checkpoint-2173/README.md

metadata

base_model: bobox/DeBERTa-ST-AllLayers-v3-checkpoints-tmp
datasets:
  - sentence-transformers/all-nli
  - tals/vitaminc
  - nyu-mll/glue
  - allenai/scitail
  - sentence-transformers/xsum
  - sentence-transformers/sentence-compression
  - allenai/sciq
  - allenai/qasc
  - sentence-transformers/msmarco-msmarco-distilbert-base-v3
  - sentence-transformers/natural-questions
  - sentence-transformers/trivia-qa
  - sentence-transformers/quora-duplicates
  - sentence-transformers/gooaq
language:
  - en
library_name: sentence-transformers
metrics:
  - pearson_cosine
  - spearman_cosine
  - pearson_manhattan
  - spearman_manhattan
  - pearson_euclidean
  - spearman_euclidean
  - pearson_dot
  - spearman_dot
  - pearson_max
  - spearman_max
  - cosine_accuracy
  - cosine_accuracy_threshold
  - cosine_f1
  - cosine_f1_threshold
  - cosine_precision
  - cosine_recall
  - cosine_ap
  - dot_accuracy
  - dot_accuracy_threshold
  - dot_f1
  - dot_f1_threshold
  - dot_precision
  - dot_recall
  - dot_ap
  - manhattan_accuracy
  - manhattan_accuracy_threshold
  - manhattan_f1
  - manhattan_f1_threshold
  - manhattan_precision
  - manhattan_recall
  - manhattan_ap
  - euclidean_accuracy
  - euclidean_accuracy_threshold
  - euclidean_f1
  - euclidean_f1_threshold
  - euclidean_precision
  - euclidean_recall
  - euclidean_ap
  - max_accuracy
  - max_accuracy_threshold
  - max_f1
  - max_f1_threshold
  - max_precision
  - max_recall
  - max_ap
pipeline_tag: sentence-similarity
tags:
  - sentence-transformers
  - sentence-similarity
  - feature-extraction
  - generated_from_trainer
  - dataset_size:138757
  - loss:AdaptiveLayerLoss
  - loss:GISTEmbedLoss
  - loss:OnlineContrastiveLoss
  - loss:MultipleNegativesSymmetricRankingLoss
  - loss:MultipleNegativesRankingLoss
widget:
  - source_sentence: >-
      Because of the accounting charge , the company now says it lost $ 1.04
      billion , or 32 cents a share , in the quarter ended June 30 .
    sentences:
      - >-
        " As long as we don 't march on Tehran , I think we are going to be in
        pretty good shape , " he said .
      - >-
        HPs shipments increased 48 per cent year-on-year , compared to an
        increase of 31 per cent for Dell .
      - >-
        Including the charge , the Santa Clara , Calif.-based company said
        Monday it lost $ 1.04 billion , or 32 cents per share , in the period
        ending June 30 .
  - source_sentence: >-
      In more than 130 countries , more than 143,000 COVID-19 cases have been
      confirmed .
    sentences:
      - >-
        over 144,000 cases have been confirmed in more than 130 countries and
        territories , with major outbreaks in mainland China , Italy , South
        Korea , and Iran .
      - >-
        more than 650,000 cases have been reported worldwide ; more than 30,200
        people have died and more than 139,000 have recovered , with the US
        having overtaken China and Italy to have the highest number of confirmed
        cases in the world .
      - >-
        As of 26 March , more than 519,000 cases of COVID-19 have been reported
        in over 200 countries and territories , resulting in approximately
        23,500 deaths and more than 123,000 recoveries .
  - source_sentence: Young roots and leaves are the major sites of gibberellin production.
    sentences:
      - What are the major sites of gibberellin production?
      - >-
        In science, what do you call something that always applies under the
        same conditions?
      - How does carbon dioxide chemically weather rocks?
  - source_sentence: Lasers are used to produce a beam that is coherent.
    sentences:
      - Roundworms reproduce sexually.. Roundworm eggs are comparatively large..
      - >-
        a laser is used for producing light. Lasers, for instance, produce
        coherent light..
      - >-
        Some touch receptors sense differences in temperature or pain..
        Temperature Temperature is an measurement of the amount of heat..
  - source_sentence: >-
      PTI / New York January 19, 2013, 13:05 Johnny Depp's girlfriend Amber
      Heard has reportedly moved on from her fling with the actor and is now
      dating female French model, Marie de Villepin.
    sentences:
      - BAA first half loss quadruples
      - Amber Heard dating female French model?
      - Five people shot in San Francisco
model-index:
  - name: SentenceTransformer based on bobox/DeBERTa-ST-AllLayers-v3-checkpoints-tmp
    results:
      - task:
          type: semantic-similarity
          name: Semantic Similarity
        dataset:
          name: StS test
          type: StS-test
        metrics:
          - type: pearson_cosine
            value: 0.8796347799632783
            name: Pearson Cosine
          - type: spearman_cosine
            value: 0.8923792178697394
            name: Spearman Cosine
          - type: pearson_manhattan
            value: 0.8673292515085165
            name: Pearson Manhattan
          - type: spearman_manhattan
            value: 0.8708405624162647
            name: Spearman Manhattan
          - type: pearson_euclidean
            value: 0.8662889333214727
            name: Pearson Euclidean
          - type: spearman_euclidean
            value: 0.867901651872447
            name: Spearman Euclidean
          - type: pearson_dot
            value: 0.7952693748718531
            name: Pearson Dot
          - type: spearman_dot
            value: 0.8035752306525754
            name: Spearman Dot
          - type: pearson_max
            value: 0.8796347799632783
            name: Pearson Max
          - type: spearman_max
            value: 0.8923792178697394
            name: Spearman Max
      - task:
          type: binary-classification
          name: Binary Classification
        dataset:
          name: mrpc test
          type: mrpc-test
        metrics:
          - type: cosine_accuracy
            value: 0.7578947368421053
            name: Cosine Accuracy
          - type: cosine_accuracy_threshold
            value: 0.6727536916732788
            name: Cosine Accuracy Threshold
          - type: cosine_f1
            value: 0.8397212543554007
            name: Cosine F1
          - type: cosine_f1_threshold
            value: 0.6727536916732788
            name: Cosine F1 Threshold
          - type: cosine_precision
            value: 0.7392638036809815
            name: Cosine Precision
          - type: cosine_recall
            value: 0.9717741935483871
            name: Cosine Recall
          - type: cosine_ap
            value: 0.8587001102740125
            name: Cosine Ap
          - type: dot_accuracy
            value: 0.7078947368421052
            name: Dot Accuracy
          - type: dot_accuracy_threshold
            value: 16.11664390563965
            name: Dot Accuracy Threshold
          - type: dot_f1
            value: 0.810810810810811
            name: Dot F1
          - type: dot_f1_threshold
            value: 13.418485641479492
            name: Dot F1 Threshold
          - type: dot_precision
            value: 0.6976744186046512
            name: Dot Precision
          - type: dot_recall
            value: 0.967741935483871
            name: Dot Recall
          - type: dot_ap
            value: 0.8100266677537138
            name: Dot Ap
          - type: manhattan_accuracy
            value: 0.7263157894736842
            name: Manhattan Accuracy
          - type: manhattan_accuracy_threshold
            value: 71.4499740600586
            name: Manhattan Accuracy Threshold
          - type: manhattan_f1
            value: 0.8114901256732495
            name: Manhattan F1
          - type: manhattan_f1_threshold
            value: 76.88587951660156
            name: Manhattan F1 Threshold
          - type: manhattan_precision
            value: 0.7313915857605178
            name: Manhattan Precision
          - type: manhattan_recall
            value: 0.9112903225806451
            name: Manhattan Recall
          - type: manhattan_ap
            value: 0.8234340325662497
            name: Manhattan Ap
          - type: euclidean_accuracy
            value: 0.7315789473684211
            name: Euclidean Accuracy
          - type: euclidean_accuracy_threshold
            value: 3.678903102874756
            name: Euclidean Accuracy Threshold
          - type: euclidean_f1
            value: 0.8111111111111112
            name: Euclidean F1
          - type: euclidean_f1_threshold
            value: 3.678903102874756
            name: Euclidean F1 Threshold
          - type: euclidean_precision
            value: 0.75
            name: Euclidean Precision
          - type: euclidean_recall
            value: 0.8830645161290323
            name: Euclidean Recall
          - type: euclidean_ap
            value: 0.8265125766382443
            name: Euclidean Ap
          - type: max_accuracy
            value: 0.7578947368421053
            name: Max Accuracy
          - type: max_accuracy_threshold
            value: 71.4499740600586
            name: Max Accuracy Threshold
          - type: max_f1
            value: 0.8397212543554007
            name: Max F1
          - type: max_f1_threshold
            value: 76.88587951660156
            name: Max F1 Threshold
          - type: max_precision
            value: 0.75
            name: Max Precision
          - type: max_recall
            value: 0.9717741935483871
            name: Max Recall
          - type: max_ap
            value: 0.8587001102740125
            name: Max Ap
      - task:
          type: binary-classification
          name: Binary Classification
        dataset:
          name: Vitaminc test
          type: Vitaminc-test
        metrics:
          - type: cosine_accuracy
            value: 0.5710526315789474
            name: Cosine Accuracy
          - type: cosine_accuracy_threshold
            value: 0.717450737953186
            name: Cosine Accuracy Threshold
          - type: cosine_f1
            value: 0.6779026217228464
            name: Cosine F1
          - type: cosine_f1_threshold
            value: 0.4914843738079071
            name: Cosine F1 Threshold
          - type: cosine_precision
            value: 0.5186246418338109
            name: Cosine Precision
          - type: cosine_recall
            value: 0.9783783783783784
            name: Cosine Recall
          - type: cosine_ap
            value: 0.5613681349794449
            name: Cosine Ap
          - type: dot_accuracy
            value: 0.5631578947368421
            name: Dot Accuracy
          - type: dot_accuracy_threshold
            value: 22.15652847290039
            name: Dot Accuracy Threshold
          - type: dot_f1
            value: 0.6703910614525139
            name: Dot F1
          - type: dot_f1_threshold
            value: 13.749557495117188
            name: Dot F1 Threshold
          - type: dot_precision
            value: 0.5113636363636364
            name: Dot Precision
          - type: dot_recall
            value: 0.972972972972973
            name: Dot Recall
          - type: dot_ap
            value: 0.5440593805944962
            name: Dot Ap
          - type: manhattan_accuracy
            value: 0.5815789473684211
            name: Manhattan Accuracy
          - type: manhattan_accuracy_threshold
            value: 82.38890075683594
            name: Manhattan Accuracy Threshold
          - type: manhattan_f1
            value: 0.6642728904847397
            name: Manhattan F1
          - type: manhattan_f1_threshold
            value: 147.61215209960938
            name: Manhattan F1 Threshold
          - type: manhattan_precision
            value: 0.49731182795698925
            name: Manhattan Precision
          - type: manhattan_recall
            value: 1
            name: Manhattan Recall
          - type: manhattan_ap
            value: 0.5516886870709616
            name: Manhattan Ap
          - type: euclidean_accuracy
            value: 0.5894736842105263
            name: Euclidean Accuracy
          - type: euclidean_accuracy_threshold
            value: 4.052872657775879
            name: Euclidean Accuracy Threshold
          - type: euclidean_f1
            value: 0.6654676258992805
            name: Euclidean F1
          - type: euclidean_f1_threshold
            value: 7.110843658447266
            name: Euclidean F1 Threshold
          - type: euclidean_precision
            value: 0.49865229110512127
            name: Euclidean Precision
          - type: euclidean_recall
            value: 1
            name: Euclidean Recall
          - type: euclidean_ap
            value: 0.5530280125387326
            name: Euclidean Ap
          - type: max_accuracy
            value: 0.5894736842105263
            name: Max Accuracy
          - type: max_accuracy_threshold
            value: 82.38890075683594
            name: Max Accuracy Threshold
          - type: max_f1
            value: 0.6779026217228464
            name: Max F1
          - type: max_f1_threshold
            value: 147.61215209960938
            name: Max F1 Threshold
          - type: max_precision
            value: 0.5186246418338109
            name: Max Precision
          - type: max_recall
            value: 1
            name: Max Recall
          - type: max_ap
            value: 0.5613681349794449
            name: Max Ap

