MPNet base trained on AllNLI triplets

This is a sentence-transformers model finetuned from prajjwal1/bert-tiny on the pair_similarity_new_1 dataset. It maps sentences & paragraphs to a 128-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: prajjwal1/bert-tiny
• Maximum Sequence Length: 512 tokens
• Output Dimensionality: 128 dimensions
• Similarity Function: Cosine Similarity
• Training Dataset:
  • pair_similarity_new_1
• Language: en
• License: apache-2.0

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': 128, '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("Tien09/tiny_bert_ft_sim_score_2")
# Run inference
sentences = [
    'When this card destroys an opponent\'s monster by battle and sends it to the GY: You can discard 1 WATER monster to the GY; Special Summon 1 "Mermail" monster from your Deck in face-up Defense Position. You can only use the effect of "Mermail Abyssnose" once per turn.',
    'If "Obsidim, the Ashened City" is in the Field Zone, you can Special Summon this card (from your hand). You can only Special Summon "King of the Ashened City" once per turn this way. During your Main Phase: You can Special Summon 1 "Ashened" monster from your hand, except "King of the Ashened City", or if your opponent controls a monster with 2800 or more ATK, you can Special Summon it from your Deck instead. You can only use this effect of "King of the Ashened City" once per turn.',
    'Activate only when your opponent declares a direct attack and you control no monsters. Special Summon 1 Level 4 or lower Beast-Type monster from your hand in face-up Attack Position.',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 128]

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

Training Details

Training Dataset

pair_similarity_new_1

• Dataset: pair_similarity_new_1 at a250d43
• Size: 8,959 training samples
• Columns: effect_text, score, and effect_text2
• Approximate statistics based on the first 1000 samples:

  	effect_text	score	effect_text2
  type	string	float	string
  details	min: 9 tokens mean: 73.57 tokens max: 204 tokens	min: 0.0 mean: 0.43 max: 1.0	min: 4 tokens mean: 71.17 tokens max: 193 tokens

• Samples:

  effect_text	score	effect_text2
  When your opponent's monster attacks a face-up Level 4 or lower Toon Monster on your side of the field, you can make the attack a direct attack to your Life Points.	0.0	During either player's Main Phase: Special Summon this card as a Normal Monster (Reptile-Type/EARTH/Level 4/ATK 1600/DEF 1800). (This card is also still a Trap Card.)
  When your opponent Special Summons a monster, you can discard 1 card to Special Summon this card from your hand. Your opponent cannot remove cards from play.	1.0	Activate this card by discarding 1 monster, then target 1 monster in your GY whose Level is lower than the discarded monster's original Level; Special Summon it and equip it with this card. The equipped monster has its effects negated. You can only activate 1 "Overdone Burial" per turn.
  Mystical Elf" + "Curtain of the Dark Ones	0.0	A lost dog that wandered off 1000 years ago. He's still waiting for his master to come for him.

• Loss: CoSENTLoss with these parameters:

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

Evaluation Dataset

pair_similarity_new_1

• Dataset: pair_similarity_new_1 at a250d43
• Size: 1,920 evaluation samples
• Columns: effect_text, score, and effect_text2
• Approximate statistics based on the first 1000 samples:

  	effect_text	score	effect_text2
  type	string	float	string
  details	min: 6 tokens mean: 72.29 tokens max: 190 tokens	min: 0.0 mean: 0.43 max: 1.0	min: 8 tokens mean: 71.47 tokens max: 199 tokens

• Samples:

  effect_text	score	effect_text2
  2+ Level 4 monsters This Xyz Summoned card gains 500 ATK x the total Link Rating of Link Monsters linked to this card. You can detach 2 materials from this card, then target 1 4 Cyberse Link Monster in your GY; Special Summon it to your field so it points to this card, also you cannot Special Summon other monsters or attack directly for the rest of this turn.	1.0	3 Level 4 monsters Once per turn, you can also Xyz Summon "Zoodiac Tigermortar" by using 1 "Zoodiac" monster you control with a different name as Xyz Material. (If you used an Xyz Monster, any Xyz Materials attached to it also become Xyz Materials on this card.) This card gains ATK and DEF equal to the ATK and DEF of all "Zoodiac" monsters attached to it as Materials. Once per turn: You can detach 1 Xyz Material from this card, then target 1 Xyz Monster you control and 1 "Zoodiac" monster in your GY; attach that "Zoodiac" monster to that Xyz Monster as Xyz Material.
  1 Tuner + 1 or more non-Tuner Pendulum Monsters Once per turn: You can target 1 Pendulum Monster on the field or 1 card in the Pendulum Zone; destroy it, and if you do, shuffle 1 card on the field into the Deck. Once per turn: You can Special Summon 1 "Dracoslayer" monster from your Deck in Defense Position, but it cannot be used as a Synchro Material for a Summon.	0.0	If this card is Special Summoned: You can add 1 "Performapal" monster from your Deck to your hand, except a Pendulum Monster. You can only use this effect of "Performapal Longphone Bull" once per turn.
  If you control an Illusion or Spellcaster monster: Add 1 "White Forest" monster from your Deck to your hand. If this card is sent to the GY to activate a monster effect: You can Set this card. You can only use each effect of "Tales of the White Forest" once per turn.	0.25	Cannot be destroyed by your opponent's card effects while "Multi-Universe" is on the field. You can only use each of the following effects of "Krishnerd Witch" once per turn. If a card(s) in the Field Zone leaves the field by card effect (except during the Damage Step): You can Special Summon this card from your hand. When a Field Spell that is already face-up on the field activates its effect (Quick Effect): You can shuffle 1 of your monsters that is banished or in your GY into the Deck, or if that monster mentions that Field Spell, you can Special Summon it instead.

• Loss: CoSENTLoss with these parameters:

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

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• per_device_train_batch_size: 16
• per_device_eval_batch_size: 16
• num_train_epochs: 5
• warmup_ratio: 0.1
• fp16: True
• batch_sampler: no_duplicates

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: 5
• 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: None
• hub_always_push: False
• 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
• 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
• use_liger_kernel: False
• eval_use_gather_object: False
• average_tokens_across_devices: False
• prompts: None
• batch_sampler: no_duplicates
• multi_dataset_batch_sampler: proportional

Training Logs

Framework Versions

• Python: 3.10.12
• Sentence Transformers: 3.3.1
• Transformers: 4.47.1
• PyTorch: 2.5.1+cu121
• Accelerate: 1.2.1
• Datasets: 3.2.0
• Tokenizers: 0.21.0

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",
}

CoSENTLoss

@online{kexuefm-8847,
    title={CoSENT: A more efficient sentence vector scheme than Sentence-BERT},
    author={Su Jianlin},
    year={2022},
    month={Jan},
    url={https://kexue.fm/archives/8847},
}