marjorie Fielding. It is a
        faithful adaptation of the novel "The Franchise Affair" by Josephine
        Tey.
  - source_sentence: Was McDull, Kung Fu Kindergarten or Pettson and Findus created first?
    sentences:
      - >-
        Tabaluga Tabaluga is a media franchise featuring a fictional little
        green Dragon of the same name, created by German Rock musician Peter
        Maffay, children's songwriter and the author . The artist Helme Heine
        drew the image of Tabaluga as it is currently known. The character
        Tabaluga was first introduced by Peter Maffay in a musical fairy tale
        "Tabaluga ... oder die Reise zur Vernunft" (Tabaluga or... The Journey
        to Reason) in 1983. This first studio album was the step to success:
        within the next years some Helme Heine books, four sequel concept studio
        albums, two resounding tours, a stage musical, "Tabaluga und Lilli"
        ("Tabaluga and Lilli"), based on the third concept album and many TV
        Cartoons which have been broadcasting in over 100 countries round the
        world followed and a children's game show. Over 100 kindergartens and
        child care groups carry the word "Tabaluga" in their names.
      - >-
        2005–06 FC Bayern Munich season FC Bayern Munich won the domestic
        double, beating Werder Bremen by five points in Bundesliga, and
        defeating Eintracht Frankfurt 1–0 in the DFB-Pokal final, thanks to a
        goal from Claudio Pizarro. The season was in spite of that tainted due
        to a big defeat to Milan in the UEFA Champions League, losing out 5–2 on
        aggregate in the Last 16. At the end of the season, Bayern signed German
        football's wonderkid Lukas Podolski from Köln.
      - >-
        Pettson and Findus Pettson and Findus (Swedish: "Pettson och Findus" )
        is a series of children's books written and illustrated by Swedish
        author Sven Nordqvist. The books feature an old farmer (Pettson) and his
        cat (Findus) who live in a small ramshackle farmhouse in the
        countryside. The first of the Pettson och Findus book to be published
        was "Pannkakstårtan" in 1984 (first published in English in 1985 as
        "Pancake Pie").
model-index:
  - name: BGE-base-en-v1.5-Hotpotqa
    results:
      - task:
          type: triplet
          name: Triplet
        dataset:
          name: dim 384
          type: dim_384
        metrics:
          - type: cosine_accuracy
            value: 0.8853525792711784
            name: Cosine Accuracy
          - type: dot_accuracy
            value: 0.11464742072882159
            name: Dot Accuracy
          - type: manhattan_accuracy
            value: 0.8862991008045433
            name: Manhattan Accuracy
          - type: euclidean_accuracy
            value: 0.8853525792711784
            name: Euclidean Accuracy
          - type: max_accuracy
            value: 0.8862991008045433
            name: Max Accuracy
      - task:
          type: triplet
          name: Triplet
        dataset:
          name: dim 256
          type: dim_256
        metrics:
          - type: cosine_accuracy
            value: 0.8840511121628017
            name: Cosine Accuracy
          - type: dot_accuracy
            value: 0.11571225745385708
            name: Dot Accuracy
          - type: manhattan_accuracy
            value: 0.8851159488878372
            name: Manhattan Accuracy
          - type: euclidean_accuracy
            value: 0.8841694273544723
            name: Euclidean Accuracy
          - type: max_accuracy
            value: 0.8851159488878372
            name: Max Accuracy
      - task:
          type: triplet
          name: Triplet
        dataset:
          name: dim 128
          type: dim_128
        metrics:
          - type: cosine_accuracy
            value: 0.8829862754377662
            name: Cosine Accuracy
          - type: dot_accuracy
            value: 0.11831519167061051
            name: Dot Accuracy
          - type: manhattan_accuracy
            value: 0.8823946994794132
            name: Manhattan Accuracy
          - type: euclidean_accuracy
            value: 0.8836961665877898
            name: Euclidean Accuracy
          - type: max_accuracy
            value: 0.8836961665877898
            name: Max Accuracy
      - task:
          type: triplet
          name: Triplet
        dataset:
          name: dim 64
          type: dim_64
        metrics:
          - type: cosine_accuracy
            value: 0.8815664931377188
            name: Cosine Accuracy
          - type: dot_accuracy
            value: 0.12434926644581164
            name: Dot Accuracy
          - type: manhattan_accuracy
            value: 0.88180312352106
            name: Manhattan Accuracy
          - type: euclidean_accuracy
            value: 0.88180312352106
            name: Euclidean Accuracy
          - type: max_accuracy
            value: 0.88180312352106
            name: Max Accuracy

