Update README.md

2a0d263 verified over 1 year ago

21.1 kB

base_model: intfloat/multilingual-e5-large-instruct
language:
  - en
  - ne
library_name: sentence-transformers
pipeline_tag: sentence-similarity
tags:
  - sentence-transformers
  - sentence-similarity
  - feature-extraction
  - generated_from_trainer
  - dataset_size:45199
  - loss:MultipleNegativesRankingLoss
widget:
  - source_sentence: प्रधानमन्त्री नरेन्द्र मोदी सरकारका असफलताहरू के के हुन्?
    sentences:
      - >-
        पूर्वोत्तर राज्यहरूका मुख्य समस्याहरू के के हुन् र तिनीहरूको केन्द्रीय
        सरकारसँग असन्तोष के हो?
      - पूर्णांक के हो?
      - नरेन्द्र मोदी सरकारले कुन क्षेत्रमा असफल भएको छ?
  - source_sentence: >-
      मैले विचार गर्नुपर्ने कलेजहरू के के हुन्, विचार गर्नुपर्ने कारकहरू: केएमसी
      म्यानिपल वा केएमसी मंगोलमा?
    sentences:
      - मंगलोर शान्त वा हिंस्रक स्थान हो?
      - पुरुषहरूको तुलनामा महिलाहरूको लागि यौनिक आनन्द बढी हुन्छ कि हुँदैन?
      - के कसैले केएमसी मानिपाल र मंगलोरको संक्षिप्त तुलना गर्न सक्छ?
  - source_sentence: म कसरी मेरो अङ्ग्रेजी भाषा सुधार गर्न सक्छु?
    sentences:
      - म कसरी एक नेचुरल अंग्रेजी वक्ता बन्न सक्छु?
      - >-
        म जहाँ कुनै मूल अंग्रेजी वक्ताहरू छन् जो मेरो साथ मित्र बन्न चाहन्छन् र
        मलाई मद्दत गर्न चाहन्छन्?
      - ने टी २०१ 6 को लागि निजी कलेजहरूको लागि एमबीबीएसको लागि के कटअफ हुनेछ?
  - source_sentence: समय यात्रा सम्भव छ कि छैन? यदि छ भने, कसरी?
    sentences:
      - अन्धकारमय वेब सुरक्षित छ कि छैन ब्राउज गर्न?
      - >-
        यदि कुनै बितेको समय राम्रो थियो र समयको यात्रा सम्भव थियो भने म किन
        वर्तमान समयमा बाँचिरहेको छु?
      - भविष्यमा समय यात्रा सम्भव हुनेछ कि छैन?
  - source_sentence: म कसरी बिस्तारै तौल घटाउन सक्छु?
    sentences:
      - कसरी कुनै केटाले त्यो केटीसँग बदला लिन सक्छ जसले उसलाई धोका दिएको छ?
      - कस्तो प्रकारको आहार कसैले आहार नचाहने व्यक्तिका लागि उत्तम हुन्छ?
      - वजन घटाउनको लागि कुनै राम्रो आहार हो?
license: apache-2.0

SentenceTransformer based on intfloat/multilingual-e5-large-instruct

This is a sentence-transformers model finetuned from intfloat/multilingual-e5-large-instruct on the universalml0/nepali_embedding_dataset dataset. It maps sentences & paragraphs to a 1024-dimensional dense vector space and can be used for semantic textual similarity, semantic search, paraphrase mining, text classification, clustering, and more.

Model Details

Model Description

Model Type: Sentence Transformer
Base model: intfloat/multilingual-e5-large-instruct
Maximum Sequence Length: 512 tokens
Output Dimensionality: 1024 tokens
Similarity Function: Cosine Similarity
Training Dataset:
- universalml0/nepali_embedding_dataset

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: XLMRobertaModel 
  (1): Pooling({'word_embedding_dimension': 1024, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False, 'include_prompt': True})
  (2): Normalize()
)

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("universalml0/finetuned_embedding_model_e5-large-multilingual-large")
# Run inference
sentences = [
    'म कसरी बिस्तारै तौल घटाउन सक्छु?',
    'वजन घटाउनको लागि कुनै राम्रो आहार हो?',
    'कस्तो प्रकारको आहार कसैले आहार नचाहने व्यक्तिका लागि उत्तम हुन्छ?',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 1024]

