Instructions to use along26/all-MiniLM-L6-v2_multilingual_malaysian with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use along26/all-MiniLM-L6-v2_multilingual_malaysian with sentence-transformers:

from sentence_transformers import SentenceTransformer

model = SentenceTransformer("along26/all-MiniLM-L6-v2_multilingual_malaysian")

sentences = [
"Consider the set A = {1/n : n ∈ ℕ}. Is the set A compact?",
"Mengapa masih terdapat kekurangan pilihan perumahan mampu milik untuk keluarga berpendapatan rendah di Malaysia, walaupun kerajaan berusaha?",
"The Malaysian government has implemented various measures to address the income gap between the rich and the poor, but there is still a noticeable disparity. According to the World Bank, the Gini coefficient in Malaysia was 0.394 in 2019, indicating a moderate level of income inequality, where a score of 0 represents perfect equality and a score of 1 represents perfect inequality. This is higher than the average for upper-middle-income countries.\n\nSeveral factors contribute to the widening income gap in Malaysia:\n\n1. Rapid urbanization: Urban areas have experienced faster economic growth compared to rural areas, resulting in higher income levels and better job opportunities in urban centers. This has led to a growing disparity between urban and rural populations.\n2. Education and skills: Higher-paying jobs often require advanced skills and education, and there is a significant difference in access to quality education between the rich and the poor. This contributes to income disparities as those with lower levels of education and skills are more likely to be employed in lower-paying jobs.\n3. Structural issues in the economy: Certain sectors, like manufacturing and construction, have a larger proportion of lower-skilled and lower-paid workers. The reliance on these sectors for employment opportunities could contribute to income disparities.\n4. Insufficient minimum wage: Although Malaysia has a minimum wage policy, it is not uniform across the country and has been criticized for being too low to provide a decent living for workers, particularly in high-cost areas.\n5. Labor market policies: The labor market in Malaysia is often criticized for being too favorable for employers, resulting in less bargaining power for workers and lower wages.\n6. Taxation policies: The taxation system in Malaysia may not be progressive enough, with insufficient tax revenue directed to support social welfare programs that would benefit lower-income households.\n7. Social welfare programs: While there are social welfare programs in place, they may not be extensive or well-targeted enough to effectively bridge the income gap between the rich and the poor.\n\nThe Malaysian government has implemented various initiatives aimed at reducing income inequality, such as:\n\n1. Minimum wage policies\n2. Programs to improve access to quality education, particularly for low-income households\n3. Encouraging foreign investment in less developed regions\n4. Developing new economic sectors with higher paying jobs\n5. Skills training programs for the workforce\n\nHowever, more can be done to ensure these policies are effectively implemented and have a more significant impact on reducing income inequality in Malaysia.",
"Pertimbangkan set A = {1/n : n ∈ ℕ}. Adakah set A padat?"
]
embeddings = model.encode(sentences)

similarities = model.similarity(embeddings, embeddings)
print(similarities.shape)
# [4, 4]

Notebooks
Google Colab
Kaggle

SentenceTransformer based on sentence-transformers/all-MiniLM-L6-v2

This is a sentence-transformers model finetuned from sentence-transformers/all-MiniLM-L6-v2. It maps sentences & paragraphs to a 384-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: sentence-transformers/all-MiniLM-L6-v2
Maximum Sequence Length: 512 tokens
Output Dimensionality: 384 dimensions
Similarity Function: Cosine Similarity

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, 'architecture': 'BertModel'})
  (1): Pooling({'word_embedding_dimension': 384, '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("along26/all-MiniLM-L6-v2_multilingual_malaysian")
# Run inference
sentences = [
    "What is the intensity of light transmitted through two polarizers with their axes at an angle of 45 degrees to each other, if the intensity of the incident light is 12 W/m² and the polarizer absorbs 50% of the light perpendicular to its axis? Use Malus' Law to solve the problem.",
    'Apakah keamatan cahaya yang dihantar melalui dua polarizer dengan paksinya pada sudut 45 darjah antara satu sama lain, jika keamatan cahaya kejadian ialah 12 W/m² dan polarizer menyerap 50% cahaya berserenjang dengan paksinya? Gunakan Hukum Malus untuk menyelesaikan masalah.',
    'What role did the opposition parties and civil society organizations play in exposing the 1MDB scandal and holding Najib Razak accountable?',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 384]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities)
# tensor([[ 1.0000, -0.7535,  0.9687],
#         [-0.7535,  1.0000, -0.7616],
#         [ 0.9687, -0.7616,  1.0000]])

