kushalc1/sarashina-embedding-v2-1b-jsts-matryoshka

SentenceTransformer based on sbintuitions/sarashina-embedding-v2-1b

This is a sentence-transformers model finetuned from sbintuitions/sarashina-embedding-v2-1b on the jsts dataset. It maps sentences & paragraphs to a 1792-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: sbintuitions/sarashina-embedding-v2-1b
• Maximum Sequence Length: 8192 tokens
• Output Dimensionality: 1792 dimensions
• Similarity Function: Cosine Similarity
• Training Dataset:
  • jsts
• Language: jpn

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': 8192, 'do_lower_case': False, 'architecture': 'LlamaModel'})
  (1): Pooling({'word_embedding_dimension': 1792, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': False, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': True, 'include_prompt': False})
)

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("kushalc1/sarashina-embedding-v2-1b-jsts-matryoshka")
# Run inference
sentences = [
    '樹木に囲まれた芝生の上に三頭のキリンが立っています。',
    '芝生の上に数頭のキリンが歩いています。',
    '茶色のテーブルの上にピザと飲み物が置かれています。',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 1792]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities)
# tensor([[1.0000, 0.9453, 0.4754],
#         [0.9453, 1.0000, 0.5004],
#         [0.4754, 0.5004, 1.0000]])

Evaluation

Metrics

Semantic Similarity

• Datasets: sts-dev-1792, sts-test-1792 and sts-test-1792
• Evaluated with EmbeddingSimilarityEvaluator with these parameters:

  {
      "truncate_dim": 1792
  }
  

Metric	sts-dev-1792	sts-test-1792
pearson_cosine	0.8088	0.8088
spearman_cosine	0.7435	0.7435

Semantic Similarity

• Datasets: sts-dev-1280, sts-test-1280 and sts-test-1280
• Evaluated with EmbeddingSimilarityEvaluator with these parameters:

  {
      "truncate_dim": 1280
  }
  

Metric	sts-dev-1280	sts-test-1280
pearson_cosine	0.8078	0.8078
spearman_cosine	0.7442	0.7442

Semantic Similarity

• Datasets: sts-dev-768, sts-test-768 and sts-test-768
• Evaluated with EmbeddingSimilarityEvaluator with these parameters:

  {
      "truncate_dim": 768
  }
  

Metric	sts-dev-768	sts-test-768
pearson_cosine	0.8049	0.8049
spearman_cosine	0.7423	0.7423

Semantic Similarity

• Datasets: sts-dev-256, sts-test-256 and sts-test-256
• Evaluated with EmbeddingSimilarityEvaluator with these parameters:

  {
      "truncate_dim": 256
  }
  

Metric	sts-dev-256	sts-test-256
pearson_cosine	0.8022	0.8022
spearman_cosine	0.7411	0.7411

Semantic Similarity

• Datasets: sts-dev-64, sts-test-64 and sts-test-64
• Evaluated with EmbeddingSimilarityEvaluator with these parameters:

  {
      "truncate_dim": 64
  }
  

Metric	sts-dev-64	sts-test-64
pearson_cosine	0.7972	0.7972
spearman_cosine	0.7389	0.7389

Training Details

Training Dataset

jsts

• Dataset: jsts at b3d3097
• Size: 12,451 training samples
• Columns: sentence1, sentence2, and score
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2	score
  type	string	string	float
  details	min: 5 tokens mean: 10.64 tokens max: 35 tokens	min: 3 tokens mean: 10.53 tokens max: 30 tokens	min: 0.0 mean: 2.32 max: 5.0

• Samples:

  sentence1	sentence2	score
  川べりでサーフボードを持った人たちがいます。	トイレの壁に黒いタオルがかけられています。	0.0
  二人の男性がジャンボジェット機を見ています。	2人の男性が、白い飛行機を眺めています。	3.799999952316284
  男性が子供を抱き上げて立っています。	坊主頭の男性が子供を抱いて立っています。	4.0

• Loss: MatryoshkaLoss with these parameters:

  {
      "loss": "MultipleNegativesRankingLoss",
      "matryoshka_dims": [
          1792,
          1280,
          768,
          256,
          64
      ],
      "matryoshka_weights": [
          1,
          1,
          1,
          1,
          1
      ],
      "n_dims_per_step": -1
  }
  

Evaluation Dataset

jsts

• Dataset: jsts at b3d3097
• Size: 1,457 evaluation samples
• Columns: sentence1, sentence2, and score
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2	score
  type	string	string	float
  details	min: 5 tokens mean: 10.78 tokens max: 34 tokens	min: 3 tokens mean: 10.63 tokens max: 37 tokens	min: 0.0 mean: 2.22 max: 5.0

• Samples:

  sentence1	sentence2	score
  レンガの建物の前を、乳母車を押した女性が歩いています。	厩舎で馬と女性とが寄り添っています。	0.0
  山の上に顔の白い牛が2頭います。	曇り空の山肌で、牛が２匹草を食んでいます。	2.4000000953674316
  バナナを持った人が道路を通行しています。	道の上をバナナを背負った男性が歩いています。	3.5999999046325684

• Loss: MatryoshkaLoss with these parameters:

  {
      "loss": "MultipleNegativesRankingLoss",
      "matryoshka_dims": [
          1792,
          1280,
          768,
          256,
          64
      ],
      "matryoshka_weights": [
          1,
          1,
          1,
          1,
          1
      ],
      "n_dims_per_step": -1
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• per_device_train_batch_size: 16
• per_device_eval_batch_size: 16
• num_train_epochs: 4
• warmup_ratio: 0.1
• fp16: True

