README.md · d0rj/e5-large-en-ru at main

File size: 16,442 Bytes

---
license: mit
language:
- en
- ru
metrics:
- accuracy
- f1
- recall
library_name: transformers
pipeline_tag: sentence-similarity
tags:
- mteb
- retrieval
- retriever
- pruned
- e5
- sentence-transformers
- feature-extraction
- sentence-similarity
model-index:
- name: e5-large-en-ru
  results:
  - task:
      type: Classification
    dataset:
      type: mteb/amazon_counterfactual
      name: MTEB AmazonCounterfactualClassification (en)
      config: en
      split: test
      revision: e8379541af4e31359cca9fbcf4b00f2671dba205
    metrics:
    - type: accuracy
      value: 79.5671641791045
    - type: ap
      value: 44.011060753169424
    - type: f1
      value: 73.76504135120175
  - task:
      type: Reranking
    dataset:
      type: mteb/askubuntudupquestions-reranking
      name: MTEB AskUbuntuDupQuestions
      config: default
      split: test
      revision: 2000358ca161889fa9c082cb41daa8dcfb161a54
    metrics:
    - type: map
      value: 57.69669466706412
    - type: mrr
      value: 70.61370531592138
  - task:
      type: STS
    dataset:
      type: mteb/biosses-sts
      name: MTEB BIOSSES
      config: default
      split: test
      revision: d3fb88f8f02e40887cd149695127462bbcf29b4a
    metrics:
    - type: cos_sim_pearson
      value: 86.36465960226795
    - type: cos_sim_spearman
      value: 84.57602350761223
    - type: euclidean_pearson
      value: 84.31391364490506
    - type: euclidean_spearman
      value: 84.57602350761223
    - type: manhattan_pearson
      value: 84.15796224236456
    - type: manhattan_spearman
      value: 84.3645729064343
  - task:
      type: Reranking
    dataset:
      type: mteb/mind_small
      name: MTEB MindSmallReranking
      config: default
      split: test
      revision: 3bdac13927fdc888b903db93b2ffdbd90b295a69
    metrics:
    - type: map
      value: 31.105698873583098
    - type: mrr
      value: 32.163780846856206
  - task:
      type: STS
    dataset:
      type: mteb/sickr-sts
      name: MTEB SICK-R
      config: default
      split: test
      revision: a6ea5a8cab320b040a23452cc28066d9beae2cee
    metrics:
    - type: cos_sim_pearson
      value: 83.75973907678062
    - type: cos_sim_spearman
      value: 80.54994608351296
    - type: euclidean_pearson
      value: 80.58496551316748
    - type: euclidean_spearman
      value: 80.54993996457814
    - type: manhattan_pearson
      value: 80.49280884070782
    - type: manhattan_spearman
      value: 80.41230093993471
  - task:
      type: STS
    dataset:
      type: mteb/sts12-sts
      name: MTEB STS12
      config: default
      split: test
      revision: a0d554a64d88156834ff5ae9920b964011b16384
    metrics:
    - type: cos_sim_pearson
      value: 87.345503928209
    - type: cos_sim_spearman
      value: 80.4634619001261
    - type: euclidean_pearson
      value: 84.2666575030677
    - type: euclidean_spearman
      value: 80.46347579495351
    - type: manhattan_pearson
      value: 84.14370038922885
    - type: manhattan_spearman
      value: 80.36565043629274
  - task:
      type: STS
    dataset:
      type: mteb/sts13-sts
      name: MTEB STS13
      config: default
      split: test
      revision: 7e90230a92c190f1bf69ae9002b8cea547a64cca
    metrics:
    - type: cos_sim_pearson
      value: 75.14644787456163
    - type: cos_sim_spearman
      value: 75.88443166051762
    - type: euclidean_pearson
      value: 76.19117255044588
    - type: euclidean_spearman
      value: 75.88443166051762
    - type: manhattan_pearson
      value: 76.00450128624708
    - type: manhattan_spearman
      value: 75.69943934692938
  - task:
      type: STS
    dataset:
      type: mteb/sts14-sts
