checkpoint-112/README.md

metadata

language:
  - en
tags:
  - sentence-transformers
  - sentence-similarity
  - feature-extraction
  - generated_from_trainer
  - dataset_size:638837
  - loss:GISTEmbedLoss
base_model: BAAI/bge-m3
widget:
  - source_sentence: >-
      The Co-operative has beaten sales growth at its major supermarket rivals
      for the first time in five years.
    sentences:
      - >-
        Evans had been due to remain with the Saddlers until 26 January but was
        recalled from his loan on Saturday.

        The 21-year-old, who scored four goals in 16 appearances on loan at
        Walsall, moves on a three-and-a-half-year deal.

        "I'm sure he will become an even better player working in our
        environment at Reading," said manager Brian McDermott.

        Find all the latest football transfers on our dedicated page.
      - >-
        A jury decided the case against Peter Barrowman, 35, was not proven
        following a trial at the High Court in Glasgow.

        Speaking outside court, the father-of-two from Stenhousemuir told BBC
        Scotland that the verdict was a "massive relief".

        He said: "I am so happy that I can go home. I can't really put it into
        words, but I'm just very happy."

        Mr Barrowman, who worked at Cornton Vale prison, is currently suspended
        by the Nursing and Midwifery Council.

        Asked if he would seek a return to nursing, he replied: "That's a
        decision I've still to make."

        Three women - all prisoners in the jail - claimed Mr Barrowman raped
        them between January and March 2014 while they were his patients.

        He denied any sexual contact with the women and claimed they were lying.

        The nurse was suspended from working at Cornton Vale in March 2014,
        after the allegations were made and he was dismissed in July last year.

        Mr Barrowman was also acquitted of charges of supplying drugs to
        prisoners. The charges were withdrawn by the Crown before the end of the
        trial.
      - >-
        The Co-op's sales grew by 1.4% in the 12 weeks to 31 January, according
        to research company Kantar Worldpanel.

        People shopped there most frequently, at an average 19 times over the
        period, compared with a market average of 11 visits.

        In contrast, Sainsbury's was the only major chain to report rising
        sales, up 0.6%.

        Revenues at Tesco fell by 1.6% over the last three months, although
        Kantar said it was the supermarket group's best reading since last
        September.

        At Morrisons, sales shrank by 2.2%, while at Asda, which is owned by US
        chain Walmart, revenue fell by 3.8%.

        Sainsbury's, which recently made a £1.3bn bid for Argos-owner Home
        Retail Group, marginally increased its share of the UK market to 16.8%,
        ahead of Asda's 16.2% share and Morrisons with 10.8%. Tesco remains the
        market leader with 28.5%.

        The Co-op maintained its 5.9% share of the British market.

        Kantar said it was the first time since 2011 that the Co-op had
        surpassed its non-discounter rivals in terms of sales.

        The retail-to-funerals group, the UK's largest mutual, has been beset
        with problems in recent years, most notably at the Co-operative Bank,
        which is now majority owned by bondholders, although Co-op Group retains
        a 20% stake.