SentenceTransformer based on bobox/DeBERTa-ST-AllLayers-v3-checkpoints-tmp

This is a sentence-transformers model finetuned from bobox/DeBERTa-ST-AllLayers-v3-checkpoints-tmp on the nli-pairs, vitaminc-pairs, qnli-contrastive, scitail-pairs-qa, scitail-pairs-pos, xsum-pairs, compression-pairs, compression-pairs2, compression-pairs3, sciq_pairs, qasc_pairs, qasc_facts_sym, openbookqa_pairs, msmarco_pairs, msmarco_pairs2, nq_pairs, nq_pairs2, trivia_pairs, quora_pairs, gooaq_pairs, gooaq_pairs2 and mrpc_pairs datasets. It maps sentences & paragraphs to a 768-dimensional dense vector space and can be used for semantic textual similarity, semantic search, paraphrase mining, text classification, clustering, and more.

Model Details

Model Description

• Model Type: Sentence Transformer
• Base model: bobox/DeBERTa-ST-AllLayers-v3-checkpoints-tmp
• Maximum Sequence Length: 512 tokens
• Output Dimensionality: 768 tokens
• Similarity Function: Cosine Similarity
• Training Datasets:
  • nli-pairs
  • vitaminc-pairs
  • qnli-contrastive
  • scitail-pairs-qa
  • scitail-pairs-pos
  • xsum-pairs
  • compression-pairs
  • compression-pairs2
  • compression-pairs3
  • sciq_pairs
  • qasc_pairs
  • qasc_facts_sym
  • openbookqa_pairs
  • msmarco_pairs
  • msmarco_pairs2
  • nq_pairs
  • nq_pairs2
  • trivia_pairs
  • quora_pairs
  • gooaq_pairs
  • gooaq_pairs2
  • mrpc_pairs
• 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': 512, 'do_lower_case': False}) with Transformer model: DebertaV2Model 
  (1): Pooling({'word_embedding_dimension': 768, '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})
)

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/DeBERTa-ST-AllLayers-v3.1bis-checkpoints-tmp")
# Run inference
sentences = [
    "PTI / New York January 19, 2013, 13:05 Johnny Depp's girlfriend Amber Heard has reportedly moved on from her fling with the actor and is now dating female French model, Marie de Villepin.",
    'Amber Heard dating female French model?',
    'Five people shot in San Francisco',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 768]

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

Evaluation

Metrics

Semantic Similarity

• Dataset: StS-test
• Evaluated with EmbeddingSimilarityEvaluator

Metric	Value
pearson_cosine	0.8796
spearman_cosine	0.8924
pearson_manhattan	0.8673
spearman_manhattan	0.8708
pearson_euclidean	0.8663
spearman_euclidean	0.8679
pearson_dot	0.7953
spearman_dot	0.8036
pearson_max	0.8796
spearman_max	0.8924

Binary Classification

• Dataset: mrpc-test
• Evaluated with BinaryClassificationEvaluator

Metric	Value
cosine_accuracy	0.7579
cosine_accuracy_threshold	0.6728
cosine_f1	0.8397
cosine_f1_threshold	0.6728
cosine_precision	0.7393
cosine_recall	0.9718
cosine_ap	0.8587
dot_accuracy	0.7079
dot_accuracy_threshold	16.1166
dot_f1	0.8108
dot_f1_threshold	13.4185
dot_precision	0.6977
dot_recall	0.9677
dot_ap	0.81
manhattan_accuracy	0.7263
manhattan_accuracy_threshold	71.45
manhattan_f1	0.8115
manhattan_f1_threshold	76.8859
manhattan_precision	0.7314
manhattan_recall	0.9113
manhattan_ap	0.8234
euclidean_accuracy	0.7316
euclidean_accuracy_threshold	3.6789
euclidean_f1	0.8111
euclidean_f1_threshold	3.6789
euclidean_precision	0.75
euclidean_recall	0.8831
euclidean_ap	0.8265
max_accuracy	0.7579
max_accuracy_threshold	71.45
max_f1	0.8397
max_f1_threshold	76.8859
max_precision	0.75
max_recall	0.9718
max_ap	0.8587

Binary Classification

• Dataset: Vitaminc-test
• Evaluated with BinaryClassificationEvaluator

Metric	Value
cosine_accuracy	0.5711
cosine_accuracy_threshold	0.7175
cosine_f1	0.6779
cosine_f1_threshold	0.4915
cosine_precision	0.5186
cosine_recall	0.9784
cosine_ap	0.5614
dot_accuracy	0.5632
dot_accuracy_threshold	22.1565
dot_f1	0.6704
dot_f1_threshold	13.7496
dot_precision	0.5114
dot_recall	0.973
dot_ap	0.5441
manhattan_accuracy	0.5816
manhattan_accuracy_threshold	82.3889
manhattan_f1	0.6643
manhattan_f1_threshold	147.6122
manhattan_precision	0.4973
manhattan_recall	1.0
manhattan_ap	0.5517
euclidean_accuracy	0.5895
euclidean_accuracy_threshold	4.0529
euclidean_f1	0.6655
euclidean_f1_threshold	7.1108
euclidean_precision	0.4987
euclidean_recall	1.0
euclidean_ap	0.553
max_accuracy	0.5895
max_accuracy_threshold	82.3889
max_f1	0.6779
max_f1_threshold	147.6122
max_precision	0.5186
max_recall	1.0
max_ap	0.5614

Training Details

Training Datasets

nli-pairs

• Dataset: nli-pairs at d482672
• Size: 10,000 training samples
• Columns: sentence1 and sentence2
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2
  type	string	string
  details	min: 5 tokens mean: 16.62 tokens max: 62 tokens	min: 4 tokens mean: 9.46 tokens max: 29 tokens

• Samples:

  sentence1	sentence2
  A person on a horse jumps over a broken down airplane.	A person is outdoors, on a horse.
  Children smiling and waving at camera	There are children present
  A boy is jumping on skateboard in the middle of a red bridge.	The boy does a skateboarding trick.

• Loss: AdaptiveLayerLoss with these parameters:

  {
      "loss": "GISTEmbedLoss",
      "n_layers_per_step": -1,
      "last_layer_weight": 1.75,
      "prior_layers_weight": 0.5,
      "kl_div_weight": 1.25,
      "kl_temperature": 0.9
  }
  

vitaminc-pairs

• Dataset: vitaminc-pairs at be6febb
• Size: 10,000 training samples
• Columns: claim and evidence
• Approximate statistics based on the first 1000 samples:

  	claim	evidence
  type	string	string
  details	min: 7 tokens mean: 17.68 tokens max: 63 tokens	min: 8 tokens mean: 36.56 tokens max: 262 tokens

• Samples:

  claim	evidence
  Captive State averaged a rating of more than 5.44/10 on Rotten Tomatoes , with more than 36 reviews .	On review aggregator Rotten Tomatoes , the film holds an approval rating of 47 % based on 38 reviews , with an average rating of 5.45/10 .
  The film Apne received favorable reviews .	The film received positive reviews from critics , many complimenting the theme of a family drama with just the right dose of emotion , drama , comedy , and action in a not so violent way .
  Cristhian Stuani 's goal helped Middlesbrough win the game at Burton Albion .	Three days later he made his first start , in the opening round of the League Cup , scoring in each half of a 3–1 win over Oldham Athletic at Boundary Park ; He scored a brace again in the second round on the 25th , as Middlesbrough came from behind to win away at Burton Albion.

• Loss: AdaptiveLayerLoss with these parameters:

  {
      "loss": "GISTEmbedLoss",
      "n_layers_per_step": -1,
      "last_layer_weight": 1.75,
      "prior_layers_weight": 0.5,
      "kl_div_weight": 1.25,
      "kl_temperature": 0.9
  }
  

qnli-contrastive

• Dataset: qnli-contrastive at bcdcba7
• Size: 10,000 training samples
• Columns: sentence1, sentence2, and label
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2	label
  type	string	string	int
  details	min: 6 tokens mean: 13.66 tokens max: 32 tokens	min: 6 tokens mean: 34.99 tokens max: 149 tokens	0: 100.00%

• Samples:

  sentence1	sentence2	label
  What did Oberhauser call himself after he faked his death?	Bond is tortured as Oberhauser discusses their shared history: after the younger Bond was orphaned, Oberhauser's father, Hannes, became his temporary guardian.	0
  Of what are deviations in the epics thought to be the influence?	The deviations from Pāṇini in the epics are generally considered to be on account of interference from Prakrits, or innovations, and not because they are pre-Paninian.	0
  What was Whitehead's criticism of the use of inert ideas in education?	He opined that "education with inert ideas is not only useless: it is, above all things, harmful."	0

• Loss: AdaptiveLayerLoss with these parameters:

  {
      "loss": "OnlineContrastiveLoss",
      "n_layers_per_step": -1,
      "last_layer_weight": 0.75,
      "prior_layers_weight": 1.25,
      "kl_div_weight": 0.8,
      "kl_temperature": 0.75
  }
  

scitail-pairs-qa

• Dataset: scitail-pairs-qa at 0cc4353
• Size: 10,000 training samples
• Columns: sentence2 and sentence1
• Approximate statistics based on the first 1000 samples:

  	sentence2	sentence1
  type	string	string
  details	min: 7 tokens mean: 15.81 tokens max: 41 tokens	min: 6 tokens mean: 14.85 tokens max: 33 tokens

• Samples:

  sentence2	sentence1
  Amplified, it becomes louder happens do the sound during a resonance.	What happens do the sound during a resonance?
  Opaque is the term for matter that does not let any light pass through.	What is the term for matter that does not let any light pass through?
  When drinking water is treated, the term for when chemicals cause solids in the water to clump together is coagulation.	When drinking water is treated, what is the term for when chemicals cause solids in the water to clump together?