      name: MTEB STS14
      config: default
      split: test
      revision: 6031580fec1f6af667f0bd2da0a551cf4f0b2375
    metrics:
    - type: cos_sim_pearson
      value: 77.60763524019471
    - type: cos_sim_spearman
      value: 77.2591077818027
    - type: euclidean_pearson
      value: 77.14021401348042
    - type: euclidean_spearman
      value: 77.25911027186999
    - type: manhattan_pearson
      value: 76.87139081109731
    - type: manhattan_spearman
      value: 76.98379627773018
  - task:
      type: STS
    dataset:
      type: mteb/sts15-sts
      name: MTEB STS15
      config: default
      split: test
      revision: ae752c7c21bf194d8b67fd573edf7ae58183cbe3
    metrics:
    - type: cos_sim_pearson
      value: 88.18321035966198
    - type: cos_sim_spearman
      value: 89.0469892725742
    - type: euclidean_pearson
      value: 88.05085809092137
    - type: euclidean_spearman
      value: 89.04698194601134
    - type: manhattan_pearson
      value: 88.03620967628684
    - type: manhattan_spearman
      value: 89.02859425307943
  - task:
      type: STS
    dataset:
      type: mteb/sts16-sts
      name: MTEB STS16
      config: default
      split: test
      revision: 4d8694f8f0e0100860b497b999b3dbed754a0513
    metrics:
    - type: cos_sim_pearson
      value: 82.39166503459249
    - type: cos_sim_spearman
      value: 83.71826060604693
    - type: euclidean_pearson
      value: 82.70145770530107
    - type: euclidean_spearman
      value: 83.71826045549452
    - type: manhattan_pearson
      value: 82.56870669205291
    - type: manhattan_spearman
      value: 83.55353737670136
  - task:
      type: STS
    dataset:
      type: mteb/sts17-crosslingual-sts
      name: MTEB STS17 (en-en)
      config: en-en
      split: test
      revision: af5e6fb845001ecf41f4c1e033ce921939a2a68d
    metrics:
    - type: cos_sim_pearson
      value: 89.58290721169323
    - type: cos_sim_spearman
      value: 89.25956993522081
    - type: euclidean_pearson
      value: 89.4716703635447
    - type: euclidean_spearman
      value: 89.25956993522081
    - type: manhattan_pearson
      value: 89.4475864648432
    - type: manhattan_spearman
      value: 89.14694174575615
  - task:
      type: Reranking
    dataset:
      type: mteb/scidocs-reranking
      name: MTEB SciDocsRR
      config: default
      split: test
      revision: d3c5e1fc0b855ab6097bf1cda04dd73947d7caab
    metrics:
    - type: map
      value: 81.4879065181404
    - type: mrr
      value: 94.81295937178291
  - task:
      type: PairClassification
    dataset:
      type: mteb/sprintduplicatequestions-pairclassification
      name: MTEB SprintDuplicateQuestions
      config: default
      split: test
      revision: d66bd1f72af766a5cc4b0ca5e00c162f89e8cc46
    metrics:
    - type: cos_sim_accuracy
      value: 99.73960396039604
    - type: cos_sim_ap
      value: 92.70840767967965
    - type: cos_sim_f1
      value: 86.90890990542557
    - type: cos_sim_precision
      value: 86.5213082259663
    - type: cos_sim_recall
      value: 87.3
    - type: dot_accuracy
      value: 99.73960396039604
    - type: dot_ap
      value: 92.70828452993575
    - type: dot_f1
      value: 86.90890990542557
    - type: dot_precision
      value: 86.5213082259663
    - type: dot_recall
      value: 87.3
    - type: euclidean_accuracy
      value: 99.73960396039604
    - type: euclidean_ap
      value: 92.7084093403562
    - type: euclidean_f1
      value: 86.90890990542557
    - type: euclidean_precision
      value: 86.5213082259663
    - type: euclidean_recall
      value: 87.3
    - type: manhattan_accuracy
      value: 99.74059405940594
    - type: manhattan_ap
      value: 92.7406819850299
    - type: manhattan_f1
      value: 87.01234567901234