        The discounters, Lidl and Aldi, both saw double-digit revenue growth, up
        18.7% and 13.7% respectively. At the same time, Waitrose's sales were up
        0.1% for the three months.
  - source_sentence: A standing man talks to a seated woman while on some sort of vehicle.
    sentences:
      - Two people are talking.
      - A photo set of two women at work
      - No two people are talking.
  - source_sentence: who knows the art of chakravyuh in mahabharat
    sentences:
      - >-
        History of the Detroit Red Wings The Detroit Red Wings professional ice
        hockey club was founded as the Detroit Cougars on September 25, 1926,
        one of three teams to join the National Hockey League (NHL) in 1926.
        With the demise of the Western Canada Hockey League (WCHL), the rights
        to the players of the Victoria Cougars were purchased by a Detroit group
        led by Charles A. Hughes who kept the name "Cougars" for their NHL club.
        The new team struggled financially; in 1930, the Cougars changed their
        name to the Detroit Falcons, and after being bought out of receivership
        by James E. Norris were renamed as the Detroit Red Wings in 1932. The
        team played their first game on November 18, 1926, and won their first
        two Stanley Cup titles in 1936 and 1937. The Red Wings have won the Cup
        eleven times, more than any other American team in NHL history.
      - >-
        Padmavyuha The Chakravyūha or Padmavyūha was a very special formation
        (vyuha), and knowledge of how to penetrate it was limited to only a
        handful of warriors on the Pandavas' side, namely: Abhimanyu, Arjuna,
        Krishna and Pradyumna, of which only Abhimanyu was present when the
        Kauravas used it on the battlefield. In the Mahabharata it is mentioned
        that Abhimanyu learnt about the Chakravyūha while in his mother's womb
        but he was not able to hear how to escape the formation. After Abhimanyu
        had penetrated the sixth tier of the formation, all the Kauravas'
        commanders attacked him simultaneously, which was against the righteous
        rules of warfare Dharmayuddha, and gradually exhausted and killed
        him.[2]
      - "Ford Power Stroke engine The 7.3Â\_L DI Power Stroke was in production until the second quarter of model year 2003 when it was replaced by the 6.0L because of its inability to meet newer emission requirements. Nearly 2 million 7.3s were produced from International's Indianapolis plant.[2]"
  - source_sentence: who elect the deputy chairman of rajya sabha
    sentences:
      - >-
        Mishael Morgan Marie-Charms Mishael Morgan (born July 15, 1986) known
        professionally as Mishael Morgan, is a Trinidadian-Canadian actress
        known for the role of Hilary Curtis on CBS Daytime soap opera, The Young
        and the Restless.
      - >-
        Rajya Sabha The Vice President of India (currently, Venkaiah Naidu) is
        the ex-officio Chairman of the Rajya Sabha, who presides over its
        sessions. The Deputy Chairman, who is elected from amongst the house's
        members, takes care of the day-to-day matters of the house in the
        absence of the Chairman. The Rajya Sabha held its first sitting on 13
        May 1952.[6] The salary and other benefits for a member of Rajya Sabha
        are same as for a member of Lok Sabha.
      - >-
        History of film D. W. Griffith had the highest standing among American
        directors in the industry, because of the dramatic excitement he
        conveyed to the audience through his films. The American film industry,
        or "Hollywood", as it was becoming known after its new geographical
        center in Hollywood, a neighborhood in Los Angeles, California, gained
        the position it has held, more or less, ever since: film factory for the
        world and exporting its product to most countries. By the 1920s, the
        United States reached what is still its era of greatest-ever output,
        producing an average of 800 feature films annually,[1] or 82% of the
        global total (Eyman, 1997). During late 1927, Warner's released The Jazz
        Singer, with the first synchronized dialogue (and singing) in a feature
        film. By the end of 1929, Hollywood was almost all-talkie, with several
        competing sound systems (soon to be standardized). Sound saved the
        Hollywood studio system in the face of the Great Depression (Parkinson,
        1995).
  - source_sentence: >-
      In Sri Lanka , the title of Chartered Accountant ( CA Sri Lanka ) can be
      used by only members of the Institute of Chartered Accountants of Sri
      Lanka .
    sentences:
      - Holly Holly was influenced by Elton John musically .
      - >-
        In Sri Lanka , the title of an accountant ( CA Sri Lanka ) can only be
        used by members of the Institute of Accountants in Sri Lanka .
      - >-
        He was the half-brother of Lord Alfred Paget , Lord George Paget and
        Lord Clarence Paget .
datasets:
  - bobox/enhanced_NLI-50K
  - sentence-transformers/natural-questions
  - tals/vitaminc
  - bobox/xSum-processed
  - google-research-datasets/paws
pipeline_tag: sentence-similarity
library_name: sentence-transformers
metrics:
  - pearson_cosine
  - spearman_cosine
  - cosine_accuracy
  - cosine_accuracy_threshold
  - cosine_f1
  - cosine_f1_threshold
  - cosine_precision
  - cosine_recall
  - cosine_ap
  - cosine_mcc
model-index:
  - name: SentenceTransformer based on BAAI/bge-m3
    results:
      - task:
          type: semantic-similarity
          name: Semantic Similarity
        dataset:
          name: sts test
          type: sts-test
        metrics:
          - type: pearson_cosine
            value: 0.9037962998590239
            name: Pearson Cosine
          - type: spearman_cosine
            value: 0.9228968065127588
            name: Spearman Cosine
      - task:
          type: binary-classification
          name: Binary Classification
        dataset:
          name: allNLI dev
          type: allNLI-dev
        metrics:
          - type: cosine_accuracy
            value: 0.7734375
            name: Cosine Accuracy
          - type: cosine_accuracy_threshold
            value: 0.7482618689537048
            name: Cosine Accuracy Threshold
          - type: cosine_f1
            value: 0.6769230769230768
            name: Cosine F1
          - type: cosine_f1_threshold
            value: 0.694869875907898
            name: Cosine F1 Threshold
          - type: cosine_precision
            value: 0.5789473684210527
            name: Cosine Precision
          - type: cosine_recall
            value: 0.8148148148148148
            name: Cosine Recall
          - type: cosine_ap
            value: 0.6542135747366074
            name: Cosine Ap
          - type: cosine_mcc
            value: 0.5058060457173612
            name: Cosine Mcc
      - task:
          type: binary-classification
          name: Binary Classification
        dataset:
          name: Qnli dev
          type: Qnli-dev
        metrics:
          - type: cosine_accuracy
            value: 0.69921875
            name: Cosine Accuracy
          - type: cosine_accuracy_threshold
            value: 0.6639883518218994
            name: Cosine Accuracy Threshold
          - type: cosine_f1
            value: 0.6956521739130435
            name: Cosine F1
          - type: cosine_f1_threshold
            value: 0.6639883518218994
            name: Cosine F1 Threshold
          - type: cosine_precision
            value: 0.676923076923077
            name: Cosine Precision
          - type: cosine_recall
            value: 0.7154471544715447
            name: Cosine Recall
          - type: cosine_ap
            value: 0.7520437823272439
            name: Cosine Ap
          - type: cosine_mcc
            value: 0.3994014078011956
            name: Cosine Mcc

SentenceTransformer based on BAAI/bge-m3

This is a sentence-transformers model finetuned from BAAI/bge-m3 on the NLI, natural-questions, vitaminc, xsum, paws and global_dataset datasets. It maps sentences & paragraphs to a 1024-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-m3
• Maximum Sequence Length: 8192 tokens
• Output Dimensionality: 1024 dimensions
• Similarity Function: Cosine Similarity
• Training Datasets:
  • NLI
  • natural-questions
  • vitaminc
  • xsum
  • paws
  • global_dataset
• Language: en

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': 8192, 'do_lower_case': False}) with Transformer model: XLMRobertaModel 
  (1): AdvancedWeightedPooling(
    (mha): MultiheadAttention(
      (out_proj): NonDynamicallyQuantizableLinear(in_features=1024, out_features=1024, bias=True)
    )
    (MLP): Sequential(
      (0): SwiGLUBlock(
        (in_proj_swish): Linear(in_features=1024, out_features=2048, bias=True)
        (in_proj_gate): Linear(in_features=1024, out_features=2048, bias=True)
      )
      (1): Dropout(p=0.05, inplace=False)
      (2): Linear(in_features=2048, out_features=1024, bias=True)
    )
    (layernorm): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
  )
)

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("bobox/XLMRoBERTaM3-CustomPoolin-v1-s1-checkpoints-tmp")
# Run inference
sentences = [
    'In Sri Lanka , the title of Chartered Accountant ( CA Sri Lanka ) can be used by only members of the Institute of Chartered Accountants of Sri Lanka .',
    'In Sri Lanka , the title of an accountant ( CA Sri Lanka ) can only be used by members of the Institute of Accountants in Sri Lanka .',
    'Holly Holly was influenced by Elton John musically .',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 1024]

