metadata

base_model: indobenchmark/indobert-base-p2
datasets:
  - afaji/indonli
language:
  - id
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
pipeline_tag: sentence-similarity
tags:
  - sentence-transformers
  - sentence-similarity
  - feature-extraction
  - generated_from_trainer
  - dataset_size:6915
  - loss:MultipleNegativesRankingLoss
widget:
  - source_sentence: >-
      Pesta Olahraga Asia Tenggara atau Southeast Asian Games, biasa disingkat
      SEA Games, adalah ajang olahraga yang diadakan setiap dua tahun dan
      melibatkan 11 negara Asia Tenggara.
    sentences:
      - Sekarang tahun 2017.
      - >-
        Warna kulit tidak mempengaruhi waktu berjemur yang baik untuk
        mengatifkan pro-vitamin D3.
      - Pesta Olahraga Asia Tenggara diadakan setiap tahun.
  - source_sentence: Menjalani aktivitas Ramadhan di tengah wabah Corona tentunya tidak mudah.
    sentences:
      - Tidak ada observasi yang pernah dilansir oleh Business Insider.
      - Wabah Corona membuat aktivitas Ramadhan tidak mudah dijalani.
      - Piala Sudirman pertama digelar pada tahun 1989.
  - source_sentence: >-
      Dalam bidang politik, partai ini memperjuangkan agar kekuasaan sepenuhnya
      berada di tangan rakyat.
    sentences:
      - Galileo tidak berhasil mengetes hasil dari Hukum Inert.
      - Kudeta 14 Februari 1946 gagal merebut kekuasaan Belanda.
      - Partai ini berusaha agar kekuasaan sepenuhnya berada di tangan rakyat.
  - source_sentence: >-
      Keluarga mendiang Prince menuduh layanan musik streaming Tidal memasukkan
      karya milik sang penyanyi legendaris tanpa izin .
    sentences:
      - Rosier adalah pelayan setia Lord Voldemort.
      - Bangunan ini digunakan untuk penjualan.
      - >-
        Keluarga mendiang Prince sudah memberi izin kepada TImbal untuk
        menggunakan lagu milik Prince.
  - source_sentence: >-
      Tujuan dari acara dengar pendapat CRTC adalah untuk mengumpulkan respons
      dari pada pemangku kepentingan industri ini dan dari masyarakat umum.
    sentences:
      - Pembuat Rooms hanya bisa membuat meeting yang terbuka.
      - >-
        Masyarakat umum dilibatkan untuk memberikan respon dalam acara dengar
        pendapat CRTC.
      - Eminem dirasa tidak akan memulai kembali kariernya tahun ini.
model-index:
  - name: SentenceTransformer based on indobenchmark/indobert-base-p2
    results:
      - task:
          type: semantic-similarity
          name: Semantic Similarity
        dataset:
          name: IndoNLI dev
          type: IndoNLI-dev
        metrics:
          - type: pearson_cosine
            value: 0.054645724410651776
            name: Pearson Cosine
          - type: spearman_cosine
            value: 0.05813131922360566
            name: Spearman Cosine
          - type: pearson_manhattan
            value: 0.06440629731537877
            name: Pearson Manhattan
          - type: spearman_manhattan
            value: 0.06617214306439209
            name: Spearman Manhattan
          - type: pearson_euclidean
            value: 0.06472911547924179
            name: Pearson Euclidean
          - type: spearman_euclidean
            value: 0.06670189814323607
            name: Spearman Euclidean
          - type: pearson_dot
            value: 0.02146795646141896
            name: Pearson Dot
          - type: spearman_dot
            value: 0.014015602655765296
            name: Spearman Dot
          - type: pearson_max
            value: 0.06472911547924179
            name: Pearson Max
          - type: spearman_max
            value: 0.06670189814323607
            name: Spearman Max
      - task:
          type: semantic-similarity
          name: Semantic Similarity
        dataset:
          name: IndoNLI test
          type: IndoNLI-test
        metrics:
          - type: pearson_cosine
            value: -0.027420454797600895
            name: Pearson Cosine
          - type: spearman_cosine
            value: -0.03327545125556324
            name: Spearman Cosine
          - type: pearson_manhattan
            value: -0.04660713875385687
            name: Pearson Manhattan
          - type: spearman_manhattan
            value: -0.0317801498705458
            name: Spearman Manhattan
          - type: pearson_euclidean
            value: -0.04697128223611728
            name: Pearson Euclidean
          - type: spearman_euclidean
            value: -0.03186507233227842
            name: Spearman Euclidean
          - type: pearson_dot
            value: -0.014150904875791395
            name: Pearson Dot
          - type: spearman_dot
            value: -0.01615774720436149
            name: Spearman Dot
          - type: pearson_max
            value: -0.014150904875791395
            name: Pearson Max
          - type: spearman_max
            value: -0.01615774720436149
            name: Spearman Max