BGE-base-en-v1.5-Hotpotqa

This is a sentence-transformers model finetuned from BAAI/bge-small-en on the sentence-transformers/hotpotqa dataset. 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
Maximum Sequence Length: 512 tokens
Output Dimensionality: 384 tokens
Similarity Function: Cosine Similarity
Training Dataset:
- sentence-transformers/hotpotqa
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': 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("sentence_transformers_model_id")
# Run inference
sentences = [
    'Was McDull, Kung Fu Kindergarten or Pettson and Findus created first?',
    'Pettson and Findus Pettson and Findus (Swedish: "Pettson och Findus" ) is a series of children\'s books written and illustrated by Swedish author Sven Nordqvist. The books feature an old farmer (Pettson) and his cat (Findus) who live in a small ramshackle farmhouse in the countryside. The first of the Pettson och Findus book to be published was "Pannkakstårtan" in 1984 (first published in English in 1985 as "Pancake Pie").',
    'Tabaluga Tabaluga is a media franchise featuring a fictional little green Dragon of the same name, created by German Rock musician Peter Maffay, children\'s songwriter and the author . The artist Helme Heine drew the image of Tabaluga as it is currently known. The character Tabaluga was first introduced by Peter Maffay in a musical fairy tale "Tabaluga ... oder die Reise zur Vernunft" (Tabaluga or... The Journey to Reason) in 1983. This first studio album was the step to success: within the next years some Helme Heine books, four sequel concept studio albums, two resounding tours, a stage musical, "Tabaluga und Lilli" ("Tabaluga and Lilli"), based on the third concept album and many TV Cartoons which have been broadcasting in over 100 countries round the world followed and a children\'s game show. Over 100 kindergartens and child care groups carry the word "Tabaluga" in their names.',
]
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

Triplet

Dataset: dim_384
Evaluated with TripletEvaluator

Metric	Value
cosine_accuracy	0.8854
dot_accuracy	0.1146
manhattan_accuracy	0.8863
euclidean_accuracy	0.8854
max_accuracy	0.8863

Triplet

Dataset: dim_256
Evaluated with TripletEvaluator

Metric	Value
cosine_accuracy	0.8841
dot_accuracy	0.1157
manhattan_accuracy	0.8851
euclidean_accuracy	0.8842
max_accuracy	0.8851

Triplet

Dataset: dim_128
Evaluated with TripletEvaluator

Metric	Value
cosine_accuracy	0.883
dot_accuracy	0.1183
manhattan_accuracy	0.8824
euclidean_accuracy	0.8837
max_accuracy	0.8837

Triplet

Dataset: dim_64
Evaluated with TripletEvaluator

Metric	Value
cosine_accuracy	0.8816
dot_accuracy	0.1243
manhattan_accuracy	0.8818
euclidean_accuracy	0.8818
max_accuracy	0.8818

Training Details

Training Dataset

sentence-transformers/hotpotqa

Dataset: sentence-transformers/hotpotqa at f07d3cd
Size: 76,064 training samples
Columns: anchor, positive, and negative

Approximate statistics based on the first 1000 samples:

	anchor	positive	negative
type	string	string	string
details	min: 7 tokens mean: 25.02 tokens max: 103 tokens	min: 19 tokens mean: 100.08 tokens max: 315 tokens	min: 13 tokens mean: 89.42 tokens max: 375 tokens