Training Details

Training Dataset

Unnamed Dataset

Size: 210,285 training samples
Columns: sentence_0, sentence_1, and sentence_2

Approximate statistics based on the first 1000 samples:

	sentence_0	sentence_1	sentence_2
type	string	string	string
details	min: 11 tokens mean: 215.9 tokens max: 512 tokens	min: 14 tokens mean: 257.41 tokens max: 512 tokens	min: 15 tokens mean: 236.24 tokens max: 512 tokens

Samples:

sentence_0	sentence_1	sentence_2
`What are the four main functions of the human liver, and how is its unique anatomic structure suited to perform these functions?`	`Apakah empat fungsi utama hati manusia, dan bagaimanakah struktur anatomi uniknya sesuai untuk melaksanakan fungsi ini?`	`Why is the Malaysian government not doing enough to address the rising cost of living and income inequality?`
Changing the temperature affects the equilibrium constant (Kc) and the formation of the Fe(SCN)2+ complex ion from Fe3+ and SCN- ions according to Le Chatelier's principle. Le Chatelier's principle states that if a system at equilibrium is subjected to a change in temperature, pressure, or concentration of reactants or products, the system will adjust its position to counteract the change and re-establish equilibrium. In the case of the reaction between Fe3+ and SCN- ions to form the Fe(SCN)2+ complex ion, the balanced chemical equation is: Fe3+ (aq) + SCN- (aq) ⇌ Fe(SCN)2+ (aq) The equilibrium constant (Kc) for this reaction is given by: Kc = [Fe(SCN)2+] / ([Fe3+] [SCN-]) Now, let's consider the effect of temperature on this reaction. The reaction between Fe3+ and SCN- ions is an exothermic reaction, meaning it releases heat as it proceeds. According to Le Chatelier's principle, if the temperature of the system is increased, the equilibrium will shift in the direction that absorb...	Menukar suhu memberi kesan kepada pemalar keseimbangan (Kc) dan pembentukan ion kompleks Fe(SCN)2+ daripada ion Fe3+ dan SCN- mengikut prinsip Le Chatelier. Prinsip Le Chatelier menyatakan bahawa jika sistem pada keseimbangan tertakluk kepada perubahan suhu, tekanan, atau kepekatan bahan tindak balas atau produk, sistem akan menyesuaikan kedudukannya untuk mengatasi perubahan dan mewujudkan semula keseimbangan. Dalam kes tindak balas antara ion Fe3+ dan SCN- untuk membentuk ion kompleks Fe(SCN)2+, persamaan kimia yang seimbang ialah: Fe3+ (aq) + SCN- (aq) ⇌ Fe(SCN)2+ (aq) Pemalar keseimbangan (Kc) untuk tindak balas ini diberikan oleh: Kc = [Fe(SCN)2+] / ([Fe3+] [SCN-]) Sekarang, mari kita pertimbangkan kesan suhu pada tindak balas ini. Tindak balas antara ion Fe3+ dan SCN- ialah tindak balas eksotermik, bermakna ia membebaskan haba semasa ia berjalan. Mengikut prinsip Le Chatelier, jika suhu sistem dinaikkan, keseimbangan akan beralih ke arah yang menyerap haba, yang dalam kes in...	`Why does Malaysia have one of the highest income disparities in the world, with a significant portion of the population living in poverty despite being a middle-income country?`
The use of laws like the Official Secrets Act (OSA) and Sedition Act in Malaysia has been criticized for stifling free speech and discouraging whistleblowing in the country's anti-corruption efforts. The Official Secrets Act (OSA) is a law that dates back to the colonial era and is intended to protect national security and sensitive information. However, critics argue that the law is overly broad and has been used to suppress freedom of speech and silence whistleblowers who expose corruption and abuse of power. The law imposes strict penalties for the unauthorized disclosure of confidential information, which has created a chilling effect on whistleblowers and investigative journalists who might otherwise expose corruption. Similarly, the Sedition Act is a law that criminalizes speech that is deemed seditious, including speech that is likely to cause public disorder, insult the rulers, or question the legitimacy of the government. Critics argue that the law is overly broad and has be...	Penggunaan undang-undang seperti Akta Rahsia Rasmi (OSA) dan Akta Hasutan di Malaysia telah dikritik kerana menyekat kebebasan bersuara dan menghalang pemberi maklumat dalam usaha anti-rasuah negara. Akta Rahsia Rasmi (OSA) ialah undang-undang yang bermula sejak zaman penjajah dan bertujuan untuk melindungi keselamatan negara dan maklumat sensitif. Walau bagaimanapun, pengkritik berpendapat bahawa undang-undang itu terlalu luas dan telah digunakan untuk menyekat kebebasan bersuara dan menutup mulut pemberi maklumat yang mendedahkan rasuah dan penyalahgunaan kuasa. Undang-undang mengenakan penalti yang ketat untuk pendedahan maklumat sulit yang tidak dibenarkan, yang telah mewujudkan kesan menyeramkan kepada pemberi maklumat dan wartawan penyiasat yang mungkin mendedahkan rasuah. Begitu juga, Akta Hasutan ialah undang-undang yang menjenayahkan ucapan yang dianggap menghasut, termasuk ucapan yang berkemungkinan menyebabkan gangguan awam, menghina pemerintah, atau mempersoalkan kesahiha...	`How do the surface properties of metal catalysts influence the selectivity and activity of the oxidation reaction of hydrocarbons?`