• Loss: AdaptiveLayerLoss with these parameters:

  {
      "loss": "GISTEmbedLoss",
      "n_layers_per_step": -1,
      "last_layer_weight": 1.75,
      "prior_layers_weight": 0.5,
      "kl_div_weight": 1.25,
      "kl_temperature": 0.9
  }
  

scitail-pairs-pos

• Dataset: scitail-pairs-pos at 0cc4353
• Size: 8,600 training samples
• Columns: sentence1 and sentence2
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2
  type	string	string
  details	min: 7 tokens mean: 23.74 tokens max: 71 tokens	min: 7 tokens mean: 15.41 tokens max: 38 tokens

• Samples:

  sentence1	sentence2
  Sometimes the two sides of a fracture moved due to the pressure and a fault was formed.	A fault is the fracture caused when rocks on both sides move.
  If it loses one or more electrons, it becomes a positively charged ion.	Remember electrons are negatively charged, so ions with a positive charge have lost a(n) electron.
  Like protein, one gram of carbohydrate converts into four calories.	One gram of carbohydrates provides four calories of energy.

• Loss: AdaptiveLayerLoss with these parameters:

  {
      "loss": "GISTEmbedLoss",
      "n_layers_per_step": -1,
      "last_layer_weight": 1.75,
      "prior_layers_weight": 0.5,
      "kl_div_weight": 1.25,
      "kl_temperature": 0.9
  }
  

xsum-pairs

• Dataset: xsum-pairs at 788ddaf
• Size: 3,500 training samples
• Columns: sentence1 and sentence2
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2
  type	string	string
  details	min: 10 tokens mean: 195.07 tokens max: 506 tokens	min: 5 tokens mean: 25.76 tokens max: 54 tokens

• Samples:

  sentence1	sentence2
  North Wales Police responded to a report of the incident close to the Jade Jones Pavilion at about 14:00 BST on Sunday. Det Ch Insp Arwyn Jones said members of the public detained the man until police arrived. A 50-year-old man was arrested. The child was not hurt and is "safe". "I would like to reassure the public that the child is safe with her family, and the male remains in police custody," said Det Ch Insp Jones.	A man has been arrested after a suspected attempted abduction of a three-year-old girl in Flint, police have said.
  The Neopalpa donaldtrumpi was discovered in California by researcher Vazrick Nazari, of Ottawa, Canada. The name was inspired by the striking golden flakes covering its head, which he likened to Donald Trump's famous mop. Nine species were named after President Barack Obama during his term in office. What Mr Trump will make of the honour remains to be seen - especially as the tiny moth, with its wingspan of just more than a centimetre, is native to southern California and the Mexican region of Baja California. However, Mr Nazari said he hoped it would inspire Mr Trump to prioritise ecological issues during his term in office. He told Live Science: "I hope that the president will make conservation of such fragile ecosystems in the US his top priority. These ecosystems still contain many undiscovered and undescribed species, and deserve to be protected for future generations."	A tiny moth with a very recognisable "hairstyle" has become the first creature named after the soon-to-be 45th president of the United States.
  The best chance of a tight first half was missed by Thierry Audel for the visitors, with his shot being saved by Mitch Walker after a fine run. Paul Cox's men lit up proceedings early in the second, though with Jordan White peeling off his marker nicely and firmly heading in Bradley Bauress's left-wing cross to break the deadlock. Dover plugged away for the equaliser and were rewarded when Femi Ilesanmi crossed for Ryan Bird to head in powerfully. Bedsente Gomis missed a gilt-edged chance to win it for Barrow, hitting over from close range. Match report supplied by the Press Association. Match ends, Dover Athletic 1, Barrow 1. Second Half ends, Dover Athletic 1, Barrow 1. Substitution, Barrow. Daniel Cockerline replaces Harry Panayiotou. Goal! Dover Athletic 1, Barrow 1. Ryan Bird (Dover Athletic). Substitution, Dover Athletic. Tobi Sho-Silva replaces Jamie Allen. Substitution, Dover Athletic. Kadell Daniel replaces Kane Richards. Goal! Dover Athletic 0, Barrow 1. Jordan White (Barrow). Second Half begins Dover Athletic 0, Barrow 0. First Half ends, Dover Athletic 0, Barrow 0. Substitution, Dover Athletic. George Essuman replaces Giancarlo Gallifuoco. First Half begins. Lineups are announced and players are warming up.	Barrow checked Dover's progress by holding them to a draw at the Crabble Athletic Ground.

• Loss: AdaptiveLayerLoss with these parameters:

  {
      "loss": "MultipleNegativesSymmetricRankingLoss",
      "n_layers_per_step": -1,
      "last_layer_weight": 0.75,
      "prior_layers_weight": 1.25,
      "kl_div_weight": 0.8,
      "kl_temperature": 0.75
  }
  

compression-pairs

• Dataset: compression-pairs at 605bc91
• Size: 5,500 training samples
• Columns: sentence1 and sentence2
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2
  type	string	string
  details	min: 12 tokens mean: 31.3 tokens max: 153 tokens	min: 5 tokens mean: 9.99 tokens max: 24 tokens

• Samples:

  sentence1	sentence2
  Opposition Leader Tony Abbott has urged Australians to change the government when they go to the polls.	Abbott urges change of government
  Xyratex Ltd, a leading provider of enterprise class data storage subsystems and storage process technology, announced today that it has expanded its product portfolio to include Hard Disk Drive recording head slider and head gimbal assembly automation and metrology equipment.	Xyratex expands product portfolio
  Manchester City was held to a fifth consecutive draw in the Premier League on Saturday after finishing up 3-3 with Burnley and handing fourth place to Tottenham, which beat Sunderland 2-0.	Man City held to a fifth consecutive draw

• Loss: AdaptiveLayerLoss with these parameters:

  {
      "loss": "MultipleNegativesSymmetricRankingLoss",
      "n_layers_per_step": -1,
      "last_layer_weight": 0.75,
      "prior_layers_weight": 1.25,
      "kl_div_weight": 0.8,
      "kl_temperature": 0.75
  }
  

compression-pairs2

• Dataset: compression-pairs2 at 605bc91
• Size: 4,500 training samples
• Columns: sentence1 and sentence2
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2
  type	string	string
  details	min: 13 tokens mean: 31.11 tokens max: 98 tokens	min: 5 tokens mean: 9.92 tokens max: 22 tokens

• Samples:

  sentence1	sentence2
  President Obama chose not to attend this year's NAACP convention because of ``scheduling'' issues, aides said, explaining it was not because he did not want to appear before the nation's most respected civil rights organization.	Obama not attending NAACP convention due to 'scheduling' issues
  Syrian President Bashar al-Assad warned Monday that a military attack on Iran over its nuclear program would have grave consequences for the US, Israel and the world.	Assad warns against attacking Iran
  A Belleville man and two Clyde residents were arrested on drug charges Thursday and Friday by the North Central Kansas Special Tactics and Response Team.	3 arrested on drug charges

• Loss: AdaptiveLayerLoss with these parameters:

  {
      "loss": "GISTEmbedLoss",
      "n_layers_per_step": 3,
      "last_layer_weight": 0.15,
      "prior_layers_weight": 2.5,
      "kl_div_weight": 0.75,
      "kl_temperature": 0.5
  }
  

compression-pairs3

• Dataset: compression-pairs3 at 605bc91
• Size: 4,500 training samples
• Columns: sentence1 and sentence2
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2
  type	string	string
  details	min: 10 tokens mean: 32.56 tokens max: 448 tokens	min: 5 tokens mean: 10.14 tokens max: 31 tokens

• Samples:

  sentence1	sentence2
  The Ministry of Justice has confirmed that limited court sittings will continue on Tuesday at the Christchurch Court House.	Limited court sittings to continue on Tuesday
  As a new mother, a few things come to mind when I watch the funny ``woman falls in fountain while texting'' video: 1.	Woman falls in fountain while texting:
  Congress plans to come out with a chargesheet'' against Nitish Kumar government in poll bound Bihar to expose its hollow claims'' of development and will highlight the achievements of the Centre to woo electorate in the Assembly elections there.	Congress to come out with 'chargesheet' against Nitish

• Loss: AdaptiveLayerLoss with these parameters:

  {
      "loss": "MultipleNegativesRankingLoss",
      "n_layers_per_step": 3,
      "last_layer_weight": 0.05,
      "prior_layers_weight": 10,
      "kl_div_weight": 5,
      "kl_temperature": 0.2
  }
  

sciq_pairs

• Dataset: sciq_pairs at 2c94ad3
• Size: 10,000 training samples
• Columns: sentence1 and sentence2
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2
  type	string	string
  details	min: 7 tokens mean: 17.11 tokens max: 74 tokens	min: 2 tokens mean: 85.83 tokens max: 512 tokens

• Samples:

  sentence1	sentence2
  A sequence of amino acids in single polypeptide is the primary structure for what?	Protein Structure. Primary protein structure is the sequence of amino acids in a single polypeptide. Secondary protein structure refers to internal shapes, such as alpha helices and beta pleated sheets, that a single polypeptide takes on due to bonds between atoms in different parts of the polypeptide. Tertiary protein structure is the overall three-dimensional shape of a protein consisting of one polypeptide. Quaternary protein structure is the shape of a protein consisting of two or more polypeptides. For a brief animation of protein structure, see www. stolaf. edu/people/giannini/flashanimat/proteins/protein%20structure. swf .
  Where do birds store and moisten food that is waiting to be digested?	Birds have a sac-like structure called a crop to store and moisten food that is waiting to be digested. They also have an organ called a gizzard that contains swallowed stones. The stones make up for the lack of teeth by grinding food, which can then be digested more quickly. Both structures make it easier for the digestive system to produce a steady supply of nutrients from food.
  Concentration of what, the substance left behind when ocean water evaporates, is about 3.5 percent?	Dissolved mineral salts wash into the ocean. As ocean water evaporates, it leaves the salts behind. This makes the water saltier. Ocean water is about 3.5 percent salts. The main salt is sodium chloride.

• Loss: AdaptiveLayerLoss with these parameters:

  {
      "loss": "GISTEmbedLoss",
      "n_layers_per_step": -1,
      "last_layer_weight": 1.75,
      "prior_layers_weight": 0.5,
      "kl_div_weight": 1.25,
      "kl_temperature": 0.9
  }
  

qasc_pairs

• Dataset: qasc_pairs at a34ba20
• Size: 7,889 training samples
• Columns: sentence1 and sentence2
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2
  type	string	string
  details	min: 4 tokens mean: 11.55 tokens max: 29 tokens	min: 18 tokens mean: 35.3 tokens max: 65 tokens

• Samples:

  sentence1	sentence2
  Tumors are a collection of cancer what?	Cancer cells divide more often than normal cells, and grow out of control.. Cancer cells, however, grow out of control and develop into a tumor.. Cancer cells divide more often than normal cells, and often develop into tumors.
  What allows candles to burn and wax to melt?	melting is when solids are heated above their melting point. Softer melt point wax is used to produce an even burning candle.. Heat allows candles to burn and wax to melt.
  How do ships find submarines?	sonar is used to find the location of an object. Surface ships use sonar to locate and track submarines.. Ships use sonar to find submarines

• Loss: AdaptiveLayerLoss with these parameters:

  {
      "loss": "GISTEmbedLoss",
      "n_layers_per_step": -1,
      "last_layer_weight": 1.75,
      "prior_layers_weight": 0.5,
      "kl_div_weight": 1.25,
      "kl_temperature": 0.9
  }
  

qasc_facts_sym

• Dataset: qasc_facts_sym at a34ba20
• Size: 7,889 training samples
• Columns: combinedfact and facts
• Approximate statistics based on the first 1000 samples:

  	combinedfact	facts
  type	string	string
  details	min: 5 tokens mean: 11.73 tokens max: 30 tokens	min: 12 tokens mean: 25.05 tokens max: 47 tokens

• Samples:

  combinedfact	facts
  Head hair is a type of insulation	Head hair is especially important in preventing heat loss from the body.. Insulation serves to both prevent heat loss in the winter and heat gain in the summer..
  fats and proteins can be used for energy by the cells of most organisms	Glucose is used for energy by the cells of most organisms.. After hours of no glucose ingestion, fats and proteins can be used for energy..
  competition may cause animals to fight towards members of a property of genus	competition may cause animals to fight towards members of their own species. Species is a property of genus, genus is a property of family, etc..

