syubraj/sentence_similarity_nepali_v2

SentenceTransformer based on Rajan/NepaliBERT

This is a sentence-transformers model finetuned from Rajan/NepaliBERT. 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: Rajan/NepaliBERT
• Maximum Sequence Length: 512 tokens
• Output Dimensionality: 768 tokens
• 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}) 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("syubraj/sentence_similarity_nepali_v2")
# Run inference
sentences = [
    'रातो, डबल डेकर बस।',
    'रातो डबल डेकर बस।',
    'दुई कालो कुकुर हिउँमा हिंड्दै।',
]
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: stsb-dev-nepali
• Evaluated with EmbeddingSimilarityEvaluator

Metric	Value
pearson_cosine	0.6971
spearman_cosine	0.6623
pearson_manhattan	0.6332
spearman_manhattan	0.6079
pearson_euclidean	0.634
spearman_euclidean	0.609
pearson_dot	0.4848
spearman_dot	0.5306
pearson_max	0.6971
spearman_max	0.6623

Training Details

Training Dataset

Unnamed Dataset

• Size: 4,599 training samples
• Columns: sentence_0, sentence_1, and label
• Approximate statistics based on the first 1000 samples:

  	sentence_0	sentence_1	label
  type	string	string	float
  details	min: 6 tokens mean: 19.5 tokens max: 81 tokens	min: 6 tokens mean: 19.43 tokens max: 75 tokens	min: 0.0 mean: 0.54 max: 1.0

• Samples:

  sentence_0	sentence_1	label
  एक व्यक्ति प्याज काट्दै छ।	एउटा बिरालो शौचालयमा पपिङ गर्दैछ।	0.0
  क्यानडाको तेल रेल विस्फोटमा थप मृत्यु हुने अपेक्षा गरिएको छ	क्यानडामा रेल दुर्घटनामा पाँच जनाको मृत्यु भएको छ	0.5599999904632569
  एउटी महिला झिंगा माझ्दै छिन्।	एउटी महिला केही झिंगा माझ्दै।	1.0

• Loss: CosineSimilarityLoss with these parameters:

  {
      "loss_fct": "torch.nn.modules.loss.MSELoss"
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• per_device_train_batch_size: 16
• per_device_eval_batch_size: 16
• num_train_epochs: 100
• multi_dataset_batch_sampler: round_robin

All Hyperparameters

Click to expand

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: steps
• prediction_loss_only: True
• per_device_train_batch_size: 16
• per_device_eval_batch_size: 16
• per_gpu_train_batch_size: None
• per_gpu_eval_batch_size: None
• gradient_accumulation_steps: 1
• eval_accumulation_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: 100
• 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
• use_ipex: False
• bf16: False
• 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
• batch_sampler: batch_sampler
• multi_dataset_batch_sampler: round_robin

