Deehan1866/finetuned-deberta

SentenceTransformer based on microsoft/deberta-v3-base

This is a sentence-transformers model finetuned from microsoft/deberta-v3-base on the PiC/phrase_similarity 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: microsoft/deberta-v3-base
• Maximum Sequence Length: 512 tokens
• Output Dimensionality: 768 tokens
• Similarity Function: Cosine Similarity
• Training Dataset:
  • PiC/phrase_similarity
• Language: en

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: DebertaV2Model 
  (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("Deehan1866/deberta")
# Run inference
sentences = [
    "She wants to write about Keima but suffers a major case of writer's block.",
    "She wants to write about Keima but suffers a huge occurrence of writer's block.",
    'specific medical status of movement and the general condition of movement both are conditions under which contradictions can move.',
]
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

Binary Classification

• Dataset: quora-duplicates-dev
• Evaluated with BinaryClassificationEvaluator

Metric	Value
cosine_accuracy	0.722
cosine_accuracy_threshold	0.9618
cosine_f1	0.7511
cosine_f1_threshold	0.8653
cosine_precision	0.6703
cosine_recall	0.854
cosine_ap	0.7484
dot_accuracy	0.676
dot_accuracy_threshold	553.1699
dot_f1	0.7329
dot_f1_threshold	531.9863
dot_precision	0.6181
dot_recall	0.9
dot_ap	0.5892
manhattan_accuracy	0.734
manhattan_accuracy_threshold	168.0139
manhattan_f1	0.7526
manhattan_f1_threshold	226.917
manhattan_precision	0.6667
manhattan_recall	0.864
manhattan_ap	0.7567
euclidean_accuracy	0.725
euclidean_accuracy_threshold	8.8652
euclidean_f1	0.7526
euclidean_f1_threshold	14.1391
euclidean_precision	0.6703
euclidean_recall	0.858
euclidean_ap	0.7558
max_accuracy	0.734
max_accuracy_threshold	553.1699
max_f1	0.7526
max_f1_threshold	531.9863
max_precision	0.6703
max_recall	0.9
max_ap	0.7567

Training Details

Training Dataset

PiC/phrase_similarity

• Dataset: PiC/phrase_similarity at fc67ce7
• Size: 7,004 training samples
• Columns: sentence1, sentence2, and label
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2	label
  type	string	string	int
  details	min: 12 tokens mean: 25.5 tokens max: 57 tokens	min: 12 tokens mean: 25.9 tokens max: 58 tokens	0: ~48.80% 1: ~51.20%

• Samples:

  sentence1	sentence2	label
  newly formed camp is released from the membrane and diffuses across the intracellular space where it serves to activate pka.	recently made encampment is released from the membrane and diffuses across the intracellular space where it serves to activate pka.	0
  According to one data, in 1910, on others – in 1915, the mansion became Natalya Dmitriyevna Shchuchkina's property.	According to a particular statistic, in 1910, on others – in 1915, the mansion became Natalya Dmitriyevna Shchuchkina's property.	1
  Note that Fact 1 does not assume any particular structure on the set formula_65.	Note that Fact 1 does not assume any specific edifice on the set formula_65.	0

• Loss: SoftmaxLoss

Evaluation Dataset

PiC/phrase_similarity

• Dataset: PiC/phrase_similarity at fc67ce7
• Size: 1,000 evaluation samples
• Columns: sentence1, sentence2, and label
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2	label
  type	string	string	int
  details	min: 10 tokens mean: 25.46 tokens max: 58 tokens	min: 11 tokens mean: 25.84 tokens max: 59 tokens	0: ~50.00% 1: ~50.00%

• Samples:

  sentence1	sentence2	label
  after theo's apparent death, she decides to leave first colony and ends up traveling with the apostles.	after theo's apparent death, she decides to leave original settlement and ends up traveling with the apostles.	0
  The guard assigned to Vivian leaves her to prevent the robbery, allowing her to connect to the bank's network.	The guard assigned to Vivian leaves her to prevent the robbery, allowing her to connect to the bank's locations.	0
  Two days later Louis XVI banished Necker by a "lettre de cachet" for his very public exchange of pamphlets.	Two days later Louis XVI banished Necker by a "lettre de cachet" for his very free forum of pamphlets.	0

• Loss: SoftmaxLoss

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• per_device_train_batch_size: 16
• per_device_eval_batch_size: 16
• learning_rate: 2e-05
• num_train_epochs: 5
• warmup_ratio: 0.1
• load_best_model_at_end: 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
• 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: 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: 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: True
• 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	quora-duplicates-dev_max_ap
0	0	-	-	0.6315
0.2283	100	-	0.6770	0.6409
0.4566	200	-	0.6208	0.6930
0.6849	300	-	0.5337	0.7567
0.9132	400	-	0.5565	0.7696
1.1416	500	0.5876	0.5897	0.7682
1.3699	600	-	0.5412	0.7747
1.5982	700	-	0.5578	0.7757
1.8265	800	-	0.5836	0.7689
2.0548	900	-	0.5749	0.7727
2.2831	1000	0.3988	0.6549	0.7706
2.5114	1100	-	0.6600	0.7731
2.7397	1200	-	0.6223	0.7757
2.9680	1300	-	0.7001	0.7604
3.1963	1400	-	0.7821	0.7633
3.4247	1500	0.2379	0.8848	0.7602
3.6530	1600	-	0.8800	0.7615
3.8813	1700	-	0.9374	0.7670
4.1096	1800	-	0.9646	0.7668
4.3379	1900	-	0.9942	0.7673
4.5662	2000	0.1714	0.9849	0.7667
4.7945	2100	-	1.0025	0.7671
5.0	2190	-	-	0.7567

• The bold row denotes the saved checkpoint.

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 and SoftmaxLoss

@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",
}