• Loss: AdaptiveLayerLoss with these parameters:

  {
      "loss": "MultipleNegativesSymmetricRankingLoss",
      "n_layers_per_step": -1,
      "last_layer_weight": 0.75,
      "prior_layers_weight": 1.25,
      "kl_div_weight": 0.8,
      "kl_temperature": 0.75
  }
  

openbookqa_pairs

• Dataset: openbookqa_pairs
• Size: 4,505 training samples
• Columns: question and fact
• Approximate statistics based on the first 1000 samples:

  	question	fact
  type	string	string
  details	min: 3 tokens mean: 13.81 tokens max: 78 tokens	min: 4 tokens mean: 11.49 tokens max: 30 tokens

• Samples:

  question	fact
  What is animal competition?	if two animals eat the same prey then those animals compete for that pey
  If you wanted to make a metal bed frame, where would you start?	alloys are made of two or more metals
  Places lacking warmth have few what	cold environments contain few organisms

• Loss: AdaptiveLayerLoss with these parameters:

  {
      "loss": "GISTEmbedLoss",
      "n_layers_per_step": -1,
      "last_layer_weight": 1.75,
      "prior_layers_weight": 0.5,
      "kl_div_weight": 1.25,
      "kl_temperature": 0.9
  }
  

msmarco_pairs

• Dataset: msmarco_pairs at 28ff31e
• Size: 5,500 training samples
• Columns: sentence1 and sentence2
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2
  type	string	string
  details	min: 4 tokens mean: 8.67 tokens max: 41 tokens	min: 19 tokens mean: 76.09 tokens max: 258 tokens

• Samples:

  sentence1	sentence2
  how is democracy in ghana	How Democracy Looks Different in Ghana. By On 12/10/16 at 7:00 AM. Opinion Ghana Ghana elections 2016. Updated
  what cell parts are found only in plant cells? what are found only in animal cells?	Vacuoles are large, liquid-filled organelles found only in plant cells. Vacuoles can occupy up to 90% of a cell's volume and have a single membrane. Their main function is as a space-filler in the cell, but they can also fill digestive functions similar to lysosomes (which are also present in plant cells).
  social security retirement number	Social Security Retirement Customer Support Service Phone Number. The customer support phone number of Social Security Retirement is 1-800-772-1213 (Click phone number to call).

• Loss: AdaptiveLayerLoss with these parameters:

  {
      "loss": "GISTEmbedLoss",
      "n_layers_per_step": -1,
      "last_layer_weight": 1.75,
      "prior_layers_weight": 0.5,
      "kl_div_weight": 1.25,
      "kl_temperature": 0.9
  }
  

msmarco_pairs2

• Dataset: msmarco_pairs2 at 28ff31e
• Size: 4,500 training samples
• Columns: sentence1 and sentence2
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2
  type	string	string
  details	min: 4 tokens mean: 8.68 tokens max: 22 tokens	min: 12 tokens mean: 77.19 tokens max: 209 tokens

• Samples:

  sentence1	sentence2
  how many core wire do i need for trailer	Wiring standard 5 core wire for Trailers Australian standards.
  3.#TAB#What what percentage of people with Autism have an average or above average intellect?	Each individual with autism is unique. Many of those on the autism spectrum have exceptional abilities in visual skills, music and academic skills. About 40 percent have average to above average intellectual abilities.Indeed, many persons on the spectrum take deserved pride in their distinctive abilities and “atypical” ways of viewing the world. Others with autism have significant disability and are unable to live independently.irst and foremost, we now know that there is no one cause of autism just as there is no one type of autism. Over the last five years, scientists have identified a number of rare gene changes, or mutations, associated with autism.
  why is anansi such a great folktale hero of the ashanti people	In the language of the Ashanti people, Anansi means spider. the word nan means to spin. Ashanti folk tales are known as Ananisem, which means story and which may or may not be about spiders. About the story In this folk tale, Anansi, trickster figure and hero of many West African tales, seeks to own all the stories known in the world. Anansi the spider, goes to the Sky God, who in reality owns the storiess. The Sky God agrees to sell Anansi all of the stories if he will bring him the hornets, the great python, and the leopard. The tale provides us with insight into beliefs and traditions of the Ashanti people, and reveals the great value they place on their folklore.

• Loss: AdaptiveLayerLoss with these parameters:

  {
      "loss": "GISTEmbedLoss",
      "n_layers_per_step": 3,
      "last_layer_weight": 0.15,
      "prior_layers_weight": 2.5,
      "kl_div_weight": 0.75,
      "kl_temperature": 0.5
  }
  

nq_pairs

• Dataset: nq_pairs at f9e894e
• Size: 2,934 training samples
• Columns: sentence1 and sentence2
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2
  type	string	string
  details	min: 9 tokens mean: 11.82 tokens max: 25 tokens	min: 23 tokens mean: 134.79 tokens max: 512 tokens

• Samples:

  sentence1	sentence2
  when did 8 out of 10 cats does countdown start	8 Out of 10 Cats Does Countdown The first episode aired as part of Channel 4's "mash-up night" on 2 January 2012; the first full series was aired in July 2013 and multiple series have since been commissioned. Episodes feature Jimmy Carr as host; two teams of two contestants feature in most episodes, with Sean Lock and Jon Richardson as permanent team captains (various other personalities, such as Lee Mack, Sarah Millican, Bill Bailey and Claudia Winkleman, deputise when either captain is unavailable to film). Rachel Riley and Susie Dent take up their regular roles from Countdown; Joe Wilkinson appears in many earlier episodes as Rachel's assistant and has also stood in as a team captain.
  who qualifies for the fifa club world cup	FIFA Club World Cup The current format of the tournament involves seven teams competing for the title at venues within the host nation over a period of about two weeks; the winners of that year's AFC Champions League (Asia), CAF Champions League (Africa), CONCACAF Champions League (North America), Copa Libertadores (South America), OFC Champions League (Oceania) and UEFA Champions League (Europe), along with the host nation's national champions, participate in a straight knock-out tournament. The host nation's national champions dispute a play-off against the Oceania champions, from which the winner joins the champions of Asia, Africa and North America at the quarter-finals. The quarter-final winners go on to face the European and South American champions, who enter at the semi-final stage, for a place in the final.
  countries that don't need visa to enter australia	Visa policy of Australia Australia maintains a universal visa regime, meaning that every non-citizen in Australia must have a visa, either as a result of an application, or one granted automatically by law.[2] As of 2015 there is no intention to provide visa free access for any country,[3] however Australia gives a visitor visa exemption to:

• Loss: AdaptiveLayerLoss with these parameters:

  {
      "loss": "GISTEmbedLoss",
      "n_layers_per_step": -1,
      "last_layer_weight": 1.75,
      "prior_layers_weight": 0.5,
      "kl_div_weight": 1.25,
      "kl_temperature": 0.9
  }
  

nq_pairs2

• Dataset: nq_pairs2 at f9e894e
• Size: 2,401 training samples
• Columns: sentence1 and sentence2
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2
  type	string	string
  details	min: 9 tokens mean: 11.85 tokens max: 23 tokens	min: 6 tokens mean: 132.09 tokens max: 512 tokens

• Samples:

  sentence1	sentence2
  who played general marshall in saving private ryan	Harve Presnell His film career was revived when he played William H. Macy's testy father-in-law in Fargo (1996). Subsequent films included The Whole Wide World (1996), Larger than Life (1996), The Chamber (1996), Face/Off (1997), Julian Po (1997), Saving Private Ryan (1998, as General George Marshall), Patch Adams (1998), Walking Across Egypt (1999), The Legend of Bagger Vance (2000), The Family Man (2000), Escanaba in da Moonlight (2001), Mr. Deeds (2002), Super Sucker (2003), Flags of Our Fathers (2006), and Evan Almighty (2007).[2]
  what does tpc stand for in tpc sawgrass	Tournament Players Club Tournament Players Club (TPC) is a chain of public and private golf courses operated by the PGA Tour. Most of the courses either are or have been hosts for PGA Tour events, with the remainder having frequently hosted events on the second-tier Web.com Tour or the over-50s PGA Tour Champions.
  where does the distributor get its power from	Distributor A distributor consists of a rotating arm or rotor inside the distributor cap, on top of the distributor shaft, but insulated from it and the body of the vehicle (ground). The distributor shaft is driven by a gear on the camshaft on most overhead valve engines, and attached directly to a camshaft on most overhead cam engines. (The distributor shaft may also drive the oil pump.) The metal part of the rotor contacts the high voltage cable from the ignition coil via a spring-loaded carbon brush on the underside of the distributor cap. The metal part of the rotor arm passes close to (but does not touch) the output contacts which connect via high tension leads to the spark plug of each cylinder. As the rotor spins within the distributor, electric current is able to jump the small gaps created between the rotor arm and the contacts due to the high voltage created by the ignition coil.[2]

• Loss: AdaptiveLayerLoss with these parameters:

  {
      "loss": "GISTEmbedLoss",
      "n_layers_per_step": 3,
      "last_layer_weight": 0.15,
      "prior_layers_weight": 2.5,
      "kl_div_weight": 0.75,
      "kl_temperature": 0.5
  }
  

trivia_pairs

• Dataset: trivia_pairs at a7c36e3
• Size: 9,700 training samples
• Columns: sentence1 and sentence2
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2
  type	string	string
  details	min: 8 tokens mean: 16.94 tokens max: 44 tokens	min: 36 tokens mean: 446.46 tokens max: 512 tokens

