Bharatdeep-H/stella_finetuned_en_dataset_to_mine_negatives_from

SentenceTransformer based on NovaSearch/stella_en_400M_v5

This is a sentence-transformers model finetuned from NovaSearch/stella_en_400M_v5 on the csv dataset. It maps sentences & paragraphs to a 1024-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: NovaSearch/stella_en_400M_v5
• Maximum Sequence Length: 512 tokens
• Output Dimensionality: 1024 dimensions
• Similarity Function: Cosine Similarity
• Training Dataset:
  • csv

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: NewModel 
  (1): Pooling({'word_embedding_dimension': 1024, '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})
  (2): Dense({'in_features': 1024, 'out_features': 1024, 'bias': True, 'activation_function': 'torch.nn.modules.linear.Identity'})
)

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("Bharatdeep-H/stella_finetuned_en_dataset_to_mine_negatives_from")
# Run inference
sentences = [
    'day I was acquitted 12/12/12 i hocus45th GP SERVICES USA CDC CENTERS FOR DISH CONTROL AND P EXCLUSIVE: Per the CDC There Are Nearly Twice As Many Vaccine Related Deaths SO FAR in 2021 (1,755) Than All the Vaccine Deaths this Past Decade (994) For information about vaccines. visit who.int.',
    '“Thousands of COVID Vaccine Injuries and 13 U.S. Deaths Reported in December Alone”; “In December, 3,916 COVID vaccine-related adverse events, including 13 deaths, were reported to VAERS”',
    'Side effects of the first published vaccine According to Pfizer documents, 1,200 deaths.',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 1024]

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

Evaluation

Metrics

Triplet

• Evaluated with TripletEvaluator

Metric	Value
cosine_accuracy	0.9714

Training Details

Training Dataset

csv

• Dataset: csv
• Size: 164,619 training samples
• Columns: query, answer, and negative
• Approximate statistics based on the first 1000 samples:

  	query	answer	negative
  type	string	string	string
  details	min: 5 tokens mean: 83.66 tokens max: 512 tokens	min: 6 tokens mean: 19.17 tokens max: 141 tokens	min: 6 tokens mean: 18.69 tokens max: 120 tokens

• Samples:

  query	answer	negative
  YEAST EXTRACK C DDees.co NO AFTER 40 YEARS OF RESEARCH, THERE IS A VACCINE AGAINST AIDS AFTER 76 YEARS OF RESEARCH, NO THERE IS A VACCINE AGAINST INFLUENZA AFTER 100 YEARS OF RESEARCH, NO THERE IS A VACCINE AGAINST CANCER BUT AFTER ONLY 6 MONTHS THERE IS A VACCINE AGAINST A "VIRUS" THAT APPEARED "FROM SUDDEN" AND THOSE WHO	After years of research there is no vaccine against AIDS, cancer and influenza, but in 6 months there is a vaccine against covid-19	Chronology of decline in vaccine effectiveness shows that immunizers are not effective
  Donald J. Trump [USER] Trump We have reports of NIGERIA supporting Iran ACTIVELY and PASSIVELY....just got of the phone with secretary Pompeo....the USA MUST and WILL respond to terrorist nations IMMEDIATELY! 12:11 AM - 08 Jan 2020 244,391 Retweets 319,284 Likes 167K [244K Follow 319K	US President Donald Trump threatens to attack Nigeria	Vice President of the United States Mike Pence unfollowed President Donald Trump on Twitter and put a picture of Joe Biden and Kamala Harris on the profile cover
  4 k 40 years worth of research... *no vaccine for HIV cancer MA At least 100 years of research...no vaccine for Ongoing research... no vaccine for the common cold Less than a year for a Covid vaccine? I think I'll pass on that shot!! PIC COLLAGE	Vaccines for HIV, cold and cancer should deter you from getting the Covid-19 vaccine	Anticovid vaccines cause deterioration of the immune system and AIDS

• Loss: TripletLoss with these parameters:

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

Evaluation Dataset

csv

• Dataset: csv
• Size: 164,619 evaluation samples
• Columns: query, answer, and negative
• Approximate statistics based on the first 1000 samples:

  	query	answer	negative
  type	string	string	string
  details	min: 5 tokens mean: 73.2 tokens max: 512 tokens	min: 5 tokens mean: 19.27 tokens max: 103 tokens	min: 5 tokens mean: 19.3 tokens max: 179 tokens

• Samples:

  query	answer	negative
  --- Nora Al Riyadh Tik Tok Replying to [USER]	A restaurant in Riyadh in one of the malls, the opening of the meal is free, of course, Corona in the farthest corner is surprising	Video of Rakhi Sawant wearing a hijab to support protestors in Karnataka
  SQUID FOR BRAZIL BELTER SALT Corumbau pasil Milk	Lula expelled from the city of Itanagrà in Bahia in May 2021. The Army had to provide security.	All workers in Gardenia Philippines bread factory COVID-19 positive in July 2020
  I just ran out of words William Barr, Attorney General of the America literally most important person of all the American court system just publicly denounced that there has been electoral fraud 2rad10 TM [USER].6h US Attorney General William Barr denounces Vote-by-mail fraud. OM BLITZER [USER] THE WITH THE WITH SITUATION WOLFOOTION WOLF S BLITZE ROOM OUTZER TH HE WITH DERNIE CNN EXCLUSIVE Jimmy Carter & James Baker WOLF ONE-ON-ONE WITH ATTORNEY GENERAL WILLIAM BARR CAN WELL DEEST	The US attorney general denounces that there has been electoral fraud	Hillary Clinton appeared before the US justice on June 2, 2020

• Loss: TripletLoss with these parameters:

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

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• per_device_train_batch_size: 2
• learning_rate: 3e-05
• max_steps: 4000
• lr_scheduler_type: cosine
• warmup_ratio: 0.2
• bf16: 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: 2
• per_device_eval_batch_size: 8
• 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: 3e-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: 3.0
• max_steps: 4000
• lr_scheduler_type: cosine
• lr_scheduler_kwargs: {}
• warmup_ratio: 0.2
• 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: True
• 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: 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

Epoch	Step	Training Loss	Validation Loss	cosine_accuracy
0.0016	100	1.6958	-	-
0.0032	200	1.3647	-	-
0.0049	300	1.1698	-	-
0.0065	400	0.8551	-	-
0.0081	500	0.8275	-	-
0.0097	600	0.8878	-	-
0.0113	700	0.9717	-	-
0.0130	800	1.0219	-	-
0.0146	900	0.9074	-	-
0.0162	1000	0.903	0.8201	0.9452
0.0178	1100	0.9236	-	-
0.0194	1200	0.7935	-	-
0.0211	1300	1.0483	-	-
0.0227	1400	1.0878	-	-
0.0243	1500	0.9258	-	-
0.0259	1600	1.011	-	-
0.0275	1700	0.7785	-	-
0.0292	1800	0.7643	-	-
0.0308	1900	0.9918	-	-
0.0324	2000	0.7941	0.7678	0.9387
0.0340	2100	1.106	-	-
0.0356	2200	0.7571	-	-
0.0373	2300	0.6687	-	-
0.0389	2400	0.6914	-	-
0.0405	2500	0.5925	-	-
0.0421	2600	0.8085	-	-
0.0437	2700	0.5775	-	-
0.0454	2800	0.5051	-	-
0.0470	2900	0.6894	-	-
0.0486	3000	0.4202	0.4875	0.9667
0.0502	3100	0.4704	-	-
0.0518	3200	0.4511	-	-
0.0535	3300	0.3991	-	-
0.0551	3400	0.4166	-	-
0.0567	3500	0.3402	-	-
0.0583	3600	0.6621	-	-
0.0599	3700	0.5999	-	-
0.0616	3800	0.443	-	-
0.0632	3900	0.6503	-	-
0.0648	4000	0.42	0.4156	0.9714

Framework Versions

• Python: 3.10.16
• Sentence Transformers: 3.3.1
• Transformers: 4.49.0
• PyTorch: 2.5.1+cu121
• Accelerate: 1.4.0
• Datasets: 3.3.2
• 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",
}

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}
}