Loss: TripletLoss with these parameters:

{
    "distance_metric": "TripletDistanceMetric.EUCLIDEAN",
    "triplet_margin": 5
}

Training Hyperparameters

Non-Default Hyperparameters

per_device_train_batch_size: 32
per_device_eval_batch_size: 32
num_train_epochs: 10
fp16: True
multi_dataset_batch_sampler: round_robin

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
eval_strategy: no
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
torch_empty_cache_steps: None
learning_rate: 5e-05
weight_decay: 0.0
adam_beta1: 0.9
adam_beta2: 0.999
adam_epsilon: 1e-08
max_grad_norm: 1
num_train_epochs: 10
max_steps: -1
lr_scheduler_type: linear
lr_scheduler_kwargs: {}
warmup_ratio: 0.0
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
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}
parallelism_config: None
deepspeed: None
label_smoothing_factor: 0.0
optim: adamw_torch_fused
optim_args: None
adafactor: False
group_by_length: False
length_column_name: length
project: huggingface
trackio_space_id: trackio
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: None
hub_always_push: False
hub_revision: None
gradient_checkpointing: False
gradient_checkpointing_kwargs: None
include_inputs_for_metrics: False
include_for_metrics: []
eval_do_concat_batches: True
fp16_backend: auto
push_to_hub_model_id: None
push_to_hub_organization: None
mp_parameters:
auto_find_batch_size: False
full_determinism: False
torchdynamo: None
ray_scope: last
ddp_timeout: 1800
torch_compile: False
torch_compile_backend: None
torch_compile_mode: None
include_tokens_per_second: False
include_num_input_tokens_seen: no
neftune_noise_alpha: None
optim_target_modules: None
batch_eval_metrics: False
eval_on_start: False
use_liger_kernel: False
liger_kernel_config: None
eval_use_gather_object: False
average_tokens_across_devices: True
prompts: None
batch_sampler: batch_sampler
multi_dataset_batch_sampler: round_robin
router_mapping: {}
learning_rate_mapping: {}