    - type: manhattan_precision
      value: 85.95121951219512
    - type: manhattan_recall
      value: 88.1
    - type: max_accuracy
      value: 99.74059405940594
    - type: max_ap
      value: 92.7406819850299
    - type: max_f1
      value: 87.01234567901234
  - task:
      type: Reranking
    dataset:
      type: mteb/stackoverflowdupquestions-reranking
      name: MTEB StackOverflowDupQuestions
      config: default
      split: test
      revision: e185fbe320c72810689fc5848eb6114e1ef5ec69
    metrics:
    - type: map
      value: 48.566931484512196
    - type: mrr
      value: 49.23111100500807
  - task:
      type: PairClassification
    dataset:
      type: mteb/twittersemeval2015-pairclassification
      name: MTEB TwitterSemEval2015
      config: default
      split: test
      revision: 70970daeab8776df92f5ea462b6173c0b46fd2d1
    metrics:
    - type: cos_sim_accuracy
      value: 86.27287357692079
    - type: cos_sim_ap
      value: 74.20855854505362
    - type: cos_sim_f1
      value: 69.09903201787044
    - type: cos_sim_precision
      value: 65.22961574507966
    - type: cos_sim_recall
      value: 73.45646437994723
    - type: dot_accuracy
      value: 86.27287357692079
    - type: dot_ap
      value: 74.20853189774614
    - type: dot_f1
      value: 69.09903201787044
    - type: dot_precision
      value: 65.22961574507966
    - type: dot_recall
      value: 73.45646437994723
    - type: euclidean_accuracy
      value: 86.27287357692079
    - type: euclidean_ap
      value: 74.20857455896677
    - type: euclidean_f1
      value: 69.09903201787044
    - type: euclidean_precision
      value: 65.22961574507966
    - type: euclidean_recall
      value: 73.45646437994723
    - type: manhattan_accuracy
      value: 86.2192287059665
    - type: manhattan_ap
      value: 74.0513280969461
    - type: manhattan_f1
      value: 69.13344473621389
    - type: manhattan_precision
      value: 63.12118570183086
    - type: manhattan_recall
      value: 76.41160949868075
    - type: max_accuracy
      value: 86.27287357692079
    - type: max_ap
      value: 74.20857455896677
    - type: max_f1
      value: 69.13344473621389
  - task:
      type: PairClassification
    dataset:
      type: mteb/twitterurlcorpus-pairclassification
      name: MTEB TwitterURLCorpus
      config: default
      split: test
      revision: 8b6510b0b1fa4e4c4f879467980e9be563ec1cdf
    metrics:
    - type: cos_sim_accuracy
      value: 89.16055419722902
    - type: cos_sim_ap
      value: 86.03614264194854
    - type: cos_sim_f1
      value: 78.89855695205357
    - type: cos_sim_precision
      value: 73.74656938215409
    - type: cos_sim_recall
      value: 84.82445334154605
    - type: dot_accuracy
      value: 89.16055419722902
    - type: dot_ap
      value: 86.03614225282097
    - type: dot_f1
      value: 78.89855695205357
    - type: dot_precision
      value: 73.74656938215409
    - type: dot_recall
      value: 84.82445334154605
    - type: euclidean_accuracy
      value: 89.16055419722902
    - type: euclidean_ap
      value: 86.0361548355667
    - type: euclidean_f1
      value: 78.89855695205357
    - type: euclidean_precision
      value: 73.74656938215409
    - type: euclidean_recall
      value: 84.82445334154605
    - type: manhattan_accuracy
      value: 89.11786393448985
    - type: manhattan_ap
      value: 86.00799361972808
    - type: manhattan_f1
      value: 78.84721152788472
    - type: manhattan_precision
      value: 75.26776338816941
    - type: manhattan_recall
      value: 82.78410840776101
    - type: max_accuracy
      value: 89.16055419722902
    - type: max_ap
      value: 86.0361548355667
    - type: max_f1
      value: 78.89855695205357
---

