deedcon/bi-encoder-v2

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}) with Transformer model: 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("deedcon/bi-encoder-v2")
# Run inference
sentences = [
    '[CLS] [KNOWLEDGE] Kunden-Konzernstandards [CTX] Konzeption, Erstellung und Umsetzung der IT-Sicherheitsrichtlinien für die Freigabe von geheimen Daten gemäß Kunden-Konzernstandards (ISO 27001, BSI-Grundschutz)\n [SEP]',
    '[CLS] [KNOWLEDGE] Kundenberatung [CTX] Kundenberatung [SEP]',
    '[CLS] [KNOWLEDGE] Service-Katalog [CTX] Beauftragen der Hard- und Software, bzw. Leistungserbringung gem. Service-Katalog\n [SEP]',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 384]

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

Evaluation

Metrics

Semantic Similarity

• Datasets: sts-dev and sts-test
• Evaluated with EmbeddingSimilarityEvaluator

Metric	sts-dev	sts-test
pearson_cosine	0.9898	0.9897
spearman_cosine	0.9571	0.9592

Training Details

Training Dataset

Unnamed Dataset

• Size: 86,807 training samples
• Columns: text1, text2, and score
• Approximate statistics based on the first 1000 samples:

  	text1	text2	score
  type	string	string	float
  details	min: 15 tokens mean: 62.0 tokens max: 256 tokens	min: 15 tokens mean: 61.69 tokens max: 256 tokens	min: -0.08 mean: 0.45 max: 1.0

• Samples:

  text1	text2	score
  [CLS] [SKILL] Dokumentieren Netzwerkinfrastruktur [CTX] Dokumentieren und Skizzieren der Netzwerkinfrastruktur. [SEP]	[CLS] [SKILL] Abarbeitung Incidents [CTX] Abarbeitung von Changes/ Incidents in Jira [SEP]	0.1672067940235138
  [CLS] [SKILL] Durchführung Abnahmetests [CTX] Erstellung von Testplänen und Durchführung von Abnahmetests im Rahmen des Release Managements. [SEP]	[CLS] [KNOWLEDGE] Kostenrechnungswesen [CTX] Ist-Analyse der Produktionsabläufe sowie des vorhandene Kostenrechnungswesen . [SEP]	0.2705094516277313
  [CLS] [KNOWLEDGE] MS SQL Datenbankabfragen [CTX] MS SQL Datenbankabfragen . [SEP]	[CLS] [SKILL] Erstellung MS SQL Datenbank [CTX] Erstellung & Modellierung der MS SQL Datenbank über Entity Framework [SEP]	0.8471388816833496

• Loss: CosineSimilarityLoss with these parameters:

  {
      "loss_fct": "torch.nn.modules.loss.MSELoss"
  }
  

Evaluation Dataset

Unnamed Dataset

• Size: 17,361 evaluation samples
• Columns: text1, text2, and score
• Approximate statistics based on the first 1000 samples:

  	text1	text2	score
  type	string	string	float
  details	min: 14 tokens mean: 60.26 tokens max: 256 tokens	min: 15 tokens mean: 61.41 tokens max: 256 tokens	min: -0.09 mean: 0.46 max: 1.0

• Samples:

  text1	text2	score
  [CLS] [SKILL] Installation 7 [CTX] Installation, Konfiguration und Verwaltung von: Windows XP, 7, 8, 8.1, 10 [SEP]	[CLS] [SKILL] Installation 10 [CTX] Installation, Konfiguration und Verwaltung von: Windows XP, 7, 8, 8.1, 10 [SEP]	0.9105218082666396
  [CLS] [KNOWLEDGE] Oracle-Systemadministration [CTX] Installation, Administration (Backup/Recovery etc.) und Tuning von DB2 nach SAP R/3 Gesichtspunkten bzw. DB2-Applikationsprogrammierung (Auswertung und Verarbeitung von DB2-Report-Utility und DB2-Katalog nach wiederherzustellenden SAP-Tablespaces mit recoverfähiger RBA zur Automatisierung des Conditional Restart Verfahrens) u. Oracle-Systemadministration in einer SAP R/3-Basis-Umgebung unter OS/390-TSO-ISPF-LIBRARIAN, DB2 Version 5 und DB2 UDB Version 6, CLIST, RACF, JCL, COBOL, AIX, ORACLE Version 7 u.8, Open-Edition, SAP R/3-BC 4.5B., Omegamon [SEP]	[CLS] [KNOWLEDGE] RxJS [CTX] Konsumieren der Rest APIs mit HttpClient und RxJS [SEP]	-0.06634289771318436
  [CLS] [SKILL] Erstellung Prozessen [CTX] Erstellung von Dokumentationen und Arbeitsanweisungen und Prozessen Kommunikation auf allen Ebenen mit vielen Abteilungen international. [SEP]	[CLS] [KNOWLEDGE] DB2 8.1 [CTX] Systemadministration, Betrieb, Monitoring und Fehlerbehebung von über 100 SAP-Systemen unterschiedlicher Releasestände (4.6 C,6.20,6.40, EP, XI) und deren Datenbanken (Oracle 9.2, Informix 9.4, 9.3, DB2 8.1, SAP DB 7.5) auf pSeries Rechnern unter AIX 5.3 und z/OS. Durchführung und Überwachung des Transport-Managements (SAP-intern und auf OP-Ebene(AIX)) und [SEP]	-0.009634226560592651

• Loss: CosineSimilarityLoss with these parameters:

  {
      "loss_fct": "torch.nn.modules.loss.MSELoss"
  }
  

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: 1
• 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: 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: 1
• 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: batch_sampler
• multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	Training Loss	Validation Loss	sts-dev_spearman_cosine	sts-test_spearman_cosine
-1	-1	-	-	0.8023	-
0.0921	500	0.0172	0.0056	0.9152	-
0.1843	1000	0.0064	0.0043	0.9330	-
0.2764	1500	0.005	0.0037	0.9399	-
0.3686	2000	0.0047	0.0035	0.9456	-
0.4607	2500	0.0042	0.0032	0.9475	-
0.5529	3000	0.0039	0.0029	0.9514	-
0.6450	3500	0.0036	0.0027	0.9525	-
0.7372	4000	0.0034	0.0027	0.9547	-
0.8293	4500	0.0033	0.0026	0.9564	-
0.9215	5000	0.0033	0.0025	0.9571	-
-1	-1	-	-	-	0.9592

Framework Versions

• Python: 3.12.4
• Sentence Transformers: 4.1.0
• Transformers: 4.49.0
• PyTorch: 2.4.0+rocm6.3.4.git7cecbf6d
• Accelerate: 1.6.0
• Datasets: 3.5.0
• Tokenizers: 0.21.1

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",
}