• Samples:

  sentence1	sentence2
  Directed by Joss Whedon, which 2005 science fiction movie was based on a short-lived TV show that was cancelled after only eleven of its episodes were broadcast?	serenity film : definition of serenity film and synonyms of serenity film (English) $38,869,464 [1] Serenity is a 2005 space western film written and directed by Joss Whedon . It is a continuation of the short-lived 2002 Fox science fiction television series Firefly , taking place after the events of the final episode . Set in 2517, Serenity is the story of the captain and crew of Serenity , a "Firefly-class" spaceship. The captain and first mate are veterans of the Unification War, having fought on the losing side. Their lives of petty crime are interrupted by a psychic passenger who harbors a dangerous secret. The film was released in North America on September 30, 2005 by Universal Pictures . It received generally positive reviews and was number two during its opening weekend but it did not make back its budget until its release on DVD. Serenity won numerous awards, including the 2006 Hugo Award for Best Dramatic Presentation . Contents 11 External links   Plot By 2517, the Alliance has won a war with the less established planets of outer star systems. A young girl named River Tam is the most promising of a number of young people being mentally and physically conditioned by Alliance scientists. Rescued by her brother Simon , the two find refuge aboard the Firefly-class transport ship Serenity, captained by Malcolm "Mal" Reynolds . An Alliance agent, the Operative , is tasked with neutralizing River, as she has been visited by high Alliance politicians and may have learned certain top secrets due to her psychic abilities. Aboard Serenity, Mal takes River along on a bank robbery against her brother's objections. They are attacked by another group of raiders: Reavers, animalistic humans who savagely eat their victims alive. They escape, but Simon decides that he and River will leave Serenity at the next port. While Mal meets fences Fanty and Mingo at a bar, a TV commercial causes River to attack the patrons with superbly effective martial arts. Before she can shoot Mal, Simon arrives and shouts a "safe word" which causes her to fall asleep. Mal carries River back to Serenity, but the incident is captured on camera. The crew contacts a reclusive hacker known as Mr. Universe who analyzes the commercial and discovers a subliminal message being broadcast across Alliance space designed to trigger River. He notes that River whispered "Miranda" before attacking. Mal receives a call from Inara , a former passenger. Suspecting a trap but fearing she is in danger, Mal goes to her and is confronted by the Operative. The Operative offers to let Mal go free if he turns River over, but thanks to Inara's quick thinking, she and Mal escape. After another of River's outbursts, the crew discover that Miranda is a remote planet thought to have been rendered uninhabitable as the result of a terraforming accident. Serenity returns to Haven, a mining colony and home of Shepherd Book , a former passenger and friend. They discover, however, that the outpost has been destroyed and the residents slaughtered. The Operative has ordered the deaths of all of Mal's contacts to deny him a safe haven and promises that he will continue to pursue them until he gets River. The way to Miranda is blocked by a region swarming with Reavers, so Mal disguises Serenity as a Reaver ship. After sailing through a fleet of Reaver vessels, the crew discovers a habitable planet, but only corpses strew its cities. They discover a log recorded by an Alliance survey team explaining that the Alliance administered a chemical designed to suppress aggression in the residents, which worked too well, making them so passive they simply lay down and died. However, 0.1% of the population instead reacted by becoming hyper-aggressive and unstable, explaining the origin of the Reavers. Mal contacts Mr. Universe to arrange to have the log broadcast to everyone, but the Operative is already there and has Mr. Universe lure them in. Mal knows Serenity is heading into a trap, so he opens fire on one of the Reaver ships. The Reavers pursue Serenity to Mr. Universe's planet where they engag
  In which year was Ulster incorporated into the UK, under the Home Rule Act?	BBC - History - British History in depth: Irish Home Rule: An imagined future Print this page Self-government The casual observer could be forgiven for thinking that the picture entitled ‘King George and Queen Mary Opening the New Irish Parliament’ was a faithful representation of actual events. But, in fact, the occasion never happened. The image was instead one of a number of propaganda postcards produced in the years before World War One in which various artists tried to imagine what Ireland would be like under Home Rule. Yet even though Home Rule was never achieved, this image is much more than just a historical curiosity. Pictures like this help us to make the imaginative leap back to a time in Ireland’s history when the establishment of a Home Rule parliament in Dublin was to the majority of Irish people a real and immediate prospect. As one English journalist visiting Ireland in 1893 (the year of the second Home Rule Bill) recorded: 'self-government was the only topic of conversation in hotels, railway carriages, tramcars, and on the steps of the temples, at the corners of the streets, in the music halls.' Charles Stewart Parnell   © Between 1801 and 1922 Ireland formed a constituent part of the United Kingdom. At various intervals during this time, attempts were made to destabilise Anglo-Irish relations. Rebellions were launched in 1803, 1848, 1867, and 1916 to try and end British rule over Ireland. Daniel O’Connell in the 1830-1840s campaigned to repeal the Act of Union. But from the 1870s onwards Irish Nationalists (under Isaac Butt) favoured Home Rule. It was not until 1886, however, that the first attempt to legislate Home Rule was made. Nineteen years were to pass before another Home Rule Bill was introduced. The Liberal government, led by WE Gladstone and supported by the Irish Parliamentary Party under Charles Stewart Parnell, introduced a Home Rule bill in the House of Commons. British and Irish Unionists (so-called because they defended the union of 1801) defeated it. By the time another bill was introduced in 1893, Parnell was dead (having earlier been deposed following a messy divorce scandal) and his followers were acrimoniously divided. Gladstone’s second attempt was passed by the House of Commons, but was rejected by the House of Lords. Nineteen years were to pass before another Home Rule Bill was introduced in 1912. The Home Rule Bill This bill proposed the creation of a bi-cameral legislative assembly subordinate to the imperial parliament in London. It had carefully circumscribed powers over domestic issues and numerous constitutional safeguards to protect Protestants. But while the debate at Westminster focused on Home Rule finance and the protection of minorities, the discussion on the streets of Dublin, Belfast, Cork and elsewhere reflected more personal hopes, fears, and aspirations for the future. Nationalist politicians described Home Rule as the 'promised land'. The cause of Irish self-government was certainly interwoven with centuries-old memories of Catholic dispossession and Protestant ascendancy on the one hand and popish plots and moonlit intimidation on the other. Sectarian riots raged after a Catholic allegedly told a Protestant that none of "his sort" would find a job under Home Rule. Accordingly, expectations that a Dublin parliament would right old wrongs or settle old scores flourished. In 1886, for instance, sectarian riots in Belfast raged for several months after a Catholic docker allegedly told a Protestant worker that none of ‘his sort’ would find employment under Home Rule. Similarly, an English visitor to Ireland in 1893 encountered wildly optimistic expectations among Catholics and Nationalists: ‘Every man … [possessed] a visionary scheme of which he had all the absurd particulars.' Accordingly, advocates of Home Rule were aware of the need to manage popular expectations, as one argued in 1914: ‘In an autonomous Ireland public life would not be all nougat, velvet, and soft music. There will be … vehement conflicts, for that is the way of the twentieth century.' To
  Who killed at least 17 men and boys before being arrested in Milwaukee in 1991?	Newsroom

• Loss: AdaptiveLayerLoss with these parameters:

  {
      "loss": "GISTEmbedLoss",
      "n_layers_per_step": -1,
      "last_layer_weight": 1.75,
      "prior_layers_weight": 0.5,
      "kl_div_weight": 1.25,
      "kl_temperature": 0.9
  }
  

quora_pairs

• Dataset: quora_pairs at 451a485
• Size: 4,365 training samples
• Columns: sentence1 and sentence2
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2
  type	string	string
  details	min: 6 tokens mean: 13.38 tokens max: 41 tokens	min: 6 tokens mean: 13.42 tokens max: 43 tokens

• Samples:

  sentence1	sentence2
  Is it possible to gain height after 20?	How can I increase in height after 20 years?
  How will Indian GDP be affected from banning 500 and 1000 rupees notes?	How will India be affected now that 500 and 1000 rupee notes have been banned?
  What are restricted stock units? What are some examples?	What are some facts about restricted stock units?

• Loss: AdaptiveLayerLoss with these parameters:

  {
      "loss": "GISTEmbedLoss",
      "n_layers_per_step": 3,
      "last_layer_weight": 0.15,
      "prior_layers_weight": 2.5,
      "kl_div_weight": 0.75,
      "kl_temperature": 0.5
  }
  

gooaq_pairs

• Dataset: gooaq_pairs at b089f72
• Size: 5,500 training samples
• Columns: sentence1 and sentence2
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2
  type	string	string
  details	min: 8 tokens mean: 11.56 tokens max: 19 tokens	min: 12 tokens mean: 58.24 tokens max: 135 tokens

• Samples:

  sentence1	sentence2
  what are the gross error?	Gross errors are caused by experimenter carelessness or equipment failure. These "outliers" are so far above or below the true value that they are usually discarded when assessing data. The "Q-Test" (discussed later) is a systematic way to determine if a data point should be discarded.
  what is ip for url?	IP Address (Internet Protocol Address) An identifier for a computer or device on a TCP/IP network. Because IP addresses are represented by numbers and they are too complicated to remember for humans, the idea of naming sites resulted in Web URLs and e-mail addresses.
  when is the umbrella academy back?	The Umbrella Academy season 2 finally has a release date! The Hargreeves family returns on July 31, 2020.

• Loss: AdaptiveLayerLoss with these parameters:

  {
      "loss": "GISTEmbedLoss",
      "n_layers_per_step": -1,
      "last_layer_weight": 1.75,
      "prior_layers_weight": 0.5,
      "kl_div_weight": 1.25,
      "kl_temperature": 0.9
  }
  

gooaq_pairs2

• Dataset: gooaq_pairs2 at b089f72
• Size: 4,500 training samples
• Columns: sentence1 and sentence2
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2
  type	string	string
  details	min: 8 tokens mean: 11.45 tokens max: 22 tokens	min: 13 tokens mean: 57.71 tokens max: 134 tokens

• Samples:

  sentence1	sentence2
  what happens if you drink a lot of water fast?	Too much fluid will dilute the amount of sodium in the bloodstream, leading to abnormally low levels, called hyponatremia. Symptoms of hyponatremia may be mild at first, such as a feeling of nausea or bloating. Symptoms can become severe, especially when sodium levels suddenly drop.
  what is the difference between a sabre saw and a reciprocating saw?	While the reciprocating saw excels when used for demolition purposes, both types of saw will cut both metals and wood. ... The sabre saw has a smaller motor and thus isn't powerful enough for these larger jobs (in some ways the sabre saw can be considered a larger version of the jog saw, which is used for detail work).
  how to change address in my driver's license?	To change your address on your registration, you have to go to the DMV and do it in person. What you'll need: Change of address form, date of birth, driver's license number, new/old mailing address, new/old residence address, proof of identification, proof of residence, and social security number.

• Loss: AdaptiveLayerLoss with these parameters:

  {
      "loss": "GISTEmbedLoss",
      "n_layers_per_step": 3,
      "last_layer_weight": 0.15,
      "prior_layers_weight": 2.5,
      "kl_div_weight": 0.75,
      "kl_temperature": 0.5
  }
  

mrpc_pairs

• Dataset: mrpc_pairs at bcdcba7
• Size: 2,474 training samples
• Columns: sentence1 and sentence2
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2
  type	string	string
  details	min: 9 tokens mean: 26.33 tokens max: 49 tokens	min: 10 tokens mean: 26.34 tokens max: 48 tokens

• Samples:

  sentence1	sentence2
  " The bolt catcher is not as robust as it is supposed to be , " the board 's chairman , retired Adm. Hal Gehman , said .	" The bolt catcher is not as robust " as it should be , said retired Navy Admiral Harold W. Gehman Jr . , the board chairman .
  To the caller who claimed to have found a mini-cassette from the space shuttle Columbia : NASA is extremely interested in talking to you , " no-questions-asked " .	To the Sound-Off caller who said they found a mini-cassette from the space shuttle Columbia : NASA is extremely interested in talking to you on a " no-questions-asked " basis .
  " Everything is going to move everywhere , " Doug Feith , undersecretary of defence for policy , said in an interview .	" Everything is going to move everywhere , " Douglas J. Feith , undersecretary of defence force policy , is quoted as saying .