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

Evaluation

Metrics

Semantic Similarity

• Dataset: sts-test
• Evaluated with EmbeddingSimilarityEvaluator

Metric	Value
pearson_cosine	0.9038
spearman_cosine	0.9229

Binary Classification

• Datasets: allNLI-dev and Qnli-dev
• Evaluated with BinaryClassificationEvaluator

Metric	allNLI-dev	Qnli-dev
cosine_accuracy	0.7734	0.6992
cosine_accuracy_threshold	0.7483	0.664
cosine_f1	0.6769	0.6957
cosine_f1_threshold	0.6949	0.664
cosine_precision	0.5789	0.6769
cosine_recall	0.8148	0.7154
cosine_ap	0.6542	0.752
cosine_mcc	0.5058	0.3994

Training Details

Training Datasets

NLI

NLI

• Dataset: NLI at d43e6fe
• Size: 20,100 training samples
• Columns: anchor, entailment, and negative
• Approximate statistics based on the first 1000 samples:

  	anchor	entailment	negative
  type	string	string	string
  details	min: 4 tokens mean: 25.86 tokens max: 170 tokens	min: 6 tokens mean: 16.58 tokens max: 54 tokens	min: 6 tokens mean: 16.65 tokens max: 48 tokens

• Samples:

  anchor	entailment	negative
  Cugny is a commune . It is found in the region Picardie in the Aisne department in the north of France .	Cugny is a commune in the Aisne department in Picardy in northern France .	Cugny is a commune in the Var department in Provence-Alpes-Côte d`Azur in southern France.
  Two dogs on leashes look toward the door.	The dogs are ready to go out for a walk.	The dogs are not ready to go out for a walk.
  Denver Broncos Kyle Orton and Knowshon Moreno will both be listed as questionable for Sunday's game against Cincinnati.	Kyle Orton, Knowshon Moreno are questionable for Sunday	Kyle Borton, Knowshon Merino are available for Monday

• Loss: GISTEmbedLoss with these parameters:

  {'guide': SentenceTransformer(
    (0): Transformer({'max_seq_length': 512, 'do_lower_case': False}) 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()
  ), 'temperature': 0.025}
  

natural-questions

natural-questions

• Dataset: natural-questions at f9e894e
• Size: 20,100 training samples
• Columns: sentence1 and sentence2
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2
  type	string	string
  details	min: 10 tokens mean: 13.39 tokens max: 29 tokens	min: 23 tokens mean: 151.47 tokens max: 652 tokens

• Samples:

  sentence1	sentence2
  who was the first person to use fossils for dating rock layers	Relative dating The regular order of occurrence of fossils in rock layers was discovered around 1800 by William Smith. While digging the Somerset Coal Canal in southwest England, he found that fossils were always in the same order in the rock layers. As he continued his job as a surveyor, he found the same patterns across England. He also found that certain animals were in only certain layers and that they were in the same layers all across England. Due to that discovery, Smith was able to recognize the order that the rocks were formed. Sixteen years after his discovery, he published a geological map of England showing the rocks of different geologic time eras.
  how many games are in the eastern conference finals	NBA Conference Finals Initially, the BAA teams were aligned into two divisions, the Eastern Division and the Western Division. The Divisional Finals were first played in 1949, the league's third season. The first two seasons used a playoffs format where Eastern and Western Division teams would face each other before the BAA Finals, hence there were no divisional finals. In the 1949–50 season, the league realigned itself to three divisions, with the addition of the Central Division. However, the arrangement was only used for one season and the league went back into two divisions format in 1951. The two divisions format remained until 1970, when the NBA realigned itself into two conferences with two divisions each, which led to the renaming to Conference Finals. The finals was a best-of-3 series from 1949 to 1950 to; a best-of-5 series from 1951–56, and a best-of-7 series since 1957. Currently, the Conference Finals are played in a best-of-7 series like the NBA Playoffs and Finals. The t...
  what is the difference between sheraton and four points	Four Points by Sheraton In April 1995, Sheraton Hotels and Resorts introduced a new, up-scale hotel brand Four Points by Sheraton Hotels, to replace the designation of certain hotels as Sheraton Inns. In 1998, Starwood Hotels & Resorts Worldwide, Inc. acquired ITT Sheraton, outbidding Hilton. In 2000, Starwood re-launched Four Points by Sheraton, now targeted as a premier upscale five star hotel chain for business and leisure travelers.

• Loss: GISTEmbedLoss with these parameters:

  {'guide': SentenceTransformer(
    (0): Transformer({'max_seq_length': 512, 'do_lower_case': False}) 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()
  ), 'temperature': 0.025}
  

vitaminc

vitaminc

• Dataset: vitaminc at be6febb
• Size: 370,653 training samples
• Columns: claim and evidence
• Approximate statistics based on the first 1000 samples:

  	claim	evidence
  type	string	string
  details	min: 9 tokens mean: 21.17 tokens max: 80 tokens	min: 9 tokens mean: 46.62 tokens max: 293 tokens

• Samples:

  claim	evidence
  In 2013 Copa Sudamericana finals , Lanus won against Ponte Preta with Agustin Marchesin starting in both matches .	In 2013 , Marches�n won the Copa Sudamericana with Lan�s , starting in both matches of the finals against Ponte Preta of Brazil.
  More Life ( album ) debuted number one on the US Billboard 200 .	More Life received generally positive reviews and debuted at number one on the US Billboard 200 , becoming his seventh consecutive number one album , while also breaking several streaming records.
  Brandon Paak Anderson was born in Ventura County , California .	Brandon Paak Anderson ( born February 8 , 1986 ) , known professionally as Anderson .Paak , is an American singer , songwriter , rapper , drummer , and record producer from Ventura County , California .