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

Training Details

Training Dataset

universalml0/nepali_embedding_dataset

Dataset: universalml0/nepali_embedding_dataset
Size: 45,199 training samples
Columns: anchor, positive, and negative

Approximate statistics based on the first 1000 samples:

	anchor	positive	negative
type	string	string	string
details	min: 7 tokens mean: 17.53 tokens max: 486 tokens	min: 6 tokens mean: 17.68 tokens max: 512 tokens	min: 6 tokens mean: 18.9 tokens max: 156 tokens

Samples:

anchor	positive	negative
`भारतीय सरकारले ५०० र १००० रुपयाको नोटमाथि प्रतिबन्ध लगाउनुको कारण के थियो?`	`भारतीय सरकारले ५०० र १००० को नोटलाई निष्क्रिय पारेको छ तर तिनीहरूलाई ५०० र २००० को नोटहरूसँग प्रतिस्थापन गरेको छ। के यो विरोधाभासी छैन?`	`भारतीय सरकारले किन चाहेको भए सीमित मात्रामा नोटहरू मुद्रण गर्न र बजेट घाटा क्लियर गर्न सक्दैन? विशेष गरी, किन कुनै पनि देशले यो गर्न सक्दैन?`
`भारतीय हुनुको अनुभूति कस्तो हुन्छ?`	`भारतीय हुनुको अनुभूति कस्तो हुन्छ?`	`भारतीय महिला हुनुको अनुभव कस्तो हुन्छ?`
`के कुनै व्यक्तिले edWisor मार्फत कुनै नौकरी पाएको छ?`	`एडवाइजर वैध छ र के कसैले यस मार्फत कुनै नौकरी पाएको छ?`	`एलिटमसको माध्यमबाट कसैले काम पाएको छ?`

Loss: MultipleNegativesRankingLoss with these parameters:

{
    "scale": 20.0,
    "similarity_fct": "cos_sim"
}

Training Hyperparameters

Non-Default Hyperparameters

per_device_train_batch_size: 4
learning_rate: 1e-06
num_train_epochs: 1
warmup_ratio: 0.3
bf16: True
batch_sampler: no_duplicates

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
eval_strategy: no
prediction_loss_only: True
per_device_train_batch_size: 4
per_device_eval_batch_size: 8
per_gpu_train_batch_size: None
per_gpu_eval_batch_size: None
gradient_accumulation_steps: 1
eval_accumulation_steps: None
torch_empty_cache_steps: None
learning_rate: 1e-06
weight_decay: 0.0
adam_beta1: 0.9
adam_beta2: 0.999
adam_epsilon: 1e-08
max_grad_norm: 1.0
num_train_epochs: 1
max_steps: -1
lr_scheduler_type: linear
lr_scheduler_kwargs: {}
warmup_ratio: 0.3
warmup_steps: 0
log_level: passive
log_level_replica: warning
log_on_each_node: True
logging_nan_inf_filter: True
save_safetensors: True
save_on_each_node: False
save_only_model: False
restore_callback_states_from_checkpoint: False
no_cuda: False
use_cpu: False
use_mps_device: False
seed: 42
data_seed: None
jit_mode_eval: False
use_ipex: False
bf16: True
fp16: False
fp16_opt_level: O1
half_precision_backend: auto
bf16_full_eval: False
fp16_full_eval: False
tf32: 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: False
resume_from_checkpoint: None
hub_model_id: None
hub_strategy: every_save
hub_private_repo: False
hub_always_push: False
gradient_checkpointing: False
gradient_checkpointing_kwargs: None
include_inputs_for_metrics: False
eval_do_concat_batches: True
fp16_backend: auto
push_to_hub_model_id: None
push_to_hub_organization: None
mp_parameters:
auto_find_batch_size: False
full_determinism: False
torchdynamo: None
ray_scope: last
ddp_timeout: 1800
torch_compile: False
torch_compile_backend: None
torch_compile_mode: None
dispatch_batches: None
split_batches: None
include_tokens_per_second: False
include_num_input_tokens_seen: False
neftune_noise_alpha: None
optim_target_modules: None
batch_eval_metrics: False
eval_on_start: False
eval_use_gather_object: False
batch_sampler: no_duplicates
multi_dataset_batch_sampler: proportional