# E5-large-en-ru

## Model info

This is vocabulary pruned version of [intfloat/multilingual-e5-large](https://huggingface.co./intfloat/multilingual-e5-large).

Uses only russian and english tokens.

### Size

|  | intfloat/multilingual-e5-large | d0rj/e5-large-en-ru |
| --- | --- | --- |
| Model size (MB) | 2135.82 | 1394.8 |
| Params (count) | 559,890,946 | 365,638,14 |
| Word embeddings dim | 256,002,048 | 61,749,248 |

### Performance

Equal performance on SberQuAD dev benchmark.

| Metric on SberQuAD (4122 questions) | intfloat/multilingual-e5-large | d0rj/e5-large-en-ru |
| --- | --- | --- |
| recall@3 | 0.787239204269772 | **0.7882096069868996** |
| map@3 | 0.7230713245997101 | **0.723192624939351** |
| mrr@3 | 0.7241630276564784 | **0.7243651948892132** |
| recall@5 | 0.8277535177098496 | **0.8284813197476953** |
| map@5 | 0.7301603186155587 | **0.7302573588872716** |
| mrr@5 | 0.7334667637069385 | **0.7335718906679607** |
| recall@10 | **0.8716642406598738** | 0.871421639980592 |
| map@10 | **0.7314774917730316** | 0.7313000338687417 |
| mrr@10 | **0.7392223685527911** | 0.7391814537556898 |

## Usage

- Use **dot product** distance for retrieval.

- Use "query: " and "passage: " correspondingly for asymmetric tasks such as passage retrieval in open QA, ad-hoc information retrieval.

- Use "query: " prefix for symmetric tasks such as semantic similarity, bitext mining, paraphrase retrieval.

- Use "query: " prefix if you want to use embeddings as features, such as linear probing classification, clustering.

### transformers

#### Direct usage

```python
import torch.nn.functional as F
from torch import Tensor
from transformers import XLMRobertaTokenizer, XLMRobertaModel


def average_pool(last_hidden_states: Tensor, attention_mask: Tensor) -> Tensor:
    last_hidden = last_hidden_states.masked_fill(~attention_mask[..., None].bool(), 0.0)
    return last_hidden.sum(dim=1) / attention_mask.sum(dim=1)[..., None]


input_texts = [
  'query: How does a corporate website differ from a business card website?',
  'query: Где был создан первый троллейбус?',
  'passage: The first trolleybus was created in Germany by engineer Werner von Siemens, probably influenced by the idea of his brother, Dr. Wilhelm Siemens, who lived in England, expressed on May 18, 1881 at the twenty-second meeting of the Royal Scientific Society. The electrical circuit was carried out by an eight-wheeled cart (Kontaktwagen) rolling along two parallel contact wires. The wires were located quite close to each other, and in strong winds they often overlapped, which led to short circuits. An experimental trolleybus line with a length of 540 m (591 yards), opened by Siemens & Halske in the Berlin suburb of Halensee, operated from April 29 to June 13, 1882.',
  'passage: Корпоративный сайт — содержит полную информацию о компании-владельце, услугах/продукции, событиях в жизни компании. Отличается от сайта-визитки и представительского сайта полнотой представленной информации, зачастую содержит различные функциональные инструменты для работы с контентом (поиск и фильтры, календари событий, фотогалереи, корпоративные блоги, форумы). Может быть интегрирован с внутренними информационными системами компании-владельца (КИС, CRM, бухгалтерскими системами). Может содержать закрытые разделы для тех или иных групп пользователей — сотрудников, дилеров, контрагентов и пр.',
]

tokenizer = XLMRobertaTokenizer.from_pretrained('d0rj/e5-large-en-ru', use_cache=False)
model = XLMRobertaModel.from_pretrained('d0rj/e5-large-en-ru', use_cache=False)

batch_dict = tokenizer(input_texts, max_length=512, padding=True, truncation=True, return_tensors='pt')

outputs = model(**batch_dict)
embeddings = average_pool(outputs.last_hidden_state, batch_dict['attention_mask'])

embeddings = F.normalize(embeddings, p=2, dim=1)
scores = (embeddings[:2] @ embeddings[2:].T) * 100
print(scores.tolist())
# [[68.59542846679688, 81.75910949707031], [80.36100769042969, 64.77748107910156]]
```

#### Pipeline

```python
from transformers import pipeline


pipe = pipeline('feature-extraction', model='d0rj/e5-large-en-ru')
embeddings = pipe(input_texts, return_tensors=True)
embeddings[0].size()
# torch.Size([1, 17, 1024])
```

### sentence-transformers

```python
from sentence_transformers import SentenceTransformer


sentences = [
    'query: Что такое круглые тензоры?',
    'passage: Abstract: we introduce a novel method for compressing round tensors based on their inherent radial symmetry. We start by generalising PCA and eigen decomposition on round tensors...',
]

model = SentenceTransformer('d0rj/e5-large-en-ru')
embeddings = model.encode(sentences, convert_to_tensor=True)
embeddings.size()
# torch.Size([2, 1024])
```