• Loss: GISTEmbedLoss with these parameters:

  {'guide': SentenceTransformer(
    (0): Transformer({'max_seq_length': 512, 'do_lower_case': False}) 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()
  ), 'temperature': 0.025}
  

xsum

xsum

• Dataset: xsum at 044020f
• Size: 131,779 training samples
• Columns: summary and document
• Approximate statistics based on the first 1000 samples:

  	summary	document
  type	string	string
  details	min: 8 tokens mean: 30.94 tokens max: 51 tokens	min: 60 tokens mean: 336.79 tokens max: 616 tokens

• Samples:

  summary	document
  Salford Red Devils hooker Logan Tomkins and winger Greg Johnson have both signed new undisclosed-length contracts with the Super League club.	Tomkins, 24, joined Salford on a permanent deal from Wigan in 2016 and has become their first-choice hooker. Johnson, 27, has been at the AJ Bell Stadium since 2014 and scored 10 tries this season in 18 appearances. "Logan is one of the hardest workers around, he does so much unseen work," head coach Ian Watson said. "Greg is one of the best wingers in Super League when it comes to getting the team on the front foot. He's an extremely good trainer and that shows on the field," Watson added.
  Valtteri Bottas expects to be allowed to race on "equal terms" with Lewis Hamilton after being confirmed as Nico Rosberg's replacement at Mercedes.	Media playback is not supported on this device Bottas, 27, has signed a one-year deal with the option for more after arriving from Williams. "We are in this together, with similar cars and equipment to win races. I'm sure they will let us race hard and fair," the Finn told BBC Radio 5 live. "Lewis is a great driver and everyone knows how quick he is." Bottas, who made his grand prix debut with Williams in 2013, has not won a race in F1, while Briton Hamilton is a 53-time race winner and three-time world champion. I certainly think Valtteri can give Lewis a run for his money "It's going to be a big challenge for me to be quicker than him," added Bottas. "He knows this team really well, has been part of this team a long time, so for me it's going to be a great reference and we can be a strong pair." Rosberg's shock retirement, announced just five days after he wrapped up his maiden world title, presented Bottas with an unexpected chance to land the most coveted seat in the sport, and tea...
  The family of an Indian man who was lynched by a mob over rumours he consumed beef has denied reports that they no longer want a police investigation into the killing.	Mohammad Akhlaq was beaten to death by a mob in Dadri in Uttar Pradesh state in late September. His son, Mohammed Sartaj, told BBC Hindi's Salman Ravi that he "is waiting for the police to charge the suspects". Six people have been arrested in connection with the attack. "We will go to the president of India if we have to. We will also demand an inquiry by the Central Bureau of Investigation [India's top investigative body] if we feel that the Uttar Pradesh police is trying to save those involved in the case," Mr Sartaj, who works for the Indian Air Force, told BBC Hindi's Salman Ravi. He was responding to reports which said Mr Akhlaq's family had said they were satisfied with the compensation they had received and did not want further investigations. Mr Sartaj said that the police had already spoken to his family. "They have also taken statements from eye witnesses. My family has told the police about the people who were involved in the attack. Now we have to see if the police tries ...

• Loss: GISTEmbedLoss with these parameters:

  {'guide': SentenceTransformer(
    (0): Transformer({'max_seq_length': 512, 'do_lower_case': False}) 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()
  ), 'temperature': 0.025}
  

paws

paws

• Dataset: paws at 161ece9
• Size: 49,401 training samples
• Columns: sentence1 and sentence2
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2
  type	string	string
  details	min: 9 tokens mean: 31.51 tokens max: 58 tokens	min: 10 tokens mean: 31.48 tokens max: 57 tokens

• Samples:

  sentence1	sentence2
  In 2014 , Harris Publications acquired XXL '' , King '' and `` Antenna '' from Townsquare Media .	In 2014 Harris Publications '' XXL , '' King and '' Antenna `` acquired by Townsquare Media .
  Kallir is incompetent , but in politics very honest .	Kallir is incompetent , but very honest in politics .
  This association was further enhanced after the female Christian missionary , Nino , converted Mirian , his wife Nana and household into Christianity in or around 337 .	This association was further strengthened after the Christian female missionary Nino , Mirian , his wife Nana and household converted to Christianity in or around 337 .

• Loss: GISTEmbedLoss with these parameters:

  {'guide': SentenceTransformer(
    (0): Transformer({'max_seq_length': 512, 'do_lower_case': False}) 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()
  ), 'temperature': 0.025}
  

global_dataset

global_dataset

• Dataset: global_dataset
• Size: 46,804 training samples
• Columns: sentence1 and sentence2
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2
  type	string	string
  details	min: 7 tokens mean: 24.52 tokens max: 129 tokens	min: 5 tokens mean: 119.66 tokens max: 575 tokens

• Samples:

  sentence1	sentence2
  The Organizing Committee banned the unauthorized filming , and on 6 August 2008 , investigated SBS cameras inside the stadium during the ceremony as reprisals for the leak .	The organizing committee banned the unauthorized filming and investigated SBS cameras in the stadium on August 6 , 2008 during the ceremony as retaliation for the leak .
  Aaron O'Connor, Lenny Pidgeley, Lee Minshull and captain Andy Sandell are among 11 senior players to be released by Newport County.	Striker O'Connor tweeted: "Wanna thank everyone at Newport County for an amazing few years. Goalkeeper Pidgeley said "won't be getting another contract at Newport", while midfielder Minshull said "it's official, I'm a free agent". Midfielder Adam Chapman has turned down a new deal to stay at Rodney Parade. Shaun Jeffers, Max Porter, Jamie Stephens, Robbie Willmott, Ismail Yakubu and Mike Flynn are the other senior players to be released. Youngsters Joe Parker and Kyle Patten have also left County. Just seven players at the club are under contract for next season: Joe Day, Mark Byrne, Kevin Feely, Yan Klukowski, Aaron Collins, Tom Owen-Evans and Kieran Parselle. New manager Terry Butcher has offered contract extensions to Darren Jones, Ryan Jackson and Andrew Hughes. Pidgeley finished the season on loan at Mansfield, but O'Connor, Minshull and Sandell have been three of the Exiles' key players this term. O'Connor scored 11 goals in 42 appearances, Minshull also made 42 appearances for t...
  A former senior police officer has warned that Brexit could lead to "civil unrest" in Northern Ireland.	Peter Sheridan, now chief executive at Cooperation Ireland said the peace process is in a "fragile state". He said the numerous agreements that make up Northern Ireland's peace process show the brittle nature of its politics. "Less than a year ago, both governments had to step in to rescue the executive," said Mr Sheridan. Although stressing that he did not predict the violence "in the way it was [during the Troubles]", he added: "We've already seen peaceful protests along the border, even though nothing has happened on the border yet. "The history of this place is that mass protest can lead to confrontation and unrest. "We've had three shootings in the last few days, so we shouldn't be surprised that there would be people who would seek to exploit a change in the border." The former assistant chief constable denied that he was scaremongering: "I would challenge anyone who says there won't be people who will seek to exploit the border," he told the BBC's Good Morning Ulster programme. ...

• Loss: GISTEmbedLoss with these parameters:

  {'guide': SentenceTransformer(
    (0): Transformer({'max_seq_length': 512, 'do_lower_case': False}) 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()
  ), 'temperature': 0.025}
  

Evaluation Datasets

NLI

NLI

• Dataset: NLI at d43e6fe
• Size: 85 evaluation samples
• Columns: anchor, entailment, and negative
• Approximate statistics based on the first 85 samples:

  	anchor	entailment	negative
  type	string	string	string
  details	min: 10 tokens mean: 18.18 tokens max: 45 tokens	min: 6 tokens mean: 13.54 tokens max: 46 tokens	min: 7 tokens mean: 13.82 tokens max: 41 tokens

• Samples:

  anchor	entailment	negative
  A crowd is watching a group of men in suits with briefcases walk in formation down the street led by a woman holding a sign.	A group of well dressed people walk down a block with one of them holding a sign.	A group of poorly dressed people walk down a block with one of them holding a sign.
  two women loading their bicycles onto a bus rack	Two woman loading bikes onto a public bus.	Two women unloading bikes from a private bus.
  A person wearing a straw hat, standing outside working a steel apparatus with a pile of coconuts on the ground.	A person is near a pile of coconuts.	A person is far from a pile of coconuts.

• Loss: GISTEmbedLoss with these parameters:

  {'guide': SentenceTransformer(
    (0): Transformer({'max_seq_length': 512, 'do_lower_case': False}) 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()
  ), 'temperature': 0.025}
  

natural-questions

natural-questions

• Dataset: natural-questions at f9e894e
• Size: 113 evaluation samples
• Columns: sentence1 and sentence2
• Approximate statistics based on the first 113 samples:

  	sentence1	sentence2
  type	string	string
  details	min: 10 tokens mean: 13.02 tokens max: 20 tokens	min: 38 tokens mean: 149.73 tokens max: 590 tokens

• Samples:

  sentence1	sentence2
  who is the chief minister of all india	List of current Indian chief ministers In the Republic of India, a chief minister is the head of government of each of twenty-nine states and two union territories (Delhi and Puducherry). According to the Constitution of India, at the state-level, the governor is de jure head, but de facto executive authority rests with the chief minister. Following elections to the state legislative assembly, the governor usually invites the party (or coalition) with a majority of seats to form the government. The governor appoints the chief minister, whose council of ministers are collectively responsible to the assembly. Given he has the assembly's confidence, the chief minister's term is usually for a maximum of five years; there are no limits to the number of terms he/she can serve.[1]
  when does elena come back from being katherine	No Exit (The Vampire Diaries) Damon wakes up chained on the Salvatore house's basement. Stefan is there and Damon tries to warn him that because of his cravings, one day he will kill him but Stefan tells him that they will find a solution. Damon points to Stefan that "Elena" provoked him to feed on her and then she kicked a stake towards him so he will kill him. Stefan does not believe that Elena would want that, he locks Damon up and gets upstairs where Caroline is. The two of them discuss what happened between Stefan and "Elena" and Caroline also tells him about Nadia and Matt and the text he sent to her. From Matt's text, who used "K" and not "E", they put the pieces together and they realize that Katherine is in Elena's body.
  is the eureka tower the tallest building in australia	Eureka Tower Eureka Tower is a 297.3-metre (975 ft) skyscraper located in the Southbank precinct of Melbourne, Victoria, Australia.[3] Construction began in August 2002 and the exterior completed on 1 June 2006. The plaza was finished in June 2006 and the building was officially opened on 11 October 2006.[4] The project was designed by Melbourne architectural firm Fender Katsalidis Architects and was built by Grocon (Grollo Australia). The developer of the tower was Eureka Tower Pty Ltd, a joint venture consisting of Daniel Grollo (Grocon), investor Tab Fried and one of the Tower's architects Nonda Katsalidis. It was the world's tallest residential tower when measured to its highest floor,[5] until surpassed by Ocean Heights and the HHHR Tower in Dubai. It is the second tallest building in Australia, behind Q1, Queensland, and is the tallest to roof (excluding spire).[6] As of 2016 it is the 15th tallest residential building in the world.

