LamaDiab/MiniLM-SemanticEngine

SentenceTransformer based on sentence-transformers/all-MiniLM-L6-v2

This is a sentence-transformers model finetuned from sentence-transformers/all-MiniLM-L6-v2. It maps sentences & paragraphs to a 384-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: sentence-transformers/all-MiniLM-L6-v2
• Maximum Sequence Length: 256 tokens
• Output Dimensionality: 384 dimensions
• 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': 256, 'do_lower_case': False, 'architecture': 'BertModel'})
  (1): Pooling({'word_embedding_dimension': 384, '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): Normalize()
)

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("LamaDiab/MiniLM-SemanticEngine")
# Run inference
sentences = [
    'hiit biker shorts - black',
    'black shorts',
    'winter slippers for ladies christmas themed',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 384]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities)
# tensor([[ 1.0000,  0.7103, -0.0705],
#         [ 0.7103,  1.0000, -0.0356],
#         [-0.0705, -0.0356,  1.0000]])

Evaluation

Metrics

Triplet

• Evaluated with TripletEvaluator

Metric	Value
cosine_accuracy	0.9472

Training Details

Training Dataset

Unnamed Dataset

• Size: 169,967 training samples
• Columns: anchor and positive
• Approximate statistics based on the first 1000 samples:

  	anchor	positive
  type	string	string
  details	min: 3 tokens mean: 8.82 tokens max: 237 tokens	min: 3 tokens mean: 14.99 tokens max: 256 tokens

• Samples:

  anchor	positive
  orasi barista almond milk is a premium, plant-based milk designed specifically for coffee lovers. crafted to create the perfect froth, it delivers a smooth and creamy texture that enhances the flavor of your lattes, cappuccinos, and other coffee drinks.	groceries
  this toy is a "modern fashion" doll, combining beauty and innovation in its design. the doll has long and pink hair that adds a modern and attractive character to it. it comes with a wide variety of clothes and cool accessories that allow children to switch outfits and try different looks.
  features:
  modern and attractive design: the doll has a stylish and modern design that suits the tastes of children of different ages.
  long and colorful hair: long and colorful hair gives the doll a distinctive and beautiful look, enhancing the possibilities of play and creativity.
  wide range of clothes: the game has a large assortment of clothes that allow children to choose the appropriate outfits for the doll character according to their imagination.
  multiple accessories: it comes with various accessories that add a touch of distinction and elegance to the doll, allowing to experiment with different styles.
  stimulate creativity and imagination: the game helps enhance children's imagination by...	kids
  zinnia ice box vivid gen.2 - blue	blue ice box

• Loss: MultipleNegativesSymmetricRankingLoss with these parameters:

  {
      "scale": 20.0,
      "similarity_fct": "cos_sim",
      "gather_across_devices": true
  }
  

Evaluation Dataset

Unnamed Dataset

• Size: 16,216 evaluation samples
• Columns: anchor, positive, and negative
• Approximate statistics based on the first 1000 samples:

  	anchor	positive	negative
  type	string	string	string
  details	min: 3 tokens mean: 9.79 tokens max: 52 tokens	min: 2 tokens mean: 19.21 tokens max: 256 tokens	min: 3 tokens mean: 9.76 tokens max: 67 tokens

• Samples:

  anchor	positive	negative
  dosado ring	dosado or dos- à- dos: a wavy movement of two people around eachother, without turning & facing the same direction. material: 18k gold plated hammered brass. size: one size, adjustable. care instructions: to keep the jewelry pieces looking as good as new, please make sure that you store them in an airtight container. they should not come in contact with sweat, water or pefume, alcohol, sanitizers etc. polish with a microfiber cloth.	kiprun ks light men's running shoes - black
  puzzle city of fog	this amazing puzzle offers a unique opportunity to explore the beauty of san francisco, also known as the "city by the bay," through assembling a 2000-piece jigsaw. you'll immerse yourself in a world full of colors and details, as your eyes wander across the iconic golden gate bridge, towering buildings, distinctive hilly streets, and sailing ships in the harbor. it’s a panoramic depiction of san francisco, providing a comprehensive view of the city and its landmarks. features: explore san francisco: enjoy a virtual exploration of san francisco without leaving your home. get up close with famous landmarks such as the golden gate bridge and the harbor. improves cognitive skills: assembling the puzzle enhances focus, memory, and fine motor skills while boosting problem-solving and decision-making abilities. relaxation and stress relief: puzzle assembly is a fun and engaging activity that helps to relax and reduce stress, especially when concentrating on the appealing details of san franc...	unicorn
  my fault series	mercedes ron book	sophie's world

• Loss: MultipleNegativesSymmetricRankingLoss with these parameters:

  {
      "scale": 20.0,
      "similarity_fct": "cos_sim",
      "gather_across_devices": true
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• per_device_train_batch_size: 64
• per_device_eval_batch_size: 64
• weight_decay: 0.01
• num_train_epochs: 5
• warmup_ratio: 0.2
• fp16: True
• dataloader_num_workers: 2
• dataloader_prefetch_factor: 2
• push_to_hub: True
• hub_model_id: LamaDiab/MiniLM-SemanticEngine
• 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: 64
• per_device_eval_batch_size: 64
• 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: 0.01
• 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: 0.2
• 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: 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: 2
• dataloader_prefetch_factor: 2
• 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: True
• resume_from_checkpoint: None
• hub_model_id: LamaDiab/MiniLM-SemanticEngine
• 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: no_duplicates
• multi_dataset_batch_sampler: proportional
• router_mapping: {}
• learning_rate_mapping: {}

Training Logs

Epoch	Step	Training Loss	Validation Loss	cosine_accuracy
0.0004	1	1.6989	-	-
0.1883	500	1.6103	1.4441	0.9124
0.3765	1000	1.1942	1.3155	0.9233
0.5648	1500	0.9831	1.2584	0.9257
0.7530	2000	0.8867	1.2368	0.9254
0.9413	2500	0.8094	1.1874	0.9274
1.1295	3000	0.5818	1.1431	0.9348
1.3178	3500	0.6978	1.1291	0.9374
1.5060	4000	0.6652	1.0936	0.9389
1.6943	4500	0.6287	1.0889	0.9369
1.8825	5000	0.5986	1.0780	0.9404
2.0708	5500	0.4376	1.0783	0.9386
2.2590	6000	0.511	1.0674	0.9405
2.4473	6500	0.4997	1.0412	0.9427
2.6355	7000	0.4985	1.0160	0.9441
2.8238	7500	0.4798	1.0264	0.9434
3.0120	8000	0.3477	1.0153	0.9455
3.2003	8500	0.4117	1.0177	0.9461
3.3886	9000	0.4302	1.0071	0.9451
3.5768	9500	0.4046	1.0171	0.9460
3.7651	10000	0.414	0.9819	0.9474
3.9533	10500	0.3786	0.9982	0.9463
4.1416	11000	0.2952	0.9920	0.9461
4.3298	11500	0.3655	0.9959	0.9455
4.5181	12000	0.3655	0.9961	0.9464
4.7063	12500	0.3662	0.9826	0.9467
4.8946	13000	0.3545	0.9864	0.9472

Framework Versions

• Python: 3.11.13
• Sentence Transformers: 5.1.2
• Transformers: 4.53.3
• PyTorch: 2.6.0+cu124
• Accelerate: 1.9.0
• Datasets: 4.4.1
• Tokenizers: 0.21.2

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",
}