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
• 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: 1.0
• num_train_epochs: 4
• 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}
• 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
• 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: False
• 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: False
• prompts: None
• batch_sampler: batch_sampler
• multi_dataset_batch_sampler: proportional
• router_mapping: {}
• learning_rate_mapping: {}

Training Logs

Epoch	Step	Training Loss	Validation Loss	sts-dev-1792_spearman_cosine	sts-dev-1280_spearman_cosine	sts-dev-768_spearman_cosine	sts-dev-256_spearman_cosine	sts-dev-64_spearman_cosine	sts-test-1792_spearman_cosine	sts-test-1280_spearman_cosine	sts-test-768_spearman_cosine	sts-test-256_spearman_cosine	sts-test-64_spearman_cosine
0.1284	100	4.911	6.2373	0.7573	0.7572	0.7517	0.7350	0.7114	-	-	-	-	-
0.2567	200	5.8664	7.9813	0.6739	0.6692	0.6636	0.6500	0.6146	-	-	-	-	-
0.3851	300	7.259	7.9829	0.6831	0.6815	0.6797	0.6698	0.6529	-	-	-	-	-
0.5135	400	7.1234	7.5810	0.6878	0.6887	0.6881	0.6819	0.6679	-	-	-	-	-
0.6418	500	7.233	6.9384	0.6628	0.6694	0.6658	0.6662	0.6583	-	-	-	-	-
0.7702	600	7.0228	7.0102	0.6352	0.6364	0.6346	0.6310	0.6246	-	-	-	-	-
0.8986	700	6.539	6.7671	0.6411	0.6415	0.6403	0.6394	0.6346	-	-	-	-	-
1.0270	800	6.2863	7.5846	0.6120	0.6342	0.6314	0.6266	0.6189	-	-	-	-	-
1.1553	900	5.7608	6.7480	0.6773	0.6790	0.6758	0.6748	0.6691	-	-	-	-	-
1.2837	1000	5.672	6.6481	0.6846	0.6836	0.6817	0.6834	0.6794	-	-	-	-	-
1.4121	1100	5.7371	6.6843	0.6945	0.6953	0.6966	0.6939	0.6891	-	-	-	-	-
1.5404	1200	5.8827	6.6863	0.6903	0.6940	0.6922	0.6883	0.6834	-	-	-	-	-
1.6688	1300	5.6242	6.6517	0.6856	0.6857	0.6847	0.6809	0.6762	-	-	-	-	-
1.7972	1400	5.5211	6.1428	0.7134	0.7123	0.7117	0.7065	0.7022	-	-	-	-	-
1.9255	1500	5.4882	6.0439	0.7227	0.7227	0.7214	0.7192	0.7134	-	-	-	-	-
2.0539	1600	5.4436	6.0361	0.7199	0.7203	0.7191	0.7201	0.7143	-	-	-	-	-
2.1823	1700	4.366	6.1447	0.7286	0.7290	0.7274	0.7283	0.7231	-	-	-	-	-
2.3107	1800	4.6607	6.1692	0.7365	0.7356	0.7344	0.7303	0.7263	-	-	-	-	-
2.4390	1900	4.3651	6.2109	0.7178	0.7169	0.7149	0.7134	0.7125	-	-	-	-	-
2.5674	2000	4.4692	6.1421	0.7237	0.7233	0.7214	0.7192	0.7150	-	-	-	-	-
2.6958	2100	4.434	5.9462	0.7275	0.7267	0.7260	0.7253	0.7203	-	-	-	-	-
2.8241	2200	4.2634	6.0055	0.7218	0.7216	0.7205	0.7196	0.7177	-	-	-	-	-
2.9525	2300	4.2524	5.8834	0.7297	0.7308	0.7302	0.7282	0.7245	-	-	-	-	-
3.0809	2400	3.5146	6.2635	0.7430	0.7425	0.7416	0.7402	0.7357	-	-	-	-	-
3.2092	2500	3.0137	6.1396	0.7455	0.7441	0.7430	0.7410	0.7377	-	-	-	-	-
3.3376	2600	2.9956	6.2779	0.7426	0.7427	0.7407	0.7402	0.7371	-	-	-	-	-
3.4660	2700	3.0125	6.2415	0.7459	0.7457	0.7435	0.7425	0.7378	-	-	-	-	-
3.5944	2800	3.2683	6.2214	0.7407	0.7407	0.7385	0.7378	0.7342	-	-	-	-	-
3.7227	2900	2.7818	6.2854	0.7444	0.7442	0.7422	0.7411	0.7390	-	-	-	-	-
3.8511	3000	2.7216	6.2760	0.7425	0.7429	0.7411	0.7401	0.7378	-	-	-	-	-
3.9795	3100	2.8901	6.2306	0.7435	0.7442	0.7423	0.7411	0.7389	-	-	-	-	-
-1	-1	-	-	-	-	-	-	-	0.7435	0.7442	0.7423	0.7411	0.7389

Framework Versions

• Python: 3.12.6
• Sentence Transformers: 5.2.0
• Transformers: 4.56.0
• PyTorch: 2.8.0+cu129
• Accelerate: 1.10.1
• Datasets: 4.4.2
• Tokenizers: 0.22.0

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",
}

MatryoshkaLoss

@misc{kusupati2024matryoshka,
    title={Matryoshka Representation Learning},
    author={Aditya Kusupati and Gantavya Bhatt and Aniket Rege and Matthew Wallingford and Aditya Sinha and Vivek Ramanujan and William Howard-Snyder and Kaifeng Chen and Sham Kakade and Prateek Jain and Ali Farhadi},
    year={2024},
    eprint={2205.13147},
    archivePrefix={arXiv},
    primaryClass={cs.LG}
}

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}
}