• Loss: GISTEmbedLoss with these parameters:

  {'guide': SentenceTransformer(
    (0): Transformer({'max_seq_length': 512, 'do_lower_case': False}) 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()
  ), 'temperature': 0.025}
  

vitaminc

vitaminc

• Dataset: vitaminc at be6febb
• Size: 63,054 evaluation samples
• Columns: claim and evidence
• Approximate statistics based on the first 1000 samples:

  	claim	evidence
  type	string	string
  details	min: 10 tokens mean: 22.27 tokens max: 44 tokens	min: 11 tokens mean: 37.93 tokens max: 81 tokens

• Samples:

  claim	evidence
  By March 23 , 99,000 patients had recovered from coronavirus .	As of 23 March , more than 341,000 cases of COVID-19 have been reported in over 190 countries and territories , resulting in more than 14,700 deaths and over 99,000 recoveries .
  More than 681,500 cases of COVID-19 have been reported along with less than 32,000 deaths .	more than 683,000 cases of COVID-19 have been reported in over 190 countries and territories , resulting in approximately 32,100 deaths .
  On March 13th , over 145,000 cases of COVID-19 have been confirmed in more than 130 countries .	As of March 13th , over 145,000 cases of COVID-19 have been confirmed in more than 130 countries and territories , with major outbreaks in mainland China , Italy , South Korea , and Iran .

• Loss: GISTEmbedLoss with these parameters:

  {'guide': SentenceTransformer(
    (0): Transformer({'max_seq_length': 512, 'do_lower_case': False}) 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()
  ), 'temperature': 0.025}
  

xsum

xsum

• Dataset: xsum at 044020f
• Size: 131,779 evaluation samples
• Columns: summary and document
• Approximate statistics based on the first 1000 samples:

  	summary	document
  type	string	string
  details	min: 17 tokens mean: 30.55 tokens max: 47 tokens	min: 42 tokens mean: 306.88 tokens max: 554 tokens

• Samples:

  summary	document
  Jim Atchison, the chief executive of SeaWorld, has resigned.	The aquatic amusement park has struggled to attract visitors in the wake of a 2013 film, Blackfish, which criticised its treatment of killer whales. SeaWorld acknowledged in August that the film had hurt revenues at its San Diego, California park. The company's share price has fallen 44% this year, and now trades around $16 per share. In a statement, SeaWorld said that current board chairman David F D'Alessandro would serve as interim chief executive, and that the firm would also continue with its plans to reorganise in order to save $50m by the end of 2015. The company said that part of the restructuring would involve job cuts. SeaWorld operates 11 theme parks globally. In its most recent earnings report, it said attendance had dropped to 8.4 million visitors in the third-quarter of 2014 from 8.9 million in the same period a year earlier. It attributed the decline to "a combination of factors including negative media attention in California along with a challenging competitive environ...
  Organised criminal gangs are believed to be behind the thefts of more than 100 sheep and other animals.	North Wales Police have warned farmers after livestock was taken from farms in Gwynedd, Conwy county and Denbighshire in the last few weeks. Officers have said 80 sheep have gone missing in Bala, Llanrwst, Betws y Coed and Rhuallt in the last week They said the thefts appeared to be a professional operation. Rob Taylor, from the force's rural crime team, said the animals are likely being sold on or slaughtered illegally.
  Scottish FA performance director Malky Mackay will this week attempt to convince clubs to back proposed reductions to the number of funded academies in Scotland.	The suggestion is part of Project Brave - the outcome of an SFA working group tasked with improving and increasing the development of elite players. It recommends reducing the 29 funded academies to a maximum of 16. Mackay will consult member clubs in detail at four events this week. Under the plans, clubs will have their academies assessed by an independent company to assess which ones meet the new criteria. "We need to focus on the very best players in the very best academies with our limited resources we have," SFA chief executive Stewart Regan said. "One of the recommendations from the working group was to have no more than 16 academies in Scotland defined as elite. Any club can put a bid in, and they will be independently audited against a defined set of criteria. If they are successful, they will be included in Scotland's list of elite academies." Project Brave's recommendations: The working group for the strategy was formed in March last year, and initiated by Mackay's predecess...

• Loss: GISTEmbedLoss with these parameters:

  {'guide': SentenceTransformer(
    (0): Transformer({'max_seq_length': 512, 'do_lower_case': False}) 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()
  ), 'temperature': 0.025}
  

paws

paws

• Dataset: paws at 161ece9
• Size: 8,000 evaluation samples
• Columns: sentence1 and sentence2
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2
  type	string	string
  details	min: 15 tokens mean: 30.68 tokens max: 52 tokens	min: 15 tokens mean: 30.42 tokens max: 49 tokens

• Samples:

  sentence1	sentence2
  The music was written by Shyam and lyrics was composed by Sreekumaran Thampi and Sathyan Anthikkad .	The music was written by Shyam and the lyrics by Sreekumaran Thampi and Sathyan Anthikkad composed .
  Wright moved from Chapel Hill , NC to New York .	Wright moved to New York from Chapel Hill , NC .
  Massé was born in Holland , Michigan , grew up in Westchester County , New York , and lived during her youth in Europe .	Massé was born in Holland , Michigan , grew up in Westchester County , New York , and lived in Europe during her teens .

• Loss: GISTEmbedLoss with these parameters:

  {'guide': SentenceTransformer(
    (0): Transformer({'max_seq_length': 512, 'do_lower_case': False}) 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()
  ), 'temperature': 0.025}
  

global_dataset

global_dataset

• Dataset: global_dataset
• Size: 256 evaluation samples
• Columns: sentence1 and sentence2
• Approximate statistics based on the first 256 samples:

  	sentence1	sentence2
  type	string	string
  details	min: 8 tokens mean: 24.09 tokens max: 56 tokens	min: 6 tokens mean: 114.19 tokens max: 572 tokens

• Samples:

  sentence1	sentence2
  Cardiff City will want at least £5m for striker Joe Mason, BBC Wales Sport has learned.	The 24-year-old is understood to have attracted interest from Championship rivals Wolverhampton Wanderers. But a senior Cardiff source has indicated the Bluebirds will want "at least £5m or £6m" for the former Republic of Ireland Under-21 forward. Mason joined Cardiff from Plymouth Argyle in 2011 for £250,000 and has scored 22 goals in 60 League starts. This season Mason has chipped in with six goals, enjoying a prolonged run in the team having had three loan spells at Bolton Wanderers during previous campaigns. Cardiff are under a transfer embargo, but chief executive Ken Choo made it clear the club would not be forced to sell their best players as a result of the restriction. The club are also looking to draft in at least three players on loan. Mason and midfielder Joe Ralls, 22, are seen by the club as prized young assets.
  A crowded city street with lots of pedestrians.	There are many pedestrians on the city street.
  Until 1951 , he was active as a socialist in post-war legislation when he decided to focus on local politics .	He was active as a socialist in the post-war legislature until 1951 , when he decided to focus on local politics .