Training Logs

Click to expand

Epoch	Step	Training Loss
0.0088	100	0.8671
0.0177	200	0.8234
0.0265	300	0.8223
0.0354	400	0.7423
0.0442	500	0.6605
0.0531	600	0.5558
0.0619	700	0.4076
0.0708	800	0.3617
0.0796	900	0.3087
0.0885	1000	0.2747
0.0973	1100	0.2409
0.1062	1200	0.229
0.1150	1300	0.209
0.1239	1400	0.2556
0.1327	1500	0.2536
0.1416	1600	0.2092
0.1504	1700	0.2464
0.1593	1800	0.1727
0.1681	1900	0.281
0.1770	2000	0.2289
0.1858	2100	0.2065
0.1947	2200	0.1751
0.2035	2300	0.231
0.2124	2400	0.2127
0.2212	2500	0.1908
0.2301	2600	0.2131
0.2389	2700	0.1704
0.2478	2800	0.1923
0.2566	2900	0.1635
0.2655	3000	0.2061
0.2743	3100	0.1843
0.2832	3200	0.1443
0.2920	3300	0.1513
0.3009	3400	0.1879
0.3097	3500	0.2372
0.3186	3600	0.1542
0.3274	3700	0.2523
0.3363	3800	0.2055
0.3451	3900	0.1474
0.3540	4000	0.1647
0.3628	4100	0.1615
0.3717	4200	0.1271
0.3805	4300	0.1451
0.3894	4400	0.1887
0.3982	4500	0.1334
0.4071	4600	0.1962
0.4159	4700	0.1695
0.4248	4800	0.1561
0.4336	4900	0.1146
0.4425	5000	0.1381
0.4513	5100	0.1452
0.4602	5200	0.2388
0.4690	5300	0.1951
0.4779	5400	0.1142
0.4867	5500	0.182
0.4956	5600	0.1968
0.5044	5700	0.1744
0.5133	5800	0.1868
0.5221	5900	0.1452
0.5310	6000	0.1345
0.5398	6100	0.1318
0.5487	6200	0.218
0.5575	6300	0.2118
0.5664	6400	0.1972
0.5752	6500	0.0935
0.5841	6600	0.1991
0.5929	6700	0.1252
0.6018	6800	0.1128
0.6106	6900	0.1585
0.6195	7000	0.2293
0.6283	7100	0.2104
0.6372	7200	0.1416
0.6460	7300	0.2004
0.6549	7400	0.1446
0.6637	7500	0.1171
0.6726	7600	0.1386
0.6814	7700	0.1291
0.6903	7800	0.1546
0.6991	7900	0.1484
0.7080	8000	0.129
0.7168	8100	0.1873
0.7257	8200	0.1333
0.7345	8300	0.1713
0.7434	8400	0.1016
0.7522	8500	0.1519
0.7611	8600	0.1851
0.7699	8700	0.144
0.7788	8800	0.1488
0.7876	8900	0.1568
0.7965	9000	0.1672
0.8053	9100	0.1236
0.8142	9200	0.0973
0.8230	9300	0.1491
0.8319	9400	0.2251
0.8407	9500	0.1433
0.8496	9600	0.2634
0.8584	9700	0.1723
0.8673	9800	0.2373
0.8761	9900	0.1065
0.8850	10000	0.1578
0.8938	10100	0.1127
0.9027	10200	0.1632
0.9115	10300	0.19
0.9204	10400	0.0958
0.9292	10500	0.1029
0.9381	10600	0.1183
0.9469	10700	0.1779
0.9558	10800	0.1571
0.9646	10900	0.1666
0.9735	11000	0.1405
0.9823	11100	0.147
0.9912	11200	0.1428
1.0	11300	0.1724

Framework Versions

Python: 3.9.5
Sentence Transformers: 3.0.1
Transformers: 4.44.2
PyTorch: 2.3.0+cu121
Accelerate: 0.33.0
Datasets: 2.21.0
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",
}

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}
}