Samples:

anchor	positive	negative
`What type of songs is the singer of Saahore Baahubali best known for?`	`Saahore Baahubali "Saahore Baahubali" (English: Glory be to Baahubali) is a Telugu song from the 2017 film . Sung by Daler Mehndi, the song is composed by M. M. Keeravani, with lyrics penned by his father Siva Shakti Datta and Kodi Ramakrishna. Most of the lyrics were composed in Sanskrit.`	Anupama Deshpande Anupama Deshapande is a Bollywood playback singer who has won the Filmfare Award for Best Female Playback Singer for her folk song "Sohni Chinab Di" in the film "Sohni Mahiwal" (1984). This song was originally meant for Asha Bhonsle who since was busy those days. Therefore, Annu Malik recorded this song in the voice of Anupama Deshpande so that it could later on dubbed by Asha Bhonsle. But on listening the song, Asha Bhonsle sportingly advised to retain the song as it was, in the voice of Anupama Deshpande by giving full credit to the anupama's singing talent. She has sung a total of 124 songs in 92 films.
`'Dot TV' was owned and operated by a Pan-European satellite broadcasting, on-demand internet streaming media, broadband and telephone services company with headquarters where?`	.tv (TV channel) .tv (Pronounced as 'Dot TV', referred to onscreen as .tv - the technology channel) was a British television channel dedicated to technology. .tv was owned and operated by British Sky Broadcasting. The channel began broadcasting on 1 September 1996 as "The Computer Channel" and broadcast between 18:00 and 20:00. The broadcasting hours were increased to midday-midnight when "The Computer Channel" (later .tv) started broadcasting on British Sky Broadcasting's digital satellite platform, Sky Digital in 1998. In 1999 the channel interviewed then Microsoft CEO Bill Gates.	`Movistar TV Movistar TV is an IPTV service operated by Telefónica. The service was started as a commercial test pilot in the city of Alicante in 2001 and later extended to some major cities such as Madrid and Barcelona in April 2004. In 2013, Movistar Imagenio was rebranded to Movistar TV.`
`Elvira Madigan's father was born in what year?`	Gisela Brož Gisela Antonia Brož (Brosch) (also sometimes referred to as Gisela Madigan), (4 April 1865 - 1945) was an Austrian-American circus performer, tight rope dancer, and clown. Her parents were shoemaker Joseph Brož and his wife Maria. She went to convent school in Siebenbürgen and at the age of 15 she got to know the circus family Madigans with John and Laura who at that time toured with circus Krembser in Vienna. Gisela became their foster child and got to learn tight rope dancing, this along with the couple's two year younger daughter Elvira Madigan.	`Elvira Casazza Elvira Casazza (15 November 1887 – 24 January 1965) was an Italian mezzo-soprano opera singer (also known as Elvira Mari-Casazza). One of Toscanini's favourite singers, she was considered an outstanding interpreter of Mistress Quickly in Verdi's "Falstaff" during the 1920s and created several roles in Italian operas of the early 20th century.`

Loss: MatryoshkaLoss with these parameters:

{
    "loss": "TripletLoss",
    "matryoshka_dims": [
        384,
        256,
        128,
        64
    ],
    "matryoshka_weights": [
        1,
        1,
        1,
        1
    ],
    "n_dims_per_step": -1
}

Evaluation Dataset

sentence-transformers/hotpotqa

Dataset: sentence-transformers/hotpotqa at f07d3cd
Size: 8,452 evaluation samples
Columns: anchor, positive, and negative

Approximate statistics based on the first 1000 samples:

	anchor	positive	negative
type	string	string	string
details	min: 10 tokens mean: 25.14 tokens max: 130 tokens	min: 20 tokens mean: 102.4 tokens max: 512 tokens	min: 13 tokens mean: 88.09 tokens max: 358 tokens

Samples:

anchor	positive	negative
`When was the English former professional footballer which Tslil Sela has an alledged relationship with born?`	`Tslil Sela Tslil Sela (Hebrew: צליל סלע‎ ‎ , born 26 October 1987) is an Israeli model, most known for her modeling work and for her alleged relationship with English footballer Rio Ferdinand. Sela is leading the campaign for KOOI fashion 2010, and Sanyang Motorcycles (SYM Motors) in Israel.`	`Sam Collins (English footballer) Samuel Jason Collins (born 5 June 1977) is an English football manager and former footballer who played as a defender. His brother, Simon, is also a former professional footballer and manager.`
`Gebhard Leberecht von Blucher the Prussian Generalfieldmarschall led his army against this famous commander in the Battle of Lepzig?`	`Gebhard Leberecht von Blücher Gebhard Leberecht von Blücher, Fürst von Wahlstatt (] ; 16 December 1742 – 12 September 1819), "Graf" (count), later elevated to "Fürst" (sovereign prince) von Wahlstatt, was a Prussian "Generalfeldmarschall" (field marshal). He earned his greatest recognition after leading his army against Napoleon I at the Battle of the Nations at Leipzig in 1813 and the Battle of Waterloo in 1815.`	Karl Freiherr von Müffling Friedrich Karl Ferdinand Freiherr von Müffling, called Weiss (12 June 177510 January 1851) was a Prussian "Generalfeldmarschall" and military theorist. He served as Blücher's liaison officer in Wellington's headquarters during the Battle of Waterloo and was one of the organizers of the final victory over Napoleon. After the wars he served a diplomatic role at the Congress of Aix-la-Chappelle and was a major contributor to the development of the Prussian General Staff as Chief. Müffling also specialized in military topography and cartography.
`The Platonia Dilemma was introduced in the book "Metamagical Themas" which was written by an author born in what year?`	Platonia dilemma In the platonia dilemma introduced in Douglas Hofstadter's book "Metamagical Themas", an eccentric trillionaire gathers 20 people together, and tells them that if one and only one of them sends him a telegram (reverse charges) by noon the next day, that person will receive a billion dollars. If he receives more than one telegram, or none at all, no one will get any money, and cooperation between players is forbidden. In this situation, the superrational thing to do is to send a telegram with probability 1/20.	John Alexander Stewart (philosopher) John Alexander Stewart (19 October 1846 – 27 December 1933) was a Scottish writer, educator and philosopher. He was a university professor and classical lecturer at Christ Church, Oxford from 1875 to 1883, White's Professor of Moral Philosophy at Oxford, and professorial fellow of Corpus Christi College, from 1897 to his retirement in 1927. Throughout his academic career, he was an editor and author of works on Aristotle and considered one of the foremost experts on the subject. His best known books were "Notes on the Nicomachean Ethics of Aristotle" (1892) and "The Myths of Plato" (1905).

Loss: MatryoshkaLoss with these parameters:

{
    "loss": "TripletLoss",
    "matryoshka_dims": [
        384,
        256,
        128,
        64
    ],
    "matryoshka_weights": [
        1,
        1,
        1,
        1
    ],
    "n_dims_per_step": -1
}

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: steps
per_device_train_batch_size: 32
per_device_eval_batch_size: 32
gradient_accumulation_steps: 16
learning_rate: 2e-05
num_train_epochs: 20
lr_scheduler_type: cosine
warmup_ratio: 0.1
bf16: True
tf32: True
load_best_model_at_end: True
optim: adamw_torch_fused
resume_from_checkpoint: bge-small-hotpotwa-matryoshka
batch_sampler: no_duplicates

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
eval_strategy: steps
prediction_loss_only: True
per_device_train_batch_size: 32
per_device_eval_batch_size: 32
per_gpu_train_batch_size: None
per_gpu_eval_batch_size: None
gradient_accumulation_steps: 16
eval_accumulation_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: 20
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: True
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: bge-small-hotpotwa-matryoshka
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
batch_sampler: no_duplicates
multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	Training Loss	loss	dim_128_cosine_accuracy	dim_256_cosine_accuracy	dim_384_cosine_accuracy	dim_64_cosine_accuracy
0.3366	50	19.5492	19.2604	0.9585	0.9657	0.9663	0.9432
0.6731	100	19.1976	18.2958	0.9359	0.9392	0.9425	0.9276
1.0097	150	18.4746	16.9846	0.9053	0.9075	0.9085	0.8996
1.3462	200	18.0684	16.6869	0.9030	0.9051	0.9049	0.8959
1.6828	250	17.8979	16.5780	0.9017	0.9030	0.9016	0.8954
2.0194	300	17.7545	16.5135	0.8977	0.8991	0.8984	0.8925
2.3559	350	17.6046	16.4917	0.8894	0.8894	0.8907	0.8862
2.6925	400	17.4434	16.4926	0.8874	0.8862	0.8875	0.8858
3.0290	450	17.3278	16.4757	0.8854	0.8861	0.8869	0.8859
3.3656	500	17.247	16.4735	0.8830	0.8841	0.8854	0.8816

Framework Versions

Python: 3.10.10
Sentence Transformers: 3.0.1
Transformers: 4.41.2
PyTorch: 2.1.2+cu121
Accelerate: 0.33.0
Datasets: 2.19.1
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",
}

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}
}

TripletLoss

@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}
}