• Loss: GISTEmbedLoss with these parameters:

  {'guide': SentenceTransformer(
    (0): Transformer({'max_seq_length': 512, 'do_lower_case': False}) 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()
  ), 'temperature': 0.025}
  

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• per_device_train_batch_size: 128
• per_device_eval_batch_size: 256
• gradient_accumulation_steps: 3
• learning_rate: 0.001
• weight_decay: 0.001
• lr_scheduler_type: cosine_with_min_lr
• lr_scheduler_kwargs: {'num_cycles': 0.5, 'min_lr': 0.0003333333333333333}
• warmup_ratio: 0.2
• save_safetensors: False
• fp16: True
• remove_unused_columns: False
• push_to_hub: True
• hub_model_id: bobox/XLMRoBERTaM3-CustomPoolin-v1-s1-checkpoints-tmp
• hub_strategy: all_checkpoints
• hub_private_repo: False
• 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: 128
• per_device_eval_batch_size: 256
• per_gpu_train_batch_size: None
• per_gpu_eval_batch_size: None
• gradient_accumulation_steps: 3
• eval_accumulation_steps: None
• torch_empty_cache_steps: None
• learning_rate: 0.001
• weight_decay: 0.001
• adam_beta1: 0.9
• adam_beta2: 0.999
• adam_epsilon: 1e-08
• max_grad_norm: 1.0
• num_train_epochs: 3
• max_steps: -1
• lr_scheduler_type: cosine_with_min_lr
• lr_scheduler_kwargs: {'num_cycles': 0.5, 'min_lr': 0.0003333333333333333}
• warmup_ratio: 0.2
• warmup_steps: 0
• log_level: passive
• log_level_replica: warning
• log_on_each_node: True
• logging_nan_inf_filter: True
• save_safetensors: False
• 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: True
• 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: False
• 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: True
• resume_from_checkpoint: None
• hub_model_id: bobox/XLMRoBERTaM3-CustomPoolin-v1-s1-checkpoints-tmp
• hub_strategy: all_checkpoints
• hub_private_repo: False
• 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
• 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