SentenceTransformer based on indobenchmark/indobert-base-p2

This is a sentence-transformers model finetuned from indobenchmark/indobert-base-p2 on the afaji/indonli dataset. 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: indobenchmark/indobert-base-p2
Maximum Sequence Length: 512 tokens
Output Dimensionality: 768 tokens
Similarity Function: Cosine Similarity
Training Dataset:
- afaji/indonli
Language: id

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: BertModel 
  (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("cassador/3bs4lr2")
# Run inference
sentences = [
    'Tujuan dari acara dengar pendapat CRTC adalah untuk mengumpulkan respons dari pada pemangku kepentingan industri ini dan dari masyarakat umum.',
    'Masyarakat umum dilibatkan untuk memberikan respon dalam acara dengar pendapat CRTC.',
    'Pembuat Rooms hanya bisa membuat meeting yang terbuka.',
]
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: IndoNLI-dev
Evaluated with EmbeddingSimilarityEvaluator

Metric	Value
pearson_cosine	0.0546
spearman_cosine	0.0581
pearson_manhattan	0.0644
spearman_manhattan	0.0662
pearson_euclidean	0.0647
spearman_euclidean	0.0667
pearson_dot	0.0215
spearman_dot	0.014
pearson_max	0.0647
spearman_max	0.0667

Semantic Similarity

Dataset: IndoNLI-test
Evaluated with EmbeddingSimilarityEvaluator

Metric	Value
pearson_cosine	-0.0274
spearman_cosine	-0.0333
pearson_manhattan	-0.0466
spearman_manhattan	-0.0318
pearson_euclidean	-0.047
spearman_euclidean	-0.0319
pearson_dot	-0.0142
spearman_dot	-0.0162
pearson_max	-0.0142
spearman_max	-0.0162

Training Details

Training Dataset

afaji/indonli

Dataset: afaji/indonli
Size: 6,915 training samples
Columns: premise, hypothesis, and label
Approximate statistics based on the first 1000 samples:
premise hypothesis label
type string string int
details
min: 12 tokens
mean: 29.26 tokens
max: 135 tokens

min: 6 tokens
mean: 12.13 tokens
max: 36 tokens

0: ~51.00%
1: ~49.00%

	premise	hypothesis	label
type	string	string	int
details	min: 12 tokens mean: 29.26 tokens max: 135 tokens	min: 6 tokens mean: 12.13 tokens max: 36 tokens	0: ~51.00% 1: ~49.00%

Samples:

premise	hypothesis	label
`Presiden Joko Widodo (Jokowi) menyampaikan prediksi bahwa wabah virus Corona (COVID-19) di Indonesia akan selesai akhir tahun ini.`	`Prediksi akhir wabah tidak disampaikan Jokowi.`	`0`
`Meski biasanya hanya digunakan di fasilitas kesehatan, saat ini masker dan sarung tangan sekali pakai banyak dipakai di tingkat rumah tangga.`	`Masker sekali pakai banyak dipakai di tingkat rumah tangga.`	`1`
`Seperti namanya, paket internet sahur Telkomsel ini ditujukan bagi pengguna yang menginginkan kuota ekstra, untuk menemani momen sahur sepanjang bulan puasa.`	`Paket internet sahur tidak ditujukan untuk saat sahur.`	`0`

Loss: MultipleNegativesRankingLoss with these parameters:

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

Evaluation Dataset

afaji/indonli

Dataset: afaji/indonli
Size: 1,556 evaluation samples
Columns: premise, hypothesis, and label
Approximate statistics based on the first 1000 samples:
premise hypothesis label
type string string int
details
min: 9 tokens
mean: 28.07 tokens
max: 179 tokens

min: 6 tokens
mean: 12.15 tokens
max: 25 tokens

0: ~47.90%
1: ~52.10%

	premise	hypothesis	label
type	string	string	int
details	min: 9 tokens mean: 28.07 tokens max: 179 tokens	min: 6 tokens mean: 12.15 tokens max: 25 tokens	0: ~47.90% 1: ~52.10%