• Loss: AdaptiveLayerLoss with these parameters:

  {
      "loss": "MultipleNegativesSymmetricRankingLoss",
      "n_layers_per_step": -1,
      "last_layer_weight": 0.75,
      "prior_layers_weight": 1.25,
      "kl_div_weight": 0.8,
      "kl_temperature": 0.75
  }
  

Evaluation Datasets

nli-pairs

• Dataset: nli-pairs at d482672
• Size: 160 evaluation samples
• Columns: anchor and positive
• Approximate statistics based on the first 1000 samples:

  	anchor	positive
  type	string	string
  details	min: 5 tokens mean: 17.27 tokens max: 36 tokens	min: 5 tokens mean: 9.57 tokens max: 21 tokens

• Samples:

  anchor	positive
  Two women are embracing while holding to go packages.	Two woman are holding packages.
  Two young children in blue jerseys, one with the number 9 and one with the number 2 are standing on wooden steps in a bathroom and washing their hands in a sink.	Two kids in numbered jerseys wash their hands.
  A man selling donuts to a customer during a world exhibition event held in the city of Angeles	A man selling donuts to a customer.

• Loss: AdaptiveLayerLoss with these parameters:

  {
      "loss": "GISTEmbedLoss",
      "n_layers_per_step": -1,
      "last_layer_weight": 1.75,
      "prior_layers_weight": 0.5,
      "kl_div_weight": 1.25,
      "kl_temperature": 0.9
  }
  

vitaminc-pairs

• Dataset: vitaminc-pairs at be6febb
• Size: 160 evaluation samples
• Columns: claim and evidence
• Approximate statistics based on the first 1000 samples:

  	claim	evidence
  type	string	string
  details	min: 9 tokens mean: 20.73 tokens max: 41 tokens	min: 11 tokens mean: 34.56 tokens max: 79 tokens

• Samples:

  claim	evidence
  Dragon Con had over 5000 guests .	Among the more than 6000 guests and musical performers at the 2009 convention were such notables as Patrick Stewart , William Shatner , Leonard Nimoy , Terry Gilliam , Bruce Boxleitner , James Marsters , and Mary McDonnell .
  COVID-19 has reached more than 185 countries .	As of , more than cases of COVID-19 have been reported in more than 190 countries and 200 territories , resulting in more than deaths .
  In March , Italy had 3.6x times more cases of coronavirus than China .	As of 12 March , among nations with at least one million citizens , Italy has the world 's highest per capita rate of positive coronavirus cases at 206.1 cases per million people ( 3.6x times the rate of China ) and is the country with the second-highest number of positive cases as well as of deaths in the world , after China .

• Loss: AdaptiveLayerLoss with these parameters:

  {
      "loss": "GISTEmbedLoss",
      "n_layers_per_step": -1,
      "last_layer_weight": 1.75,
      "prior_layers_weight": 0.5,
      "kl_div_weight": 1.25,
      "kl_temperature": 0.9
  }
  

qnli-contrastive

• Dataset: qnli-contrastive at bcdcba7
• Size: 160 evaluation samples
• Columns: sentence1, sentence2, and label
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2	label
  type	string	string	int
  details	min: 7 tokens mean: 14.57 tokens max: 29 tokens	min: 4 tokens mean: 37.21 tokens max: 115 tokens	0: 100.00%

• Samples:

  sentence1	sentence2	label
  What came into force after the new constitution was herald?	As of that day, the new constitution heralding the Second Republic came into force.	0
  What is the first major city in the stream of the Rhine?	The most important tributaries in this area are the Ill below of Strasbourg, the Neckar in Mannheim and the Main across from Mainz.	0
  What is the minimum required if you want to teach in Canada?	In most provinces a second Bachelor's Degree such as a Bachelor of Education is required to become a qualified teacher.	0

• Loss: AdaptiveLayerLoss with these parameters:

  {
      "loss": "OnlineContrastiveLoss",
      "n_layers_per_step": -1,
      "last_layer_weight": 0.75,
      "prior_layers_weight": 1.25,
      "kl_div_weight": 0.8,
      "kl_temperature": 0.75
  }
  

scitail-pairs-qa

• Dataset: scitail-pairs-qa at 0cc4353
• Size: 160 evaluation samples
• Columns: sentence2 and sentence1
• Approximate statistics based on the first 1000 samples:

  	sentence2	sentence1
  type	string	string
  details	min: 8 tokens mean: 16.02 tokens max: 31 tokens	min: 8 tokens mean: 15.31 tokens max: 28 tokens

• Samples:

  sentence2	sentence1
  The axis of the plant has a root end and a shoot end.	What part of the plant has a root end and a shoot end?
  All cells have the small size in common.	What do all cells have in common?
  In the nuclear fusion process, two light nuclei combine to produce a heavier nucleus and great energy.	In which process do two light nuclei combine to produce a heavier nucleus and great energy?

• Loss: AdaptiveLayerLoss with these parameters:

  {
      "loss": "GISTEmbedLoss",
      "n_layers_per_step": -1,
      "last_layer_weight": 1.75,
      "prior_layers_weight": 0.5,
      "kl_div_weight": 1.25,
      "kl_temperature": 0.9
  }
  

scitail-pairs-pos

• Dataset: scitail-pairs-pos at 0cc4353
• Size: 160 evaluation samples
• Columns: sentence1 and sentence2
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2
  type	string	string
  details	min: 7 tokens mean: 23.01 tokens max: 61 tokens	min: 8 tokens mean: 15.46 tokens max: 36 tokens

• Samples:

  sentence1	sentence2
  An introduction to atoms and elements, compounds, atomic structure and bonding, the molecule and chemical reactions.	Replace another in a molecule happens to atoms during a substitution reaction.
  Wavelength The distance between two consecutive points on a sinusoidal wave that are in phase;	Wavelength is the distance between two corresponding points of adjacent waves called.
  humans normally have 23 pairs of chromosomes.	Humans typically have 23 pairs pairs of chromosomes.

• Loss: AdaptiveLayerLoss with these parameters:

  {
      "loss": "GISTEmbedLoss",
      "n_layers_per_step": -1,
      "last_layer_weight": 1.75,
      "prior_layers_weight": 0.5,
      "kl_div_weight": 1.25,
      "kl_temperature": 0.9
  }
  

xsum-pairs

• Dataset: xsum-pairs at 788ddaf
• Size: 160 evaluation samples
• Columns: sentence1 and sentence2
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2
  type	string	string
  details	min: 21 tokens mean: 179.72 tokens max: 343 tokens	min: 13 tokens mean: 25.66 tokens max: 40 tokens

• Samples:

  sentence1	sentence2
  A Met Office yellow warning to "be aware" of snow is in place for north-west England and much of the Midlands, while snow has also fallen in parts of Scotland. Here are images from the areas that have seen plenty of the frozen side of winter so far. This robin stood out among the snowy landscape in County Durham. Cows were seen being fed on a snowy farm in Loch Lomond, which is located between central Scotland and the Highlands. This shopping trolley got a dusting of the snow that fell in Buxton, Derbyshire. Black sheep stand out even more when they are grazing in the white fields of Ashbourne in Derbyshire. By contrast, these swans are hard to spot in the snow at Tarnfield Park in Yeadon, Leeds. Unfortunately the icy conditions have also meant driving has become treacherous in some areas - this car fell foul of the roads in Sheffield. Frozen weather was not enough to stop these dogs from going out on a walk in Bradford. These dog walkers were also out, despite the freezing conditions, alongside the A58 near Ripponden, West Yorkshire. Also out on the roads was this bus, tackling the snowy A62 near the village of Marsden, West Yorkshire.	A spell of cold weather across the UK has seen some parts of the country being covered in snow.
  The party challenged the verdict blocking its plans to prevent members who joined after 12 January from voting unless they paid £25 extra. It said its ruling NEC, not courts, was the "ultimate arbiter" of the rules. The Court of Appeal upheld the challenge.	Labour has won an appeal against a High Court ruling that allowed new members a vote in its leadership contest.
  26 August 2016 Last updated at 10:29 BST An archaeological excavation in Clones, County Monaghan, had failed to uncover the castle but, unknown to experts, it was standing nearby. County heritage officer Shirley Clerkin took BBC News NI's Julian Fowler on a tour.	A 17th century castle "lost" for more than 250 years has been rediscovered in the centre of a town on the Irish border.

• Loss: AdaptiveLayerLoss with these parameters:

  {
      "loss": "MultipleNegativesSymmetricRankingLoss",
      "n_layers_per_step": -1,
      "last_layer_weight": 0.75,
      "prior_layers_weight": 1.25,
      "kl_div_weight": 0.8,
      "kl_temperature": 0.75
  }
  

compression-pairs

• Dataset: compression-pairs at 605bc91
• Size: 160 evaluation samples
• Columns: sentence1 and sentence2
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2
  type	string	string
  details	min: 11 tokens mean: 33.44 tokens max: 177 tokens	min: 6 tokens mean: 10.18 tokens max: 18 tokens

• Samples:

  sentence1	sentence2
  Lindsay Lohan was abandoned by girlfriend Samantha Ronson in Las Vegas.	Lindsay Lohan abandoned by Samantha Ronson
  Venezuela stayed unbeaten in Group B of the Copa America with an astonishing late show on Wednesday in Salta to draw 3-3 with Paraguay.	Venezuela stay unbeaten in late show
  Ingram Micro Inc. said Tuesday that it plans to boost profitability through a combination of operational improvements and expansion initiatives.	Ingram Micro to boost profitability through operational improvements, expansion

• Loss: AdaptiveLayerLoss with these parameters:

  {
      "loss": "MultipleNegativesSymmetricRankingLoss",
      "n_layers_per_step": -1,
      "last_layer_weight": 0.75,
      "prior_layers_weight": 1.25,
      "kl_div_weight": 0.8,
      "kl_temperature": 0.75
  }
  

sciq_pairs

• Dataset: sciq_pairs at 2c94ad3
• Size: 160 evaluation samples
• Columns: sentence1 and sentence2
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2
  type	string	string
  details	min: 7 tokens mean: 16.44 tokens max: 34 tokens	min: 2 tokens mean: 77.31 tokens max: 512 tokens

• Samples:

  sentence1	sentence2
  Precipitation, evaporation, transpiration, and condensation are part of what cycle?	The water cycle includes the ocean, atmosphere, ground, and living things. During the water cycle, water keeps changing state by processes such as evaporation, transpiration, condensation, and precipitation.
  What part of the brain is attached to the brain stem but considered a separate region of the adult brain?	Brain Stem The midbrain and hindbrain (composed of the pons and the medulla) are collectively referred to as the brain stem (Figure 13.12). The structure emerges from the ventral surface of the forebrain as a tapering cone that connects the brain to the spinal cord. Attached to the brain stem, but considered a separate region of the adult brain, is the cerebellum. The midbrain coordinates sensory representations of the visual, auditory, and somatosensory perceptual spaces. The pons is the main connection with the cerebellum. The pons and the medulla regulate several crucial functions, including the cardiovascular and respiratory systems and rates.
  Birds have a relatively large heart and a rapid what?	To keep their flight muscles well supplied with oxygen, birds evolved specialized respiratory and circulatory systems. Birds have special air sacs for storing extra air and pumping it into the lungs. They also have a relatively large heart and a rapid heart rate. These adaptations keep plenty of oxygenated blood circulating to the flight muscles.