Click to expand

Epoch	Step	Training Loss	NLI loss	natural-questions loss	vitaminc loss	xsum loss	paws loss	global dataset loss	sts-test_spearman_cosine	allNLI-dev_cosine_ap	Qnli-dev_cosine_ap
0.0027	1	7.8455	-	-	-	-	-	-	-	-	-
0.0054	2	10.2716	-	-	-	-	-	-	-	-	-
0.0081	3	8.8104	-	-	-	-	-	-	-	-	-
0.0108	4	9.723	-	-	-	-	-	-	-	-	-
0.0135	5	8.2787	-	-	-	-	-	-	-	-	-
0.0162	6	2.1849	-	-	-	-	-	-	-	-	-
0.0189	7	6.9562	-	-	-	-	-	-	-	-	-
0.0216	8	8.0012	-	-	-	-	-	-	-	-	-
0.0243	9	7.2178	-	-	-	-	-	-	-	-	-
0.0270	10	3.7548	-	-	-	-	-	-	-	-	-
0.0296	11	1.9087	-	-	-	-	-	-	-	-	-
0.0323	12	2.4871	-	-	-	-	-	-	-	-	-
0.0350	13	2.0275	-	-	-	-	-	-	-	-	-
0.0377	14	1.7456	-	-	-	-	-	-	-	-	-
0.0404	15	1.8779	-	-	-	-	-	-	-	-	-
0.0431	16	1.3286	-	-	-	-	-	-	-	-	-
0.0458	17	1.9446	-	-	-	-	-	-	-	-	-
0.0485	18	1.7784	-	-	-	-	-	-	-	-	-
0.0512	19	2.2488	-	-	-	-	-	-	-	-	-
0.0539	20	1.4934	-	-	-	-	-	-	-	-	-
0.0566	21	1.6026	-	-	-	-	-	-	-	-	-
0.0593	22	1.1284	-	-	-	-	-	-	-	-	-
0.0620	23	0.8786	-	-	-	-	-	-	-	-	-
0.0647	24	0.9379	-	-	-	-	-	-	-	-	-
0.0674	25	1.4386	-	-	-	-	-	-	-	-	-
0.0701	26	1.4041	-	-	-	-	-	-	-	-	-
0.0728	27	0.5954	-	-	-	-	-	-	-	-	-
0.0755	28	1.0351	-	-	-	-	-	-	-	-	-
0.0782	29	1.1524	-	-	-	-	-	-	-	-	-
0.0809	30	0.4302	-	-	-	-	-	-	-	-	-
0.0836	31	1.3629	-	-	-	-	-	-	-	-	-
0.0863	32	0.801	-	-	-	-	-	-	-	-	-
0.0889	33	0.7135	-	-	-	-	-	-	-	-	-
0.0916	34	1.0678	-	-	-	-	-	-	-	-	-
0.0943	35	0.8164	-	-	-	-	-	-	-	-	-
0.0970	36	0.7157	-	-	-	-	-	-	-	-	-
0.0997	37	0.4396	-	-	-	-	-	-	-	-	-
0.1024	38	0.9716	-	-	-	-	-	-	-	-	-
0.1051	39	0.9321	-	-	-	-	-	-	-	-	-
0.1078	40	0.5171	-	-	-	-	-	-	-	-	-
0.1105	41	1.9291	-	-	-	-	-	-	-	-	-
0.1132	42	1.0919	-	-	-	-	-	-	-	-	-
0.1159	43	1.4841	-	-	-	-	-	-	-	-	-
0.1186	44	1.1497	-	-	-	-	-	-	-	-	-
0.1213	45	0.7006	-	-	-	-	-	-	-	-	-
0.1240	46	1.7046	-	-	-	-	-	-	-	-	-
0.1267	47	0.8556	-	-	-	-	-	-	-	-	-
0.1294	48	1.6158	-	-	-	-	-	-	-	-	-
0.1321	49	1.3545	-	-	-	-	-	-	-	-	-
0.1348	50	1.0598	-	-	-	-	-	-	-	-	-
0.1375	51	1.3435	-	-	-	-	-	-	-	-	-
0.1402	52	0.5441	-	-	-	-	-	-	-	-	-
0.1429	53	0.8275	-	-	-	-	-	-	-	-	-
0.1456	54	1.0796	-	-	-	-	-	-	-	-	-
0.1482	55	0.9102	-	-	-	-	-	-	-	-	-
0.1509	56	0.6044	0.7544	0.1253	1.7376	0.3028	0.0228	0.3648	0.9132	0.6860	0.7579
0.1536	57	0.6791	-	-	-	-	-	-	-	-	-
0.1563	58	2.0332	-	-	-	-	-	-	-	-	-
0.1590	59	1.6908	-	-	-	-	-	-	-	-	-
0.1617	60	1.9835	-	-	-	-	-	-	-	-	-
0.1644	61	0.5596	-	-	-	-	-	-	-	-	-
0.1671	62	0.5455	-	-	-	-	-	-	-	-	-
0.1698	63	1.4403	-	-	-	-	-	-	-	-	-
0.1725	64	1.3612	-	-	-	-	-	-	-	-	-
0.1752	65	0.4134	-	-	-	-	-	-	-	-	-
0.1779	66	1.1737	-	-	-	-	-	-	-	-	-
0.1806	67	0.9298	-	-	-	-	-	-	-	-	-
0.1833	68	1.1334	-	-	-	-	-	-	-	-	-
0.1860	69	1.0759	-	-	-	-	-	-	-	-	-
0.1887	70	0.961	-	-	-	-	-	-	-	-	-
0.1914	71	0.329	-	-	-	-	-	-	-	-	-
0.1941	72	1.4607	-	-	-	-	-	-	-	-	-
0.1968	73	1.13	-	-	-	-	-	-	-	-	-
0.1995	74	0.4843	-	-	-	-	-	-	-	-	-
0.2022	75	1.1167	-	-	-	-	-	-	-	-	-
0.2049	76	0.7438	-	-	-	-	-	-	-	-	-
0.2075	77	1.0913	-	-	-	-	-	-	-	-	-
0.2102	78	1.0479	-	-	-	-	-	-	-	-	-
0.2129	79	1.1103	-	-	-	-	-	-	-	-	-
0.2156	80	0.9204	-	-	-	-	-	-	-	-	-
0.2183	81	0.9669	-	-	-	-	-	-	-	-	-
0.2210	82	1.6861	-	-	-	-	-	-	-	-	-
0.2237	83	1.076	-	-	-	-	-	-	-	-	-
0.2264	84	1.2668	-	-	-	-	-	-	-	-	-
0.2291	85	1.4458	-	-	-	-	-	-	-	-	-
0.2318	86	1.0282	-	-	-	-	-	-	-	-	-
0.2345	87	0.9422	-	-	-	-	-	-	-	-	-
0.2372	88	0.8485	-	-	-	-	-	-	-	-	-
0.2399	89	0.6634	-	-	-	-	-	-	-	-	-
0.2426	90	0.5559	-	-	-	-	-	-	-	-	-
0.2453	91	0.4284	-	-	-	-	-	-	-	-	-
0.2480	92	0.8843	-	-	-	-	-	-	-	-	-
0.2507	93	1.1379	-	-	-	-	-	-	-	-	-
0.2534	94	0.8325	-	-	-	-	-	-	-	-	-
0.2561	95	0.7055	-	-	-	-	-	-	-	-	-
0.2588	96	0.5176	-	-	-	-	-	-	-	-	-
0.2615	97	0.6018	-	-	-	-	-	-	-	-	-
0.2642	98	1.2091	-	-	-	-	-	-	-	-	-
0.2668	99	1.1578	-	-	-	-	-	-	-	-	-
0.2695	100	1.3774	-	-	-	-	-	-	-	-	-
0.2722	101	1.6864	-	-	-	-	-	-	-	-	-
0.2749	102	1.5131	-	-	-	-	-	-	-	-	-
0.2776	103	1.5626	-	-	-	-	-	-	-	-	-
0.2803	104	1.5972	-	-	-	-	-	-	-	-	-
0.2830	105	1.1608	-	-	-	-	-	-	-	-	-
0.2857	106	1.7296	-	-	-	-	-	-	-	-	-
0.2884	107	0.7756	-	-	-	-	-	-	-	-	-
0.2911	108	1.4043	-	-	-	-	-	-	-	-	-
0.2938	109	1.0434	-	-	-	-	-	-	-	-	-
0.2965	110	0.7851	-	-	-	-	-	-	-	-	-
0.2992	111	0.8959	-	-	-	-	-	-	-	-	-
0.3019	112	1.1451	0.7854	0.1174	2.1760	0.1466	0.0221	0.5730	0.9229	0.6542	0.7520

Framework Versions

• Python: 3.11.11
• Sentence Transformers: 3.4.1
• Transformers: 4.51.1
• PyTorch: 2.5.1+cu124
• Accelerate: 1.3.0
• Datasets: 3.5.0
• Tokenizers: 0.21.0

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}
}