Samples:

premise	hypothesis	label
`Manuskrip tersebut berisi tiga catatan yang menceritakan bagaimana peristiwa jatuhnya meteorit serta laporan kematian akibat kejadian tersebut seperti dilansir dari Science Alert, Sabtu (25/4/2020).`	`Manuskrip tersebut tidak mencatat laporan kematian.`	`0`
`Dilansir dari Business Insider, menurut observasi dari Mauna Loa Observatory di Hawaii pada karbon dioksida (CO2) di level mencapai 410 ppm tidak langsung memberikan efek pada pernapasan, karena tubuh manusia juga masih membutuhkan CO2 dalam kadar tertentu.`	`Tidak ada observasi yang pernah dilansir oleh Business Insider.`	`0`
`Seorang wanita asal New York mengaku sangat benci air putih.`	`Tidak ada orang dari New York yang membenci air putih.`	`0`

Loss: MultipleNegativesRankingLoss with these parameters:

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

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: epoch
per_device_train_batch_size: 4
per_device_eval_batch_size: 4
learning_rate: 2e-05
warmup_ratio: 0.1
fp16: True

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
eval_strategy: epoch
prediction_loss_only: True
per_device_train_batch_size: 4
per_device_eval_batch_size: 4
per_gpu_train_batch_size: None
per_gpu_eval_batch_size: None
gradient_accumulation_steps: 1
eval_accumulation_steps: None
learning_rate: 2e-05
weight_decay: 0.0
adam_beta1: 0.9
adam_beta2: 0.999
adam_epsilon: 1e-08
max_grad_norm: 1.0
num_train_epochs: 3
max_steps: -1
lr_scheduler_type: linear
lr_scheduler_kwargs: {}
warmup_ratio: 0.1
warmup_steps: 0
log_level: passive
log_level_replica: warning
log_on_each_node: True
logging_nan_inf_filter: True
save_safetensors: True
save_on_each_node: False
save_only_model: False
restore_callback_states_from_checkpoint: False
no_cuda: False
use_cpu: False
use_mps_device: False
seed: 42
data_seed: None
jit_mode_eval: False
use_ipex: False
bf16: 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: 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
batch_sampler: batch_sampler
multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	Training Loss	loss	IndoNLI-dev_spearman_cosine	IndoNLI-test_spearman_cosine
0	0	-	-	0.1277	-
0.0578	100	0.0488	-	-	-
0.1157	200	0.0403	-	-	-
0.1735	300	0.0173	-	-	-
0.2313	400	0.0052	-	-	-
0.2892	500	0.0077	-	-	-
0.3470	600	0.0065	-	-	-
0.4049	700	0.0199	-	-	-
0.4627	800	0.0318	-	-	-
0.5205	900	0.019	-	-	-
0.5784	1000	0.0128	-	-	-
0.6362	1100	0.0124	-	-	-
0.6940	1200	0.0224	-	-	-
0.7519	1300	0.0115	-	-	-
0.8097	1400	0.0082	-	-	-
0.8676	1500	0.0132	-	-	-
0.9254	1600	0.0225	-	-	-
0.9832	1700	0.0133	-	-	-
1.0	1729	-	0.0173	0.0465	-
1.0411	1800	0.0056	-	-	-
1.0989	1900	0.0027	-	-	-
1.1567	2000	0.0109	-	-	-
1.2146	2100	0.0021	-	-	-
1.2724	2200	0.0004	-	-	-
1.3302	2300	0.0082	-	-	-
1.3881	2400	0.001	-	-	-
1.4459	2500	0.0009	-	-	-
1.5038	2600	0.0021	-	-	-
1.5616	2700	0.0032	-	-	-
1.6194	2800	0.0061	-	-	-
1.6773	2900	0.0057	-	-	-
1.7351	3000	0.0127	-	-	-
1.7929	3100	0.0018	-	-	-
1.8508	3200	0.0007	-	-	-
1.9086	3300	0.0078	-	-	-
1.9665	3400	0.0017	-	-	-
2.0	3458	-	0.0078	0.0446	-
2.0243	3500	0.0003	-	-	-
2.0821	3600	0.0042	-	-	-
2.1400	3700	0.0005	-	-	-
2.1978	3800	0.0002	-	-	-
2.2556	3900	0.0006	-	-	-
2.3135	4000	0.0003	-	-	-
2.3713	4100	0.0048	-	-	-
2.4291	4200	0.0002	-	-	-
2.4870	4300	0.0043	-	-	-
2.5448	4400	0.0011	-	-	-
2.6027	4500	0.0005	-	-	-
2.6605	4600	0.0009	-	-	-
2.7183	4700	0.0013	-	-	-
2.7762	4800	0.0018	-	-	-
2.8340	4900	0.0004	-	-	-
2.8918	5000	0.0014	-	-	-
2.9497	5100	0.0045	-	-	-
3.0	5187	-	0.0083	0.0581	-0.0333

Framework Versions

Python: 3.10.12
Sentence Transformers: 3.0.1
Transformers: 4.41.2
PyTorch: 2.3.0+cu121
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",
}

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}
}