• Loss: AdaptiveLayerLoss with these parameters:

  {
      "loss": "GISTEmbedLoss",
      "n_layers_per_step": -1,
      "last_layer_weight": 1.75,
      "prior_layers_weight": 0.5,
      "kl_div_weight": 1.25,
      "kl_temperature": 0.9
  }
  

qasc_pairs

• Dataset: qasc_pairs at a34ba20
• Size: 160 evaluation samples
• Columns: sentence1 and sentence2
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2
  type	string	string
  details	min: 4 tokens mean: 11.39 tokens max: 20 tokens	min: 20 tokens mean: 34.17 tokens max: 55 tokens

• Samples:

  sentence1	sentence2
  What changing can cause change of state?	temperature changing can cause phase changes. Phase changes Change of State.. temperature changing can cause change of state.
  What do substances in the pituitary control?	Most pituitary hormones control other endocrine glands.. Hormone therapy Hormones are substances that occur naturally in the body.. Substances in the pituitary control the endocrine glands.
  Conserving water can be used for survival in which region?	conserving water can be used for survival in a dry environment. By definition, deserts are dry areas.. conserving water can be used for survival in the desert

• Loss: AdaptiveLayerLoss with these parameters:

  {
      "loss": "GISTEmbedLoss",
      "n_layers_per_step": -1,
      "last_layer_weight": 1.75,
      "prior_layers_weight": 0.5,
      "kl_div_weight": 1.25,
      "kl_temperature": 0.9
  }
  

qasc_facts_sym

• Dataset: qasc_facts_sym at a34ba20
• Size: 160 evaluation samples
• Columns: combinedfact and facts
• Approximate statistics based on the first 1000 samples:

  	combinedfact	facts
  type	string	string
  details	min: 5 tokens mean: 11.79 tokens max: 23 tokens	min: 15 tokens mean: 24.64 tokens max: 48 tokens

• Samples:

  combinedfact	facts
  Cells are the basic units of the structure and function of tigers	Cells are the basic units of the structure and function of living things.. Tigers are the largest of the living cats..
  water flow through soil is variable	if soil is permeable then water easily flows through that soil. Permeability is affected by many soil characteristics..
  If seeds stick to a rabbit then that seed will be transported by the animal	if seeds stick to the fur of an animal then that seed will be transported by the animal. Rabbits' brown summer fur is replaced with fur that is greyer..

• Loss: AdaptiveLayerLoss with these parameters:

  {
      "loss": "MultipleNegativesSymmetricRankingLoss",
      "n_layers_per_step": -1,
      "last_layer_weight": 0.75,
      "prior_layers_weight": 1.25,
      "kl_div_weight": 0.8,
      "kl_temperature": 0.75
  }
  

openbookqa_pairs

• Dataset: openbookqa_pairs
• Size: 160 evaluation samples
• Columns: question and fact
• Approximate statistics based on the first 1000 samples:

  	question	fact
  type	string	string
  details	min: 3 tokens mean: 13.64 tokens max: 45 tokens	min: 4 tokens mean: 11.46 tokens max: 28 tokens

• Samples:

  question	fact
  The thermal production of a stove is generically used for	a stove generates heat for cooking usually
  What creates a valley?	a valley is formed by a river flowing
  when it turns day and night on a planet, what cause this?	a planet rotating causes cycles of day and night on that planet

• Loss: AdaptiveLayerLoss with these parameters:

  {
      "loss": "GISTEmbedLoss",
      "n_layers_per_step": -1,
      "last_layer_weight": 1.75,
      "prior_layers_weight": 0.5,
      "kl_div_weight": 1.25,
      "kl_temperature": 0.9
  }
  

msmarco_pairs

• Dataset: msmarco_pairs at 28ff31e
• Size: 160 evaluation samples
• Columns: sentence1 and sentence2
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2
  type	string	string
  details	min: 4 tokens mean: 8.22 tokens max: 28 tokens	min: 24 tokens mean: 77.72 tokens max: 212 tokens

• Samples:

  sentence1	sentence2
  what foods are low in sodium and sugar	Whole grains, such as brown rice, oatmeal and whole wheat pasta, are tasty options to include when looking for foods low in fat, sugar and sodium. Whole grains are a good source of B vitamins, vitamin E, magnesium, iron and fiber.
  symptoms of esophageal achalasia	1 The cause of achalasia is unknown; however, there is degeneration of the esophageal muscles and, more importantly, the nerves that control the muscles. 2 Common symptoms of achalasia include. 3 difficulty in swallowing (dysphagia), 4 chest pain, and. regurgitation of food and liquids.
  james bond newest film	SPECTRE, the 24th James Bond film, will be the fourth outing for Daniel Craig as 007, and the second film to be directed by Sam Mendes. The new MI6 team of Ralph Fiennes (M), Naomie Harris (Moneypenny), Rory Kinnear (Tanner) and Ben Whishaw (Q) are all reprising their roles.

• Loss: AdaptiveLayerLoss with these parameters:

  {
      "loss": "GISTEmbedLoss",
      "n_layers_per_step": -1,
      "last_layer_weight": 1.75,
      "prior_layers_weight": 0.5,
      "kl_div_weight": 1.25,
      "kl_temperature": 0.9
  }
  

nq_pairs

• Dataset: nq_pairs at f9e894e
• Size: 160 evaluation samples
• Columns: sentence1 and sentence2
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2
  type	string	string
  details	min: 9 tokens mean: 11.66 tokens max: 19 tokens	min: 19 tokens mean: 132.15 tokens max: 361 tokens

• Samples:

  sentence1	sentence2
  who teaches yoda to return through the force	Yoda In the final arc of the final season, Yoda hears Qui-Gon Jinn speaking to him from beyond the grave. Yoda flees the Jedi Temple with R2-D2 to travel to Dagobah, the planet he would make his home when he enters exile after events of Revenge of the Sith, to find answers. Shown cryptic visions of the fall of the Jedi, Yoda learns he has been "chosen" to learn how to manifest his consciousness after death as a Force ghost. Yoda is tested by a group of spirit priestesses in order to overcome trials and temptations on his pilgrimage; one of these tests is to face an illusion of ancient Sith lord Darth Bane. Yoda's final test is to resist an attempt by Darth Sidious and Dooku to lure him to the dark side with a false vision of deceased Jedi Master Sifo Dyas. Yoda engages in a metaphysical battle with Sidious, and appears to sacrifice his life in order to save Anakin's - only to awaken and discover that the battle was merely a vision, and that he has passed the test. The priestesses inform Yoda that his training will resume in time.
  where does the phrase creme dela creme come from	Crème de la crème Crème de la crème (French, lit. 'cream of the cream') is an idiom meaning the "best of the best", "superlative", or "the very best".
  where did they film this is where i leave you	This Is Where I Leave You This is Where I Leave You began principal photography on May 13, 2013 in New York City.[6] The home is located in Munsey Park on Long Island. The skating rink was in The Bellmores, New York. The synagogue interior and exterior scenes were actually shot at Congregation Kneses Tifereth Israel in Port Chester, New York. [7][8] Approximately 40 members of the congregation played extras in the scenes.[9]

• Loss: AdaptiveLayerLoss with these parameters:

  {
      "loss": "GISTEmbedLoss",
      "n_layers_per_step": -1,
      "last_layer_weight": 1.75,
      "prior_layers_weight": 0.5,
      "kl_div_weight": 1.25,
      "kl_temperature": 0.9
  }
  

trivia_pairs

• Dataset: trivia_pairs at a7c36e3
• Size: 160 evaluation samples
• Columns: sentence1 and sentence2
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2
  type	string	string
  details	min: 8 tokens mean: 16.2 tokens max: 37 tokens	min: 70 tokens mean: 445.36 tokens max: 512 tokens

• Samples:

  sentence1	sentence2
  What is the smallest woodwind instrument in an orchestra?	Instrument Lab
  Who was the last woman to be executed for murder in the UK?	The British female hanged 1868 - 1955 by Robert Anderson (Evans) on Monday, the 12th of January, 1874 . Thirty one year old Mary Ann Barry was executed alongside her partner in crime, 32 year old Edwin Bailey, for the poisoning murder of his illegitimate one year old child, Sarah, whom they considered a nuisance. With them on the gallows, set up in the quadrangle of Gloucester Gaol, was Edward Butt, who had shot his girlfriend. Mary became the last woman in England to suffer short drop hanging and reportedly struggled for some three minutes on the rope and had to be forced down into the pit by Anderson .� The two men became still almost immediately. � Berry, Elizabeth hanged by James Berry at Walton prison Liverpool on Monday, the 14th of March, 1887 . Thirty one year old Elizabeth Berry poisoned her 11 year old daughter for �10 life insurance. It was an unusual coincidence that the executioner and the criminal had the same surname and had also actually met previously when they danced together at a police ball. Biggadyke, Priscilla, was hanged at 9.00 a.m. Monday, the 28th of December, 1868 , at Lincoln by Thomas Askern for poisoning her husband with arsenic. It was alleged that she killed him because he discovered she was having an affair with one of their lodgers. Thirty five year old Priscilla was the first woman to be executed in private in Britain . She ascended the steps to the platform where she said "Surely all my troubles are over" and "Shame on you, you are not going to hang me." But Askern did, in his usual clumsy way and she reportedly died hard. Britland, Mary Ann was executed by James Berry at Strangeways on Monday, the 9th of August, 1886 , becoming the first woman to be hanged there. Thirty eight year old Mary Ann Britland was convicted of poisoning Mary Dixon, with whose husband she had been having an affair. She had also previously poisoned her own husband, Thomas and daughter, Elizabeth. Bryant, Charlotte , hanged by Tom Pierrepoint at Exeter the 15th of July 1936 . Charlotte Bryant (33) was convicted of poisoning her husband with arsenic. She was having an affair with their lodger and it seemed a simple way to remove her husband from the scene. Whilst awaiting execution, her previously black hair turned completely white. Calvert, Louie , hanged by Tom Pierrepoint at Strangeways prison Manchester Thursday, the 24th of June, 1926 . Louie Calvert, also 33, had criminal tendencies and was known to the police. She battered and strangled her landlady, Mrs. Lily Waterhouse, who had confronted her over things that had gone missing from the house and had reported Louie to the police. In the condemned cell, she also admitted to the murder of a previous employer - John Frobisher - in 1922. She was the first woman to be hanged at Stangeways since Mary Ann Britland in 1886. Tuesday, the 6th of March, 1900 . Twenty four year old Ada Chard-Williams was convicted of drowning a small child whom she had "adopted" for a few pounds. She was suspected of killing other children and was another "baby farmer."� She was the last woman to hang at Newgate, subsequent female executions in London taking place at the newly converted women's prison at Holloway. Christofi, Styllou , hanged by Albert Pierrepoint at London 's Holloway women's prison on Monday, the 13th of December, 1954 . Styllou Christofi, 53, was a Greek woman who brutally murdered her German born daughter-in-law, Hella, by battering her and then strangling her. Afterwards, she tried to burn her body. It is thought that she had also committed another murder in Cyprus . She asked for a Maltese Cross to be put on the wall of the execution chamber and this wish was granted - it remained there until the room was dismantled in 1967. Coincidentally, the murder was committed in the same street where a few months later Ruth Ellis was t
  ‘Play the ball as it lies’ is one of the rules of which sport?	The R&A - Quick Guide to the Rules THE R&A Shop Quick Guide to the Rules This guide provides a simple explanation of common Rules situations. It is not a substitute for the Rules of Golf, which should be consulted whenever any doubts arises. For more information on the points covered, please refer to the relevant Rule. The basic Rules are not as hard to learn as you may think. They would include, for example, where play of a hole commences; what to do when your ball is in water hazard, lost or out of bounds' interference from immovable obstructions such as cart paths; and playing a provisional ball. Downloads BALL IN MOTION DEFLECTED OR STOPPED Relief Situations and Procedures When playing golf, you must play the ball as it lies, whether your ball is in a good lie or a bad lie, unless the Rules allow you to do otherwise. For example, the Rules allow you to move natural objects like leaves and twigs – the Rules call these “loose impediments”. The Rules also permit you to lift and move your ball if you have interference from certain conditions. Sometimes you can move your ball without penalty, e.g. when you have interference due to a man-made object – called “obstructions” - such as a road or path, or when you have interference by an abnormal ground condition, such as casual water and ground under repair. At other times, you may incur a penalty if you wish to move your ball, e.g. when your ball is in a water hazard. Have a look at the sections below to learn more:  Ball Unplayable Rules Academy The information included in the Quick Guide is also available in an on-line course, the Rules Academy. This course is based on the Etiquette Section and Quick Guide to the Rules of Golf and features video, images and diagrams of Rules situations and includes revision questions after each section. It covers the essentials that all golfers should know such as sportsmanship, integrity and respect. It adopts a tee to green approach and tells you what you need to know at each point; from where to tee your ball, to taking relief from various conditions, to holing out and returning the score card At the end of the course, you can take a Level 1 Rules Exam and receive a certificate signed by their guiding player. Visit the Rules Academy  The Etiquette of the Game The Etiquette of the game of golf provides guidelines on the manner in which the game of golf should be played.  If they are followed, all players will gain maximum enjoyment from the game. Learn More Rules Education Whether you want to learn the basics or need to understand the finer details of the Decisions, discover more about the Rules of Golf via The R&A’s publications and The R&A’s formal three-tier education programme. Learn More Explore Etiquette Good etiquette is essential in golf, helping all players to gain maximum enjoyment from the game.  Learn more by reading the etiquette section of the Rules in the Rules Explorer.

• Loss: AdaptiveLayerLoss with these parameters:

  {
      "loss": "GISTEmbedLoss",
      "n_layers_per_step": -1,
      "last_layer_weight": 1.75,
      "prior_layers_weight": 0.5,
      "kl_div_weight": 1.25,
      "kl_temperature": 0.9
  }
  

quora_pairs

• Dataset: quora_pairs at 451a485
• Size: 160 evaluation samples
• Columns: sentence1 and sentence2
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2
  type	string	string
  details	min: 7 tokens mean: 14.08 tokens max: 40 tokens	min: 7 tokens mean: 13.96 tokens max: 35 tokens

• Samples:

  sentence1	sentence2
  To whom are we the human beings slaves?	Are we still slaves?
  What are the best investment strategy for beginners?	What are some of the best investment strategies?
  How can I add my profile picture on Qoura?	How can I upload my profile picture in Quora?

• Loss: AdaptiveLayerLoss with these parameters:

  {
      "loss": "GISTEmbedLoss",
      "n_layers_per_step": 3,
      "last_layer_weight": 0.15,
      "prior_layers_weight": 2.5,
      "kl_div_weight": 0.75,
      "kl_temperature": 0.5
  }
  

gooaq_pairs

• Dataset: gooaq_pairs at b089f72
• Size: 160 evaluation samples
• Columns: sentence1 and sentence2
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2
  type	string	string
  details	min: 8 tokens mean: 11.31 tokens max: 17 tokens	min: 24 tokens mean: 59.25 tokens max: 108 tokens

• Samples:

  sentence1	sentence2
  22 may is celebrated as?	On May 22, 1992, the text of the Convention on Biological Diversity was adopted by the of the United Nations at a conference in Nairobi, Kenya. Since 2001, the International Day for Biological Diversity is celebrated each year on the anniversary of this date.
  are health insurance premiums required on w2?	The Affordable Care Act requires employers to report the cost of coverage under an employer-sponsored group health plan on an employee's Form W-2, Wage and Tax Statement, in Box 12, using Code DD.
  is there a goliath season 4?	Goliath season four was confirmed by Amazon Studios on November 14 with the news the series will be released in 2020. ... The first series was released in October 2016, the second came out in June 2018 and third in October 2019.

• Loss: AdaptiveLayerLoss with these parameters:

  {
      "loss": "GISTEmbedLoss",
      "n_layers_per_step": -1,
      "last_layer_weight": 1.75,
      "prior_layers_weight": 0.5,
      "kl_div_weight": 1.25,
      "kl_temperature": 0.9
  }
  

mrpc_pairs

• Dataset: mrpc_pairs at bcdcba7
• Size: 160 evaluation samples
• Columns: sentence1 and sentence2
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2
  type	string	string
  details	min: 11 tokens mean: 27.26 tokens max: 42 tokens	min: 12 tokens mean: 26.89 tokens max: 41 tokens

• Samples:

  sentence1	sentence2
  The vote came just two days after Kurds swept City Council elections , taking the largest single block of votes on the 30-seat council .	The vote for mayor followed City Council elections that gave Kurds the largest block of votes on the 30-seat council .
  " We are committed to helping the Iraqi people get on the path to a free society , " Rumsfeld said in a speech to the Council on Foreign Relations .	" We are committed to helping the Iraqi people get on the path to a free society , " he said .
  Kelly killed himself after being exposed as the source for a BBC report which claimed the government had embellished evidence of Iraq 's banned weapons to justify the war .	He killed himself after being exposed as the source for a BBC report which claimed the government exaggerated the case for war against Iraq .

• Loss: AdaptiveLayerLoss with these parameters:

  {
      "loss": "MultipleNegativesSymmetricRankingLoss",
      "n_layers_per_step": -1,
      "last_layer_weight": 0.75,
      "prior_layers_weight": 1.25,
      "kl_div_weight": 0.8,
      "kl_temperature": 0.75
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• per_device_train_batch_size: 32
• per_device_eval_batch_size: 32
• learning_rate: 3e-05
• weight_decay: 0.0001
• num_train_epochs: 5
• lr_scheduler_type: cosine_with_restarts
• lr_scheduler_kwargs: {'num_cycles': 3}
• warmup_ratio: 0.2
• save_safetensors: False
• fp16: True
• push_to_hub: True
• hub_model_id: bobox/DeBERTa-ST-AllLayers-v3.1bis-checkpoints-tmp
• hub_strategy: all_checkpoints
• 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: 1
• eval_accumulation_steps: None
• learning_rate: 3e-05
• weight_decay: 0.0001
• adam_beta1: 0.9
• adam_beta2: 0.999
• adam_epsilon: 1e-08
• max_grad_norm: 1.0
• num_train_epochs: 5
• max_steps: -1
• lr_scheduler_type: cosine_with_restarts
• lr_scheduler_kwargs: {'num_cycles': 3}
• 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: True
• label_names: None
• load_best_model_at_end: False
• ignore_data_skip: False
• fsdp: []
• fsdp_min_num_params: 0
• fsdp_config: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}
• 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/DeBERTa-ST-AllLayers-v3.1bis-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
• 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
• batch_sampler: no_duplicates
• multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	Training Loss	trivia pairs loss	nli-pairs loss	mrpc pairs loss	nq pairs loss	xsum-pairs loss	qnli-contrastive loss	vitaminc-pairs loss	openbookqa pairs loss	quora pairs loss	gooaq pairs loss	sciq pairs loss	scitail-pairs-qa loss	scitail-pairs-pos loss	qasc facts sym loss	compression-pairs loss	msmarco pairs loss	qasc pairs loss	StS-test_spearman_cosine	Vitaminc-test_max_ap	mrpc-test_max_ap
0	0	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	0.8972	0.5594	0.8571
0.0127	55	0.6905	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.0253	110	0.4662	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.0380	165	0.5361	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.0506	220	0.642	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.0633	275	0.6881	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.0759	330	0.494	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.0886	385	0.6271	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.1012	440	0.5304	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.1139	495	0.5437	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.1266	550	0.6317	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.1392	605	0.7664	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.1519	660	0.5386	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.1645	715	0.5238	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.1772	770	0.5771	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.1898	825	0.5162	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.2025	880	0.5143	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.2151	935	0.5172	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.2278	990	0.539	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.2405	1045	0.3987	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.2501	1087	-	0.4567	0.8455	0.0463	0.4875	0.2330	0.1238	4.8815	1.7259	0.1201	0.6029	0.2428	0.0710	0.4099	0.1146	0.0833	0.5001	0.2238	0.8924	0.5614	0.8587
0.2531	1100	0.5964	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.2658	1155	0.5515	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.2784	1210	0.4783	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.2911	1265	0.6334	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.3037	1320	0.51	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.3164	1375	0.5821	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.3290	1430	0.416	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.3417	1485	0.6191	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.3543	1540	0.5167	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.3670	1595	0.5405	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.3797	1650	0.6837	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.3923	1705	0.5547	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.4050	1760	0.6971	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.4176	1815	0.6149	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.4303	1870	0.5569	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.4429	1925	0.6145	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.4556	1980	0.6826	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.4682	2035	0.6285	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.4809	2090	0.739	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
0.4936	2145	0.6119	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-

Framework Versions

• Python: 3.10.13
• Sentence Transformers: 3.0.1
• Transformers: 4.42.3
• PyTorch: 2.1.2
• Accelerate: 0.32.1
• Datasets: 2.20.0
• 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",
}

AdaptiveLayerLoss

@misc{li20242d,
    title={2D Matryoshka Sentence Embeddings}, 
    author={Xianming Li and Zongxi Li and Jing Li and Haoran Xie and Qing Li},
    year={2024},
    eprint={2402.14776},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}

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}
}

MultipleNegativesRankingLoss

@misc{henderson2017efficient,
    title={Efficient Natural Language Response Suggestion for Smart Reply}, 
    author={Matthew Henderson and Rami Al-Rfou and Brian Strope and Yun-hsuan Sung and Laszlo Lukacs and Ruiqi Guo and Sanjiv Kumar and Balint Miklos and Ray Kurzweil},
    year={2017},
    eprint={1705.00652},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}