Training Logs

Click to expand

Epoch	Step	Training Loss	stsb-dev-nepali_spearman_max
1.0	288	-	0.5355
1.7361	500	0.0723	-
2.0	576	-	0.5794
3.0	864	-	0.6108
3.4722	1000	0.047	0.6147
4.0	1152	-	0.6259
5.0	1440	-	0.6356
5.2083	1500	0.034	-
6.0	1728	-	0.6329
6.9444	2000	0.0217	0.6375
7.0	2016	-	0.6382
8.0	2304	-	0.6468
8.6806	2500	0.0137	-
9.0	2592	-	0.6348
10.0	2880	-	0.6332
10.4167	3000	0.0102	0.6427
11.0	3168	-	0.6370
12.0	3456	-	0.6515
12.1528	3500	0.0084	-
13.0	3744	-	0.6546
13.8889	4000	0.0069	0.6400
14.0	4032	-	0.6610
15.0	4320	-	0.6495
15.625	4500	0.006	-
16.0	4608	-	0.6574
17.0	4896	-	0.6486
17.3611	5000	0.0053	0.6589
18.0	5184	-	0.6592
19.0	5472	-	0.6488
19.0972	5500	0.0047	-
20.0	5760	-	0.6436
20.8333	6000	0.0044	0.6576
21.0	6048	-	0.6515
22.0	6336	-	0.6541
22.5694	6500	0.0041	-
23.0	6624	-	0.6549
24.0	6912	-	0.6571
24.3056	7000	0.0037	0.6603
25.0	7200	-	0.6699
26.0	7488	-	0.6653
26.0417	7500	0.0037	-
27.0	7776	-	0.6609
27.7778	8000	0.0033	0.6578
28.0	8064	-	0.6606
29.0	8352	-	0.6614
29.5139	8500	0.0031	-
30.0	8640	-	0.6579
31.0	8928	-	0.6688
31.25	9000	0.0028	0.6650
32.0	9216	-	0.6639
32.9861	9500	0.0027	-
33.0	9504	-	0.6624
34.0	9792	-	0.6646
34.7222	10000	0.0025	0.6530
35.0	10080	-	0.6587
36.0	10368	-	0.6671
36.4583	10500	0.0025	-
37.0	10656	-	0.6614
38.0	10944	-	0.6602
38.1944	11000	0.0024	0.6576
39.0	11232	-	0.6665
39.9306	11500	0.0023	-
40.0	11520	-	0.6663
41.0	11808	-	0.6734
41.6667	12000	0.0021	0.6633
42.0	12096	-	0.6667
43.0	12384	-	0.6679
43.4028	12500	0.002	-
44.0	12672	-	0.6701
45.0	12960	-	0.6650
45.1389	13000	0.0019	0.6680
46.0	13248	-	0.6631
46.875	13500	0.0018	-
47.0	13536	-	0.6643
48.0	13824	-	0.6631
48.6111	14000	0.0017	0.6648
49.0	14112	-	0.6648
50.0	14400	-	0.6619
50.3472	14500	0.0017	-
51.0	14688	-	0.6633
52.0	14976	-	0.6622
52.0833	15000	0.0016	0.6612
53.0	15264	-	0.6670
53.8194	15500	0.0015	-
54.0	15552	-	0.6618
55.0	15840	-	0.6641
55.5556	16000	0.0015	0.6617
56.0	16128	-	0.6669
57.0	16416	-	0.6645
57.2917	16500	0.0014	-
58.0	16704	-	0.6642
59.0	16992	-	0.6579
59.0278	17000	0.0013	0.6592
60.0	17280	-	0.6589
60.7639	17500	0.0014	-
61.0	17568	-	0.6685
62.0	17856	-	0.6673
62.5	18000	0.0012	0.6669
63.0	18144	-	0.6665
64.0	18432	-	0.6626
64.2361	18500	0.0012	-
65.0	18720	-	0.6619
65.9722	19000	0.0012	0.6643
66.0	19008	-	0.6651
67.0	19296	-	0.6628
67.7083	19500	0.0011	-
68.0	19584	-	0.6658
69.0	19872	-	0.6615
69.4444	20000	0.0011	0.6627
70.0	20160	-	0.6657
71.0	20448	-	0.6663
71.1806	20500	0.0011	-
72.0	20736	-	0.6634
72.9167	21000	0.001	0.6649
73.0	21024	-	0.6632
74.0	21312	-	0.6658
74.6528	21500	0.001	-
75.0	21600	-	0.6639
76.0	21888	-	0.6601
76.3889	22000	0.001	0.6623
77.0	22176	-	0.6607
78.0	22464	-	0.6613
78.125	22500	0.0009	-
79.0	22752	-	0.6613
79.8611	23000	0.0009	0.6615
80.0	23040	-	0.6615
81.0	23328	-	0.6617
81.5972	23500	0.0008	-
82.0	23616	-	0.6604
83.0	23904	-	0.6605
83.3333	24000	0.0008	0.6602
84.0	24192	-	0.6628
85.0	24480	-	0.6603
85.0694	24500	0.0008	-
86.0	24768	-	0.6602
86.8056	25000	0.0008	0.6592
87.0	25056	-	0.6611
88.0	25344	-	0.6612
88.5417	25500	0.0008	-
89.0	25632	-	0.6607
90.0	25920	-	0.6598
90.2778	26000	0.0008	0.6607
91.0	26208	-	0.6615
92.0	26496	-	0.6615
92.0139	26500	0.0007	-
93.0	26784	-	0.6609
93.75	27000	0.0007	0.6607
94.0	27072	-	0.6612
95.0	27360	-	0.6624
95.4861	27500	0.0007	-
96.0	27648	-	0.6627
97.0	27936	-	0.6618
97.2222	28000	0.0007	0.6619
98.0	28224	-	0.6621
98.9583	28500	0.0007	-
99.0	28512	-	0.6623
100.0	28800	-	0.6623

Framework Versions

• Python: 3.10.13
• Sentence Transformers: 3.0.0
• Transformers: 4.41.2
• PyTorch: 2.1.2
• Accelerate: 0.30.1
• Datasets: 2.19.2
• 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",
}