Training Logs

Click to expand

Epoch	Step	Training Loss
0.0761	500	4.9612
0.1522	1000	0.4336
0.2282	1500	0.0468
0.3043	2000	0.0045
0.3804	2500	0.0033
0.4565	3000	0.001
0.5326	3500	0.0012
0.6086	4000	0.0006
0.6847	4500	0.0002
0.7608	5000	0.0009
0.8369	5500	0.0009
0.9130	6000	0.0004
0.9890	6500	0.0007
1.0651	7000	0.0003
1.1412	7500	0.0011
1.2173	8000	0.0008
1.2934	8500	0.0
1.3694	9000	0.0003
1.4455	9500	0.0004
1.5216	10000	0.0
1.5977	10500	0.0003
1.6738	11000	0.0
1.7498	11500	0.0002
1.8259	12000	0.0003
1.9020	12500	0.0004
1.9781	13000	0.0005
2.0542	13500	0.0001
2.1302	14000	0.0
2.2063	14500	0.0
2.2824	15000	0.0
2.3585	15500	0.0004
2.4346	16000	0.0002
2.5107	16500	0.0003
2.5867	17000	0.001
2.6628	17500	0.0
2.7389	18000	0.0005
2.8150	18500	0.0003
2.8911	19000	0.0
2.9671	19500	0.0001
3.0432	20000	0.0
3.1193	20500	0.0
3.1954	21000	0.0003
3.2715	21500	0.0
3.3475	22000	0.0003
3.4236	22500	0.0
3.4997	23000	0.0
3.5758	23500	0.0
3.6519	24000	0.0
3.7279	24500	0.0
3.8040	25000	0.0003
3.8801	25500	0.0003
3.9562	26000	0.0
4.0323	26500	0.0
4.1083	27000	0.0
4.1844	27500	0.0
4.2605	28000	0.0002
4.3366	28500	0.0
4.4127	29000	0.0
4.4887	29500	0.0003
4.5648	30000	0.0
4.6409	30500	0.0003
4.7170	31000	0.0
4.7931	31500	0.0
4.8691	32000	0.0005
4.9452	32500	0.0
5.0213	33000	0.0
5.0974	33500	0.0
5.1735	34000	0.0
5.2495	34500	0.0003
5.3256	35000	0.0
5.4017	35500	0.0
5.4778	36000	0.0
5.5539	36500	0.0
5.6299	37000	0.0
5.7060	37500	0.0
5.7821	38000	0.0001
5.8582	38500	0.0009
5.9343	39000	0.0
6.0103	39500	0.0
6.0864	40000	0.0
6.1625	40500	0.0
6.2386	41000	0.0004
6.3147	41500	0.0
6.3907	42000	0.0
6.4668	42500	0.0
6.5429	43000	0.0
6.6190	43500	0.0003
6.6951	44000	0.0
6.7712	44500	0.0
6.8472	45000	0.0003
6.9233	45500	0.0
6.9994	46000	0.0
7.0755	46500	0.0
7.1516	47000	0.0
7.2276	47500	0.0
7.3037	48000	0.0003
7.3798	48500	0.0
7.4559	49000	0.0003
7.5320	49500	0.0
7.6080	50000	0.0003
7.6841	50500	0.0
7.7602	51000	0.0
7.8363	51500	0.0
7.9124	52000	0.0
7.9884	52500	0.0
8.0645	53000	0.0
8.1406	53500	0.0003
8.2167	54000	0.0
8.2928	54500	0.0
8.3688	55000	0.0
8.4449	55500	0.0
8.5210	56000	0.0
8.5971	56500	0.0
8.6732	57000	0.0
8.7492	57500	0.0003
8.8253	58000	0.0003
8.9014	58500	0.0
8.9775	59000	0.0
9.0536	59500	0.0
9.1296	60000	0.0
9.2057	60500	0.0
9.2818	61000	0.0
9.3579	61500	0.0
9.4340	62000	0.0
9.5100	62500	0.0
9.5861	63000	0.0006
9.6622	63500	0.0
9.7383	64000	0.0
9.8144	64500	0.0
9.8904	65000	0.0
9.9665	65500	0.0006

Framework Versions

Python: 3.12.12
Sentence Transformers: 5.1.2
Transformers: 4.57.1
PyTorch: 2.8.0+cu126
Accelerate: 1.11.0
Datasets: 4.0.0
Tokenizers: 0.22.1

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",
}

TripletLoss

@misc{hermans2017defense,
    title={In Defense of the Triplet Loss for Person Re-Identification},
    author={Alexander Hermans and Lucas Beyer and Bastian Leibe},
    year={2017},
    eprint={1703.07737},
    archivePrefix={arXiv},
    primaryClass={cs.CV}
}